Sapphire teased a new AMD Radeon GPU on Twitter. The graphics card, which features the classic dual-fan Sapphire Pulse cooler design, is likely a Radeon RX 9070 or RX 9070 XT.

If this will indeed be an AMD Radeon RX 9070 XT, recent benchmarks (if they are accurate) of real-world raster performance seem to indicate a card highly competitive with Nvidia RTX 4070 Ti and RTX 4080 Super, so seeing it in a more reasonable dual fan configuration is nice.

Narrowing our focus back on the Sapphire Pulse GPU specifically, some things can be approximated based on the imagery. Besides being a standard dual-slot, dual-fan GPU, unlike the other AIBs we've seen, the actual GPU length also seems quite conservative and more compact to fit into your typical mini-ITX case.

Of course, bulkier AIB GPUs like Asus' various triple-fan RX 9070 and 9070 XT models are better suited for larger cases that can accommodate them, at least theoretically. We'll have to wait for benchmarks before determining the final differences between these AIB GPUs and how their cooler designs may impact performance and overclocking ability.

The new Sapphire Radeon Pulse design looks nearly identical in terms of visual design. However, the color scheme for this particular series of GPU cooler designs has evolved over the past few generations. This is mainly due to removing many white accents that used to come with Sapphire Pulse GPUs, like the Sapphire Radeon Pulse RX 5600 XT. Now, the design is much more streamlined around blacks with red highlights, as on the last-gen Sapphire Pulse RX 7900 XT.

These design changes are noted and probably blend better into more PC builds, but fans of the classic Sapphire Pulse GPU color scheming may be a little sad to see Sapphire's visual language fall more in line with other AIB-making black-and-red GPUs.

The Linux team has published an intermediate Linux 6.13-rc3 to fix a "hilarious" bug in the kernel-based Virtual Machine code — Phoronix reports, where CPUID updates are taking significantly longer time to be processed on Sapphire Rapids CPUs than they should be, resulting in much older Skylake CPUs being up to 4X faster in regards to CPUID cost.

Phoronix reveals that Google engineers originally discovered this technical issue in how Sapphire Rapids processes CPUID information in nested VMs. Specifically, Google's Sean Christopherson states, "On Intel's Emerald Rapids, CPUID is *wildly* expensive, to the point where recomputing XSAVE offsets and sizes results in a 4x increase in latency of nested VM-Enter and VM-Exit (nested transitions can trigger xstate_required_size() multiple times per transition), relative to using cached values. The issue is easily visible by running `perf top` while triggering nested transitions: kvm_update_cpuid_runtime() shows up at a whopping 50%..."

This performance bug/regression won't be purportedly fixed until Linux 6.14, but the issue is severe enough for the Linux devs to fix the outgoing Linux 6.13 kernel temporarily. Update 6.13-rc3 caches the CPUID data on Sapphire Rapid CPUs, speeding up this process. The Linux 6.14 implementation of the complete fix will result in all the CPUIDs nested in VM-Enter and VM-Exit being completely deleted.

For those unaware, CPUID is an instruction that lets software discover details of the processor it is running on. On Sapphire Rapids, the software reading CPU information through the CPUID instruction takes up to 4X longer than on much older Skylake CPUs. This only occurs inside nested virtualization transitions, which only happen when a VM runs inside a VM.

The Linux 6.13-rc3 comes with various other bug/regression fixes. However, the caching workaround for expensive CPUID handling appears to be one of the more critical updates to the kernel. Again, this caching technique is just a workaround, with Linux 6.14's complete fix for the issue.

One of the surprises brought by Intel's launch of Arc B580 and Arc B570 graphics cards was the emergence of a new add-in-boards (AIB) brand: Onix. New players rarely arrive in the AIB market due to cutthroat competition and diminishing margins. However, Onix (via BenchLife) is reportedly not a newbie but a brand belonging to Sapphire, a veteran with a history spanning over 20 years.

Apparently, Onix Technology Limited was registered in Hong Kong on August 28, 2024. According to BenchLife, Onix Technology Limited is affiliated with Sapphire Technology Limited. Sapphire is primarily known for its AMD Radeon-based graphics cards, motherboards, and servers for AI and HPC. Unfortunately, we do not know whether Onix is a part of Sapphire or a part of PC Partner Group, which has a stake in Sapphire and owns Zotac, a maker of Nvidia GeForce-based graphics cards. In any case, Onix is not a newbie on the market.

Since AMD has gradually lost its share of the discrete graphics card market, it is logical for its partners to try and adopt other GPUs to make add-in-boards and earn money. For example, Tul Corp., the owner of PowerColor, resurrected its Sparkle brand last year to sell Intel Arc-based boards, so we will not be surprised if Sapphire follows suit. In theory, Sapphire could adopt Nvidia's GPUs, too, but allocation for these processors is tight, so it is possible that the company decided to go to Intel instead.

Given that PC Partner has worked with AMD, Intel, and Nvidia for decades, establishing a separate brand for Intel Arc graphics cards would not be a problem for the company, so it is possible that Sapphire itself has little to do with Onix.

For now, Onix only offers Intel Arc B580 and Arc B570 graphics cards, which target the entry-level market and will hardly grab a significant market share with these boards. Nonetheless, this is a start, and assuming that Intel's Arc B700-series products will be faster, they will gain more attention from enthusiasts and, therefore, a larger chunk of the market.

IT Home reports that Sapphire has officially launched a new B650 micro-ATX motherboard based on the AM5 platform: the Pulse B650M. This motherboard features a unique two-DIMM slot configuration.

It is not uncommon for budget micro-ATX motherboards to remove two DIMM slots to save money. Still, the Pulse represents a more upbeat/premium micro-ATX motherboard, suggesting the two-slot configuration was done for performance reasons over price. The board features a built-in I/O shield and a cover over the entire left area of the power delivery area, a configuration generally reserved for mid-range and premium motherboards.

The Sapphire B650M Pulse has a 7+2+1 phase power delivery system featuring 60 amp power stages, powered via a single EPS 8-pin connector. The two DIMM slots are purportedly optimized for memory overclocking with support of up to "DDR5 7400+," making it more than adequate to run memory speeds Ryzen 7000 and Ryzen 9000 chips are best suited towards (i.e., DDR5-6000 to DDR5-6400).

PCIe connectivity comes in the form of a single shielded PCie 4.0x16 slot and just one PCIe 4.0x4 slot at the bottom. Sandwiched between the two slots is a secondary M.2 slot rated at PCIe Gen 4x4 speeds. The primary M.2 slot is between the AM5 socket and primary x16 slot, with a cover attached and operating at PCIe Gen 5x4 speeds. Additionally, there are four SATA 3 connectors for adding additional slower storage.

The rear I/O consists of dual USB 3.2 ports, six USB 2.0 ports, a single HDMI port, and a single DisplayPort connector for video output through AMD Ryzen APUs. Networking comes in the form of a single Realtek RTL8111K Gigabit ethernet port at the rear and a Wi-Fi 6 card featuring Bluetooth 5 functionality.

Compared to competing AM5 B650 motherboards, Sapphire's Pulse model stands firmly in mid-range territory (or less) based on features and power delivery. For example, the Pulse features better power delivery than Gigabyte's B650 Gaming Wi-Fi, which comes with a 5+2+2 configuration but falls short of Gigabyte's more premium B650M Aorus Elite, which features a 12+2+1 power delivery configuration. The same trait applies to official memory overclocking compatibility.

Wi-FiPricing for the B650M Pulse is 899 Yuan with WiFi and 799 Yuan without Wi-Fi, which translates to $127 and $112, respectively. It remains to be seen if this board will make its way to Western markets, but at the very least, it will be available in the Asian sector.

Major AMD graphics AIB partner Sapphire launched the Pure Radeon RX 7700 XT Frostpunk 2 Edition video card on Newegg. This mid-range GPU, priced at $449.99, features 12GB of GDDR6 VRAM but has a higher core clock of 2,226 MHz and a unique port configuration of two HDMI 2.1 and two DisplayPort 2.1 ports. Sapphire also gave the card ARGB lighting instead of the original one, which had only red LEDs. Aside from that, it comes with a code for the Deluxe Edition of Frostpunk 2 — slated to arrive on September 21 and priced at $75 on launch.

This Frostpunk 2 Edition GPU isn’t the most affordable option among Sapphire’s RX 7700 XT lineup, with the two-fan Sapphire Pulse at just $399.99. But if you’re a fan of the indie game, then the $50 premium you pay with this GPU is worth it, as you’re effectively getting a $25 discount if you buy it at launch. Even if you pre-purchase the title today, you’ll still save $17.50, which makes it a steal.

Furthermore, our review of the vanilla RX 7700 XT Pure, which costs a pretty penny at over $700 on Newegg at the time of writing, shows that it’s one of the quietest GPUs you can get in the 7700 XT range — so, even if you’re not interested in a game that covers snow and the breakdown of society, you’re ultimately getting tremendous value with the Frostpunk 2 Edition GPU.

Sapphire also included three Frostpunk 2 enamel pin badges, white F and P, and two keycaps for those who are really into the game. While these trinkets don’t offer much for your gaming experience, you can add them to your Frostpunk-series merch collection to show your dedication to the title.

The 7700XT isn’t a beefy graphics card designed for 4K gaming. Instead, it’s more comfortable playing games at 1440p and 1080p. This partnership also gives us a preview of the system requirements for the sequel to the popular city-builder/survival game. Steam lists its recommended specs as AMD Ryzen 7 / Intel Core i7 2.8 GHz, 16GB RAM, and an RX 5700, RTX 2060, or Arc A770 GPU, all of which sport 8GB of VRAM. So, the 12GB VRAM on the RX 7700 XT Pure should be more than enough to run the game smoothly and on higher-quality settings.

European IT distribution company EET Group just listed six Sapphire Rapids Refresh chips on their website, as shared on X by leaker momomo_us. These Sapphire Rapids Refresh chips all have Lenovo branding, so we could assume that these listings are for Lenovo workstations like the ThinkStation.

While Intel has yet to officially launch the Xeon W3500-series chips, which will compete with AMD’s Threadripper and Threadripper Pro processors, some companies would be eager to get their hands on the latest-generation Intel workstations, which will give them an advantage.

We previously reported on the benchmarks for these chips last February, showing the W5-3535X outperforming the fourth-gen Sapphire Rapids CPUs in multi-threaded workloads. Hardware detective momomo_us also shared the specs for the new Sapphire Rapids Refresh CPUs in the same month, showing us that the next-generation chips received at least four performance cores and a small bump in their TDPs.

Intel Xeon W3500 Series Specifications*

*Specifications are unconfirmed.

The additional cores on these new chips will give users who rely heavily on multi-threaded workloads additional performance, but they’re still heavily outclassed by AMD offerings. The top-of-the-line Threadripper Pro 7995WX boasts 96 cores—52 more than Intel’s most expensive offering.

The 96-core AMD Threadripper Pro is far more expensive than what Intel offers, with the costly 7995WX costing at least $10,998 on Amazon, versus Intel’s current top offering, the 36-core Intel Xeon W9-3475X, a little less than $4,000. However, AMD’s closest competitor, the 32-core Threadripper Pro 7975WX, offers a more reasonable price of $3,900.

Although Intel has already launched the next-generation Emerald Rapids processors for the most demanding applications, the Sapphire Rapids Refresh Xeon W chips are designed for creative professionals. These chips give them the power they need without the extra features (and corresponding higher prices) they won’t use anyway in their applications.

When AMD launched the Radeon RX 7900 GRE for the rest of the world — it was previously only available in China — we were impressed with the performance and value. We tested the base model Sapphire Pulse originally, and now we have the blinged out and factory overclocked Nitro+ variant. The 7900 GRE already rates as one of the best graphics cards, currently our top overall pick for AMD GPUs, but let's see how the Nitro+ mixes things up.

The biggest change is that the Nitro+ costs $60 more than the Pulse, with minor gains in performance. It will mostly be of interest to those who like its big and boxy aesthetic and RGB lighting. The higher price puts it into direct competition with the RTX 4070 Super, though of course a fancy-pants 4070 Super would also cost more than the base model Founders Edition. Here's a rundown of the reference specs compared with the other 7900 GRE cards we've tested.

As you'd expect, the key difference between the various RX 7900 GRE cards comes down to stated boost clocks, at least insofar as specifications are concerned. There are other differences in terms of aesthetics and video ports, but the base hardware always stays the same.

AMD's official reference boost clock for the 7900 GRE is 2245 MHz, but AMD doesn't make a reference model card and most of its add-in board (AIB) partners don't seem particularly interested in limiting performance that way. Even the Sapphire Pulse, which is the least expensive 7900 GRE currently selling in the U.S. market, has a boost clock 48 MHz above the reference value. That's also the lowest boost clock we can find listed for any of the cards currently selling in the U.S.

In practice, that means the Nitro+ really has a 98 MHz advantage (4.3%) over its base-MSRP sibling, rather than a 146 MHz (6.5%) advantage over the supposed reference clocks. The ASRock Steel Legend, which we'll review separately, lands between the two Sapphire cards with a 2333 MHz boost clock.

And it's not like the boost clocks actually directly equate to real-world clocks. We'll have the details on page three, but the Nitro+ averaged 2394 MHz across all the games, resolutions, and settings that we tested, shockingly only 3 MHz away from the stated boost clock. The 7900 GRE Pulse one the other hand averaged 2303 MHz (10 MHz higher) while the ASRock Steel Legend averaged 2381 MHz (48 MHz higher).

TBP (Total Board Power) is another area where AIB partners can differentiate, but not every card manufacturer provides an official TBP value. AMD's reference power is 260W for the 7900 GRE, Sapphire lists 268W for the Pulse card, and the Nitro+ kicks that up to 295W. That's a potentially significant 30W increase (13.5%) in power use, which should also allow the Nitro+ to sustain higher clocks. ASRock doesn't list a TBP, instead recommending an 800W or larger PSU.

Theoretical compute scales directly in proportion to the GPU clocks, so the Nitro+ could be about 4% faster than the Pulse model. We'll get to the numbers on the next page, but it's worth noting that differences in VRMs (Voltage Regulator Modules), firmware, fans, RGB lighting (or the lack thereof), and other aspects of the card designs can also impact power use and thus performance.

Sapphire Radeon RX 7900 GRE Nitro+

The Sapphire 7900 GRE Nitro+ design looks and feels pretty boxy. Yes, it has rounded corners, but it's a very large graphics card measuring 321x131x60 mm (our measurements). That's a full triple slot width, in comparison to the 2.5-slot Pulse card that measures 321x127x51 mm — not that you can actually use that third slot with the slightly smaller Pulse.

The Nitro+ is also quite chunky, tipping the scales at 1604g. That's not the heaviest graphics card we've ever seen (by a long shot — RTX 4090 cards continue to hold that 'crown' with some weighing over 2500g), but the Pulse and ASRock cards are quite a bit more svelte at 1100g and 1197g, respectively. All that extra mass should hopefully improve the cooling potential, but much of the weight likely comes from the full metal shroud used on the Nitro+.

The Nitro+ fans are the same 95mm 'angular velocity' design that we've seen in other recent Sapphire cards, including the Pulse. The only real difference is in the sticker on the fan hub, where the Pulse has the name on the sticker, while the center Nitro+ fan has the Nitro fox (snake?) face and the other two fans just have a silver circle. We've been impressed with Sapphire's fans in recent cards that we've tested, and expect good things from them here.

All the 7900 GRE cards that we've seen have dual 8-pin power connectors, and that includes the Nitro+. That's enough for 300W of power, plus another 75W max from the PCIe x16 slot. Even if you want to overclock the card beyond the factory settings, power limits shouldn't be a serious concern.

One final item of note with the Nitro+ is that it includes a full triple-slot IO bracket, which can help support the large card. Video outputs consist of dual DisplayPort 2.1 54 Gbps and dual HDMI 2.1 48 Gbps ports. Most other cards tend to include three DP outputs, but that should only matter if you plan on hooking up more than two monitors.

The packaging and other materials don't hold any surprises. Besides the colorful box and padded anti-static bag around the card, the only extra consists of a support bracket. It's basically the same as the one that comes with the Pulse card, only this one ships inside a thin cardboard box.

As a premium brand card, the Nitro+ also comes with a dual-BIOS switch (located next to the 8-pin power connections). This can be helpful should you ever need to update your VBIOS, but in practice it's not a super critical feature. There are three switch locations: "Silent," "Performance," and "Software controlled" — with the latter being the default. You can use Sapphire Trixx to change modes, though without that software there's no indication which mode the Nitro+ uses by default. (Based on testing, we assume it's the "Performance" mode.)

AMD Radeon RX 7900 GRE Test Setup

Our current graphics card testbed has been in use for over a year now, and so far we haven't seen any pressing need to upgrade. The Core i9-13900K is still holding its own, and while the i9-14900K or Ryzen 9 7950X3D can improve performance slightly, at higher resolutions and settings we're still almost entirely GPU limited — though perhaps not when the future RTX 50-series and RX 8000-series arrive.

Note that, due to the widely reported Core i9-13900K/14900K instability problems — which were affecting certain games until we changed BIOS settings — we have recently retested all the cards in our test suite. Most of the time, the changes in performance have been within margin of error (1~2 percent), and driver and game updates over time could easily account for some of the that as well.

We're using the latest 24.4.1 drivers from AMD for the Nitro+, which based on limited testing appear to perform the same (for our test suite) as the 24.3.1 and 24.2.1 drivers.

Our latest test suite consists of 19 games. Twelve of the games support DirectX Raytracing (DXR), but we only enable the DXR features in eight of the games. The remaining eleven games are tested in pure rasterization mode. In general, we find rasterization performance is a more useful metric for gaming, though many newer titles have added ray tracing effects. Reflections and global illumination remain the most impactful of the RT effects.

We test at 1080p (medium and ultra), 1440p ultra, and 4K ultra for most of our reviews — ultra being the highest supported preset if there is one, and in some cases maxing out other settings for good measure (except for MSAA or super sampling).

Our PC is hooked up to a Samsung Odyssey Neo G8 32, one of the best gaming monitors around, allowing us to fully experience the higher frame rates that might be available. G-Sync and FreeSync were enabled, as appropriate. As we're not testing with esports games, most of our performance results are nowhere near the 240 Hz limit, or even the 144 Hz limit of our secondary test PC.

We've installed Windows 11 22H2 and used InControl to lock our test PC to that major release for the foreseeable future. We've also used Winaero Tweaker to disable Windows Update and some other elements. Basically, we're doing our best to lock down our PC and prevent OS updates, as much as possible, so that our test results over time remain valid.

Unfortunately, there's no good way to lock down Steam and other game libraries to prevent them from updating. Regular game patches can and do affect performance, sometimes for good and sometimes not. When we notice a major discrepancy in results after new updates come out, we retest all the cards. Thankfully, that hasn't happened since our most recent round of retesting commenced.

Our test PC includes Nvidia's PCAT v2 (Power Capture and Analysis Tool) hardware, which means we can grab real power use, GPU clocks, and more during all of our gaming benchmarks. We'll cover those results on our page on power use.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Since the Nitro+ represents a third-party post-launch review, rather than a newly released GPU, we're going to condense all of our gaming performance results onto a single page. We'll work through the 1440p, 1080p, and 4K results in that order, with 1440p arguably being the most important battle ground for this level of hardware.

For each resolution and setting, we have a gallery of 22 charts. The first gives the overall geometric mean for our 19 game test suite, so each game gets equal weight in the score. Then we'll have separate rasterization-only and DXR-only charts, using the geometric mean of the 11 and nine games, respectively. Finally, we'll have the 19 individual game charts, in alphabetical order. Games that have ray tracing enabled show as "(DXR)" while the rasterization games mostly use DirectX 12, with only a few exceptions.

Please note that while we use a geometric mean to give equal weighting, our test suite does have a higher proportion of DXR games than what you would see if you just took for example every major game release of the past year (and yes, we understand how nebulous the term "major release" is). We still feel that, in general, the improved image quality from DXR doesn't often justify the performance hit.

Besides full path tracing, RT reflections and global illumination tend to be the most visually impressive aspects, with shadows and ambient occlusion benefiting far less from the RT algorithms. The performance hit for only limited use of RT shadows can be much lower than for lots of reflections, but the visual gains also tend to be far less meaningful.

There are games where RT can make a noticeable and noteworthy difference in image fidelity, but there are also a lot of DXR games where it mostly tanks performance for minimal quality gains. We've tried to select DXR games where the improvements are more noteworthy, though we'll readily admit that Avatar and Diablo IV come up short in our view.

Starting with the 1440p results, the Nitro+ outperforms the base model Pulse card by just 2.2% across all 19 games. While that's technically measurable, it's not actually noticeable if you're just playing the games. The gap does increase to 2.9% in our DXR test suite, and drops to 1.7% in the rasterization suite, indicating that the more demanding a game is, the more the boosted power and clocks on the Nitro+ card will help.

Compared with the similarly priced RTX 4070 Super, a factory overclocked card like the Nitro+ doesn't seem to make much sense. It's technically a bit faster in rasterization performance, leading by 3.6% in aggregate, but the 4070 Super leads by 20% in the DXR tests and takes a 7% lead overall. And that's still not really factoring in Nvidia's DLSS advantage, which is in about twice as many games as FSR 2/3 and also looks better. Of course if you want some RGB bling to go with the 4070 Super, then you're generally looking at spending $630 or more.

The other comparison against Nvidia hardware would be the RTX 4070 vs RX 7900 GRE, where we also gave the nod to Nvidia's GPU thanks to it's relatively similar performance and superior feature set — again, referring mostly to the DLSS and AI capabilities, plus ray tracing. That was a contentious call, and certainly we understand why some people would prefer a 7900 GRE. Regardless, boosting performance a couple of percent while increasing the price 9% doesn't make for a compelling argument in favor of the Nitro+, if you're just looking at the raw performance.

Dropping the resolution to 1080p doesn't change the story much. The overall difference in performance between the Nitro+ and the 7900 GRE Pulse shrinks slightly — there's a 1.9% gap at 1080p ultra and a mere 1.4% advantage at 1080p medium — but outside of running benchmarks, you wouldn't actually feel the minor change in performance. And in the less demanding rasterization test suite, the two GPUs are all but tied with a 1.1% difference at ultra settings and 0.5% difference at medium quality.

Nvidia's competing GPUs, the RTX 4070 Super and vanilla 4070, do slightly better at 1080p than 1440p. That's because they have less VRAM and memory bandwidth, the latter being the more critical factor, which means the reduced demands of 1080p allow the GPUs to perform a bit better.

Would you notice the difference between the 7900 GRE Nitro+ and a 4070 Super? Yes, but it's purely on a game by game basis — Borderlands 3 gives the Nitro+ a 13% lead, which you might be able to feel, though that's doubtful given we're talking about 177 fps versus 200 fps. On the other hand, Minecraft, Spider-Man, Cyberpunk 2077, Bright Memory Infinite, Control, Avatar, and Metro Exodus — all DXR enabled games, not surprisingly — give the Nvidia GPU anywhere from a 14% to a 78% lead.

Last but not least, we have the 4K ultra results. The Nitro+ does see its largest margin of victory over the base model 7900 GRE... to the tune of a 2.6% overall. So again, pure performance isn't a great reason to opt for a more expensive card using the same GPU.

Many of the individual games fail to reach playable framerates on this level of hardware, at least without upscaling. Even with Quality mode upscaling, ray tracing games will often fall short of 60 fps. For the Nitro+, half of the DXR-enabled games fail to break 30 fps with our tested settings, and none of them break 60 fps. Diablo IV, with Quality mode upscaling, comes closest at 57 fps. In fact, only seven of the rasterization games break 60 fps on the Nitro+ card, which is the same count as the 7900 GRE Pulse.

Nvidia's competing RTX 4070 Super in this case does one better in terms of games breaking 60 fps, with Diablo IV using Quality upscaling hitting 73 fps. But the 4070 Super also only has two games where it fails to average at least 30 fps, Cyberpunk 2077 and Bright Memory Infinite (without upscaling in both cases). The RTX 4070 falls below 30 fps in three games (Control joins CP77 and BMI), but also only breaks 60 fps in four games (Assassin's Creed, Far Cry 6, Forza Horizon 5, and Horizon Zero Dawn).

In short, modern 4K gaming still generally requires picking up one of the fastest GPUs available, and perhaps even then you'll still need some level of upscaling. We're fine with that, particularly with higher quality upscaling solutions like DLSS. If you're playing older, less demanding games, 4K might run well, but maxed out settings in recent releases tends to be too much for midrange GPUs.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

The performance results on the previous page don't provide a compelling reason to pick up the Nitro+ over the Pulse (or any other less expensive RX 7900 GRE). But there's more to a graphics card that raw performance. Here we'll check out the power use, clock speeds, and GPU temperatures while running our gaming test suite. We also check noise levels, which are an area where cards like the Nitro+ can truly differentiate themselves from lesser models.

Our graphics card test PC uses an Nvidia PCAT v2 device, as it gives us far more data without the need to run separate tests. PCAT with FrameView allows us to capture power, temperature, and GPU clocks from our full gaming suite — along with the usual frame times. The charts below are the geometric mean across all 19 games, and we also have full tables showing the individual results further down the page.

PCAT does not favor Nvidia GPUs in any measurable way. We checked power with our previous setup for the same workload and compared that with the PCAT, and any differences were well within the margin of error (less than 1%). PCAT is external hardware that simply monitors the power draw of the PCIe power cables as well as the PCIe x16 slot by sitting between the PSU/motherboard and the graphics card.

We have separate charts for 1080p medium/ultra, 1440p ultra, and 4K ultra in the galleries below. There's no reference design for the RX 7900 GRE, so the Sapphire Pulse will have to suffice. The Sapphire Nitro+ represents the top of what we're likely to see from the 7900 GRE, give or take — Asus, Gigabyte, PowerColor, and XFX also offer 'halo' parts that have a boost clock of 2391–2395 MHz, with the XFX 7900 GRE being the least expensive option and also the highest boost clock.

Sapphire RX 7900 GRE Nitro+ Power Use

Sapphire gives a 295W TBP (Total Board Power) for the Nitro+, 35W higher than the official reference spec from AMD. The average power draw at our four test settings doesn't quite reach that high, topping out at 290W at 4K, but individual games can get there.

The highest average power (during a gaming test) comes from Cyberpunk 2077 with 297.6W. Again, that's at 4K, with the lowest power from the games we tested for 4K still hitting 286.7W. That's not much of a spread. However, dropping to 1440p gives a power range of 228–292 watts, 1080p ultra is 134–193 watts, and 1080p medium shows 145–294 watts. Diablo IV and Flight Simulator fall well below most other games at 1080p, as they're very much CPU limited at our test settings with that resolution.

The potential concern here is that the Sapphire Pulse draws 10W to 27W less power across our test suite. That's not a huge difference, but neither is there a major performance difference between the two GPUs. The ASRock Steel Legend incidentally uses slightly more power than the Nitro+, except at 4K where it's 3W less.

Sapphire RX 7900 GRE Nitro+ Clock Speeds

Boost clocks and power limits play into the average GPU clock speed, so these charts are mostly interesting as a way to see how close the various GPUs come to their stated boost clocks.

The Nitro+ doesn't quite hit its advertised 2391 MHz at 1080p medium, in part because games like Diablo IV, Far Cry 6, and Flight Simulator bring the average down thanks to hitting CPU limits. The highest average boost clocks while playing through our gaming suite easily exceed the boost clock, however, with a few titles landing around 2550 MHz — Assassin's Creed, Avatar, Diablo IV, Forza Horizon, and Spider-Man all reach into the 2500+ MHz range, depending on resolution.

Compared to the Pulse model, the Nitro+ average clocks are 50~140 MHz higher, with the bigger delta coming at 4K. That makes sense, as the higher power limit of the Nitro+ card comes into play there, and that also explains why the performance tends to be similar at 1080p. The ASRock card, incidentally, does clock about 15~35 MHz lower than the Nitro+ at 1440p and 4K. So the boost clocks are meaningful, provided the cards are pushed to the limit.

Sapphire RX 7900 GRE Nitro+ Temperatures

GPU temperatures and fan speeds go hand in hand, and most modern GPUs are perfectly fine at up to 90C, give or take. We don't have a good way of collecting fan speed data from our test suite, sadly, but we'll let noise levels stand in for that below. The Sapphire 7900 GRE Nitro+ is physically larger than the Pulse, mostly by being half a slot width thicker. That should result in lower temperatures, and that's precisely what we see.

We test in a 21C ambient, +/-2C, and the triple 95mm "Angular Velocity" fans combined with a large heatsink do a good job at keeping the 7900 GRE cool. Average temperatures across our test settings range from 62.5–64.1 degrees Celsius, but averages here aren't as meaningful — what we really want to know are the "worst-case" thermals.

As above, lighter games — including games that hit CPU bottlenecks — don't tend to heat up the GPU as much. Other factors also come into play, however, so that the "coldest" games aren't just the ones with the lowest core clocks. Far Cry 6 doesn't heat up the GPU as much due to CPU bottlenecks, but at the same time neither do Control and Cyberpunk 2077.

Peak average temps from across our test suite are around 68C, while in less demanding games the Nitro+ might land in the low 50s. Across our entire suite, the Nitro+ runs about 4C cooler than the Pulse on average. Curiously, the ASRock card runs cooler yet, beating the Nitro+ by 3C~4C depending on the settings used. But we need to check noise levels as well to see how that might factor into the equation.

Sapphire RX 7900 GRE Nitro+ Noise Levels

The Sapphire RX 7900 GRE Nitro+ runs very cool and quiet, and yet, it's not quite as quiet as the 7900 GRE Pulse card. It's 1.5 dB(A) noisier than the Pulse in our gaming test. At the same time, it's also 5 dB(A) quieter than the ASRock card, which explains why the temperatures were lower on the Steel Legend — faster fan speeds move more air and thus cool better, but they make more noise.

It looks like Sapphire may have tuned fan speeds to be a bit more aggressive (meaning, faster) due to the higher power draw of the Nitro+. That helps it run cooler but creates slightly more noise than the 'lesser' Pulse card.

Note that we measure noise levels at 10cm from the middle GPU fan, to better pick up any differences between the various cards. The noise floor of our test environment and equipment is around 31~32 dB(A). A result of 37.3 dB(A) while running a game is still very quiet and most people wouldn't notice it above the sound of the rest of the PC's fans. That corresponds to a 26% fan speed, so there's plenty of headroom remaining.

Nvidia's competing GPUs (at least the reference Founders Edition cards) are a bit louder, even though they use less power, though they're also more compact. Smaller cards generally don't have as much surface area to dissipate heat, which means they either run hotter, or need higher fan speeds, or both.

We also test with fan speed set to a static 75%, which is more about seeing how much headroom there is for additional cooling if needed. The Nitro+ generated a lot more noise, hitting 58.4 dB(A), but we don't expect most users will ever see fan speeds anywhere near that high.

Sapphire RX 7900 GRE Nitro+ Efficiency and Value

For those who like to look at the finer details, we've provided the above tables showing all of our gaming benchmarks, which include FPS/$ and FPS/W metrics. We use the best prices we can find at the time of writing, which can and do fluctuate on a daily basis.

More expensive cards generally have worse values, as measured by FPS/$. That applies even more to premium cards like the Nitro+, which costs $60 more than the Sapphire Pulse card but only improves performance a few percent. Of the cards we're looking at today, the best value award goes to the RX 7700 XT, thanks to recent price cuts. The 7900 GRE comes relatively close at 4K, but the Nitro+ model falls further behind and basically matches the FPS/$ score of the 7900 XT at higher resolutions.

As we've seen ever since Nvidia launched its RTX 40-series GPUs, efficiency isn't even close between AMD and Nvidia. For 1440p, the tested Nvidia GPUs fall in the 0.397–0.443 FPS/W range, while the AMD GPUs land in the 0.272–0.310 FPS/W range. The RTX 4070 Super ends up being around 50% more efficient than the Nitro+ card.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Sapphire's RX 7900 GRE Nitro+ caters to the gamer that wants a good AMD GPU at the heart of their PC, and then they want to add some flare and aesthetics. If you're just after a great performing graphics card for $500–$600, Sapphire's Pulse line arguably does a better job at balancing price and performance. The Nitro+ takes the same core features, then puts them into a physically larger card with superior cooling, and adds some RGB bling for good measure.

Many people will be perfectly fine skipping out on the extra girth and bling that the Nitro+ brings to the party. At the same time, others want something more than just another boring-looking graphics card. Whichever route you prefer, you can't really go wrong. I might make fun of him for it, but I know a guy that built a $500 budget gaming PC several years back... and then he promptly went out and spent another $250 on RGB lighting and bling — there literally wasn't a single addition that improved performance, and that sucked up a third of the total cost of the PC! I won't mention any names, but what's important is that he was happier with the fancy-pants gaming PC, even it it wasn't any faster.

If you're that sort of person, and the Nitro+ aesthetic scratches your particular itch, by all means: Go for it. The 7900 GRE provides one of the best balanced cards in AMD's current GPU product stack, with great performance at the right price. Do you want that with something more than the base model card? Then the Nitro+ is for you.

Of course, if you're not interested in upgrading to an AMD graphics — either because you're simply not in the market right now, or you prefer Nvidia GPUs, or for whatever other reason — there's nothing new in the Nitro+ that would change your mind. As a value proposition, it's provably worse than Sapphire's own RX 7900 GRE Pulse.

Compared with Nvidia's RTX 4070 and RTX 4070 Super, while there are certainly those who will prefer an AMD GPU, there are also plenty of gamers that want access to Nvidia's superior feature set. We're talking about DLSS, including Ray Reconstruction and maybe even frame generation. There's also Broadcast, ChatRTX, and tons of AI projects others have built around Nvidia's tensor cores.

Looking forward, it's not clear right now whether AMD will actually launch new RX 8000-series RDNA 4 GPUs this year or if those won't come out until 2025. Nor is it clear just how much faster such GPUs might be, or how much they'll cost. There are of course faster AMD GPUs than the 7900 GRE, but you get diminishing returns for the money spent. Slower GPUs like the 7700 XT are another option, but even with upscaling such cards may not have sufficient oomph to handle 1440p and especially 4K gaming.

In short, $599 spent on the RX 7900 GRE Nitro+ gets you overall better performance than the fastest previous generation AMD RX 6950 XT, at a much lower price, with better power efficiency and new features like the AI accelerators, AV1 encoding, and DisplayPort 2.1 video output. There will inevitably be something faster and better down the road a year or two, but this is a solid card that should have you happily gaming until the latter part of the 2020s.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Notable AMD GPU hardware partner Sapphire has released a new Mini ITX AM5 motherboard currently exclusive to China, as first reported by VideoCardz. This mini ITX motherboard, the Sapphire Nitro+ B650I Mini ITX, is priced at 1,689 Chinese yuan, roughly $233 US dollars. 

One thing you'll miss for that budget price? PCIe Gen 5 support. Here, Sapphire has opted for PCIe Gen 4.

Sapphire Nitro+ B650I Specifications

• Size and Dimensions: Mini ITX; Standard MITX board size is roughly 170mm (or 6.7 inches) per side.
• CPU Socket: AMD AM5
• CPU Chipset: AMD B650
• RAM Slots: 2 DDR5 RAM slots, rated for up to 6000 MT/s operation
• SATA Ports: 4 SATA ports
• M.2 Slots: 2 NVMe Gen 4 slots (one front, one rear)
• PCIe Slots: 1 PCI Express 4.0 x16 slot
• Lighting Features: 12V 4-Pin RGB LED Header, 5V 3-Pin RGB LED Header
• Wireless Features: Included Wi-Fi 6 and Bluetooth 5 antenna 
• I/O: HDMI port (iGPU), DisplayPort port (iGPU), 2 USB 2.0 ports, 2 USB 3.2 Gen 2 10 Gbps ports, 1 USB 3.2 Gen 2x2 Type-C 20 Gbps port, 2 USB 3.2 Gen2 5 Gbps ports, 1 RJ45 Ethernet port, and 3 standard audio ports (Mic, Line In, Line Out)

For the most part, this is a fairly standard Mini ITX motherboard. The design is colorful without being gaudy. Dual NVMe slots, dual RAM slots, and a full-speed x16 slot are all present, as well as on-board wireless networking. Even the I/O is looking surprisingly good, with plenty of USB expansion and direct-to-iGPU video ports.

However, this is an AMD AM5 motherboard without support for PCI Express Gen 5. This means that the latest cutting-edge NVMe Gen 5 SSDs won't work, but all the best Gen 4 NVMe SSDs on the market should still be fine. Modern GPUs are also fortunately not even close to saturating PCIe Gen 5 bandwidth yet, much less Gen 4, so in terms of real-world performance these should be perfectly fine compromises.

Even the most SSD-demanding games of today, like Alan Wake 2, aren't going to tap out the full potential of your NVMe Gen 4 drive. In fact, that tier of speed— and much higher on PC— is essentially the baseline around which modern console titles that leverage SSDs are being built. 

Time will tell whether or not this product makes its way to other markets like the United States; But considering its design and already English-heavy labeling, we'd be surprised if it weren't up for consideration. 

The pricing seems on par with other AMD B650 motherboard offerings, as well. Cheaper AMD A620 boards with wireless may exist, but their sacrifices to PCIe lanes make them suboptimal for those hoping to leverage a modern GPU alongside a modern NVMe drive.

The Sapphire Pulse Radeon RX 7700 XT GPU is a 12GB graphics card built around AMD RDNA 3 architecture. It usually costs around $449, but today, it's marked down to $409 at Newegg and a few other vendors. However, you can use promo code CLBDQ5323 to get an extra $10 off through Newegg, taking the price to a new, all-time low of $399. This is part of a limited offer, and it's unclear how long the discount will be available.

We had the opportunity to review an AMD Radeon RX 7700 XT GPU last year and were a bit mixed in our final opinion, rating it at 3 out of 5 stars. The price can be considered steep given the performance compared to other graphics cards on the market in its class. That said, today's discount alleviates that issue.

The Sapphire Pulse Radeon RX 7700 XT GPU features AMD RDNA 3 architecture. It has 12GB of GDDR6 and uses a PCIe 4.0 x16 interface. This GPU's base speed is 2,171 MHz and can reach up to 2,544 MHz when boosted.

The Sapphire Pulse Radeon RX 7700 XT can support four displays via two HDMI ports and two DisplayPort outputs. It can output a maximum resolution of 7680 x 4320px through either port.

Visit the Sapphire Pulse Radeon RX 7700 XT GPU product page at Newegg for more details and purchase options.

A mysterious transparent chip has been found in an HP computer, dating back to 1977. Investigations have concluded that the custom chip was formed on a sapphire substrate, making it a silicon-on-sapphire IC. What was it for? Its function was far more mundane than you might expect: It was a support component on a floppy disk controller.

Computer historian and IC reverse-engineering enthusiast Ken Shirriff uncovered this interesting chip by de-capping a broken interface chip in a vintage HP computer. But the interface wasn’t from some futuristic photonic supercomputer server experiment from the company; it was in a humble floppy drive controller PCB sitting between HP's interface bus (HP-IB) and a Z80 processor.

Shirriff explains that the silicon-on-sapphire IC was found beneath the cap of a chip called the PHI (Processor-to-HP-IB Interface) and was used in multiple HP products back in the day. “It handles the bus protocol and buffered data between the interface bus and a device's microprocessor,” according to historians' investigations.

Silicon-on-sapphire may sound somewhat futuristic, but Shirriff’s blog highlights that ICs made this way have been around since 1963 or earlier. A notable example of a silicon-on-sapphire IC is the RCA 1802 processor used on the Galileo space probe that studied Jupiter and its moons.

The IC reverse engineering enthusiast also detailed some of the peculiar features of silicon-on-sapphire, such as the sapphire substrate being an insulator, effectively isolating the various silicon regions upon it. Apparently, this type of construction “reduces the capacitance between transistors, improving performance.” Moreover, sapphire’s insulation properties can prevent stray currents and protect against low-impedance short circuits and the effects of radiation – hence its use in the aforementioned space electronics applications.

Shirriff’s blog goes into a deep dive with a look inside the HP PHI chip, its construction, and die. He even examines its logic gates, first-in-first-out buffers (FIFOs), and address decoder. Please check out the blog for all these finer details and more.

In conclusion, the computer historian echoes our initial thoughts that this silicon-on-sapphire IC is “interesting as an example of a ‘technology of the future’ that didn't quite pan out.”

Shirriff also contrasts late 70s era processors built on silicon-on-sapphire vs regular silicon in terms of energy consumption and clock speeds. Would you be surprised to hear that silicon-on-sapphire ICs were far superior using these metrics? Things might have panned out differently if these transparent ICs had been mass-produced at better yields and lower manufacturing costs. A frightening statistic highlighted by Shirriff is that HP’s silicon-on-sapphire yields were a mere 9%.

AWS has announced the general availability of Amazon EC2 M7i-flex and EC2 M7i instances powered by Intel's custom 4th-Gen Xeon Scalable 'Sapphire Rapids' processors. These processors are said to deliver up to 15% better performance over comparable CPUs used by other cloud providers. The new M7i instances are said to deliver up to 19% better price performance compared to the older M6i instances and are aimed at businesses and developers seeking efficient, high-performing compute resources in the cloud.

The M7i instances are designed for general-purpose workloads requiring the largest instance sizes or continuous high CPU usage, including large application servers and databases, gaming servers, CPU-based machine learning, and video streaming. These instances are available in different sizes. Starting from two vCPUs and 8GB of DDR5 memory (m7i.large) and all the way to 192 vCPUs, 768GB of memory, and 50 Gbps network bandwidth (m7i.48xlarge).

M7i instances also support the new Intel Advanced Matrix Extensions (AMX) for accelerated matrix multiplication operations and include built-in Intel accelerators for efficient data operations. Intel says that its CPUs featuring AMX are capable of meeting the performance requirements for large language models (LLMs) with less than 20 billion parameters. 

Future additions to the M7i family will include bare-metal sizes, suited to high-transaction and latency-sensitive workloads that will rely on Intel's Data Streaming Accelerator (DSA), In-Memory Analytics Accelerator (IAA), and QuickAssist Technology (QAT).

In addition, AWS uses Intel's Sapphire Rapids processors for its M7i-Flex instances are a lower-cost variant of M7i instances, with 5% better price performance and 5% lower prices. They are well-suited for applications that do not fully use all compute resources at all times, and therefore, they offer a baseline of 40% CPU performance and can scale up to full CPU performance 95% of the time. AWS offers different m7i-flex sizes starting at two vCPUs and 8GB of memory and up to 32 vCPUs and 128GB of DDR5 memory. 

Amazon claims that its M7i-Flex offer 15% higher CPU performance compared to competing cloud offerings, which makes them more competitive for things like web and application servers, virtual desktops, batch processing, micro-services, databases, and enterprise software.

"Intel worked closely with AWS to bring our feature-rich 4th Gen Xeon processors to its cloud customers, many of which have benefited from its performance and value for months in private and public preview," said Lisa Spelman, Intel corporate vice president and general manager of the Xeon Products and Solutions Group. "Today, we are happy to bring that same real-world value to cloud customers around the globe."

At ChinaJoy 2023, AMD showcased a new, limited edition version of its midrange Radeon RX 7600, a GPU just short of making the list of best graphics cards. While it's unclear if the card will ever make it out of Chinese shores, it brings vibrancy to the PC hardware ecosystem (and the GPU market in particular) that's been somewhat lacking. In PC gaming hardware, "serious black" is the name of the design game. With Party Animals, however, Sapphire was serious enough about the partnership to let that color scheme go.

The playfully yellow card has the Party Animals personality splashed all over it, and the card even includes an exclusive skin for a Party Animals companion character, Nemo (the Part Animals rendition of a Corgi). But it's still a refreshing piece of hardware in a sea of black, white, and red color combinations available in the mainstream consumer market. If you don't get heart attacks from seeing the little dog paws sticking to your GPU, that is.

This piece of GPU and marketing engineering go hand in hand. Sapphire (via ITHome) showed it knows its target market by choosing a mainstream graphics card from a performance and price ($299) perspective. Party Animals being a Unity game, it'd be a waste to market an RX 7900 XT, no matter how wild that party would get. It's not the top of our performers in our GPU hierarchy, but it is a product in particular (and more high-volume) price/performance band.

As the multi-billion dollar gaming market expands, more and more humans will be hardware consumers, and some of those would be interested in graphics accelerators that didn't look like industrial pieces of machinery or military-stealth-based designs. There may be a market for these more eclectic colors, with all the Pantone fever periodically washing across the market.

Unfortunately, for now, it seems that more colorful and less serious hardware releases are almost the exclusive benefit of the Asian market in general, and China especially. Due to the culture and age of their target market, these products also usually present a more graphically-intensive design, with anime and manga characters frequently represented. The Sapphire Radeon RX 7600 Party Animals is one such card with a more subdued color scheme. It'd almost be a shame if it didn't reach our side of the globe.

We may have a new contender for the best graphics cards very soon. AMD is reportedly launching the Radeon RX 7900 GRE (Golden Rabbit Edition), a cutdown version of the Radeon RX 7900 XT, in the Chinese market. Two rumored launch dates are making the rounds: July 28 (via momomo_us) and August 3.

In Chinese culture, 2023 is the year of the rabbit, which is likely where the Golden Rabbit Edition moniker originated. It isn't the first time AMD has released an exclusive Radeon graphics card for the Chinese market. Back in 2020, AMD resuscitated Polaris and launched the Radeon RX 590 GME, a refreshed Radeon RX 580 with higher clock speeds, in China.

Being a Radeon RX 7900-tier SKU, the Radeon RX 7900 GRE likely features the same Navi 31 silicon AMD has used for the Radeon RX 7900 XT and Radeon RX 7900 XTX models. As a result, the GRE variant seemingly retains the 5376-shader configuration, but will likely have the same clock speeds as the Radeon RX 7900 XT. As rumor has it, the Radeon RX 7900 GRE allegedly received a substantial cutback in the memory subsystem.

The Radeon RX 7900 GRE purportedly has 16GB of GDDR6 memory, making it the first RDNA 3 graphics card to arrive with such a configuration. It has 4GB and 8GB less memory than the Radeon RX 7900 XT and Radeon RX 7900 XTX, respectively. The downgrade doesn't stop there. The GRE model seemingly has a memory interface reduced to 256-bit, instead of the 320-bit bus on the Radeon RX 7900 XT or the 384-bit one on the Radeon RX 7900 XTX.

Fortunately, AMD didn't skimp out on the GDDR6 memory chips. They still operate at 20 Gbps, but the narrower memory interface restricts the Radeon RX 7900 GRE's memory bandwidth to 640 GB/s, 20% lower than the Radeon RX 7900 XT and 33% below the Radeon RX 7900 XTX. The card should still have 80MB of Infinity Cache, like the Radeon RX 7900 XT.

Sapphire's Nitro+ AMD Radeon RX 7900 GRE has emerged with two 8-pin PCIe power connectors, suggesting that the GRE variant will probably have a similar 300W TDP as the Radeon RX 7900 XT. If so, a 700W power supply from a reputable brand should suffice to power the new RDNA 3 graphics card.

It's uncertain if July 28 or August 3 is the lift on review embargos or the actual market launch date. Either way, it shouldn't be long before we learn more about the Radeon RX 7900 GRE, such as the performance and pricing, since we already know it's a China-exclusive SKU.

The Texas Advanced Computing Center (TACC) on Monday announced Stampede3, its new supercomputer that will be used for high-end simulations as well as artificial intelligence/machine learning applications. The new system will be based on Intel's Xeon CPU Max processors, with on-package HBM2E memory as well as Data Center Max compute GPUs.

"We will continue our long partnership with Dell and Intel and leverage the NSF investments in Stampede2 for this new science resource using the latest technology processors with high bandwidth memory, and making Intel graphics processing units widely available to the NSF open science community," said Dan Stanzione, executive director of TACC.

TACC's Stampede3 will be built by Dell using the latest hardware from Intel. The new machine will comprise of 560 nodes based on 56-core Intel Xeon CPU Max processors with 64GB of on-package HBM2E memory that will bring in nearly 63,000 general purpose cores and peak performance of about 4 FP64 PetaFLOPS. In addition, Stampede 3 will include 10 Dell PowerEdge XE9640 servers featuring 40 new Intel Data Center GPU Max compute GPUs codenamed Ponte Vecchio for AI/ML capabilities.

"We believe the high bandwidth memory of the Xeon Max CPU nodes will deliver better performance than any CPU that our users have ever seen," Stanzione said. "They offer more than double the memory bandwidth performance per core over the current 2nd and 3rd Gen Intel Xeon Scalable processor nodes on Stampede2."

The new supercomputer uses the latest Omni-Path Fabric 400 Gb/s technology, with a 24TB/s backplane bandwidth for superior scalability and low latency for various applications that require highly accurate simulations.

In addition to Intel's latest hardware, Stampede3 is set to reintegrate Stampede2 nodes based of previous-generations Xeon Scalable CPUs for higher memory applications, high-throughput computing, interactive workloads, and other smaller workloads.

TACC says that the new system will use 1,858 compute nodes with more than 140,000 cores and more than 330 TBs of RAM, 13 PBs of new storage, and almost 10 PetaFLOPS of peak performance.  

Just like its predecessors — Stampede from 2012 and Stampede2 from 2017 — Stampede 3 will be a vital part of the U.S. National Science Foundation’s (NSF) ACCESS scientific supercomputing ecosystem and will be used for open science research projects.

TACC expects Stampede3 to be delivered in fall 2023 and will then go into full performance in early 2024. It will serve the open science community from 2024 through 2029.

We reported last week that Intel Intel has confirmed that it had paused shipments of some of its fourth-gen Xeon Sapphire Rapids processors due to a newly-discovered bug, but it hadn't set a specific date for shipments to resume. The company has now issued a statement to Tom's Hardware indicating that it has developed a firmware fix and has resumed shipments. 

"Last week, we informed you of an issue on a subset of 4th Generation Intel Xeon Medium Core Count Processors (SPR-MCC) that could interrupt system operation under certain conditions. Out of an abundance of caution, we temporarily paused some SPR-MCC shipments while we thoroughly evaluated a firmware mitigation. We are now confident the firmware mitigation addresses the issue. We have resumed shipping all versions of SPR-MCC and are working with customers to deploy the firmware as needed." — Intel spokesperson to Tom's Hardware. 

Intel originally paused the shipments of a subset of its oft-delayed fourth-gen Sapphire Rapids processor line after it discovered a bug that "could interrupt system operations under certain conditions," but the company still hasn't shared specifics of the now-mitigated issue.

Intel said that it decided to pause shipments of the impacted processors out of an abundance of caution while it worked on the firmware fix. Discovering new bugs in already-shipping processors certainly isn't uncommon, but it is uncommon for those types of bugs to lead to a halt in shipments, implying that this was more than a garden-variety erratum. Intel did say the firmware mitigation isn't expected to have any performance impacts, though.

Intel's fourth-gen Xeon has two types of underlying designs: The XCC package, which employs four compute tiles (die) to create a single chip, and the MCC package, which uses a single monolithic die. As shown in the slides above, the MCC design is used for chips up to 32 cores, which are the source of high-volume sales for Intel, while the XCC variants are used for the halo chips between 36 and 60 cores.

The bug impacted only the models built on the MCC die. Intel hasn't confirmed how long it paused shipments, though unofficial reports indicate that the pause began in mid-June. Intel says it is now resuming shipments of the impacted models and is distributing its firmware fix to its partners, meaning that the company won't be required to replace any of the chips it has already shipped to customers. 

Update 7/7/2023 8:15am PT: Intel has developed a firmware fix for the issue and has resumed shipments of the impacted Xeon models, as you can read about here. 

Original Article:

Intel has confirmed that it has paused shipments of some of its fourth-gen Xeon Sapphire Rapids processors due to a newly-discovered bug, and it hasn't set a specific date for shipments to resume. We received a tip that Intel had paused the shipments, and following up on the matter, we learned several details about the issue from Dylan Patel, Chief Analyst at SemiAnalysis, who says shipments have been paused for certain SKUs since mid-June. We also followed up with Intel on the matter, and the company issued the following statement to Tom's Hardware:

"We became aware of an issue on a subset of 4th Generation Intel Xeon Medium Core Count Processors (SPR-MCC) that could interrupt system operation under certain conditions and are actively investigating. This issue was not observed when running commercially available software, and other 4th Generation Intel Xeon processor SKUs (i.e., XCC and HBM) have not exhibited the issue. Out of an abundance of caution, we did temporarily pause some SPR MCC shipments while we gained confidence in the expected firmware mitigation and expect to release remaining shipments shortly." — Intel Spokesperson to Tom's Hardware.

In response to a follow-up question, Intel also told us that it doesn't expect the firmware mitigation to have an impact on performance.

Intel's oft-delayed Sapphire Rapids processors are created using two types of underlying designs: The XCC package, which employs four compute tiles (die) to create a single chip, and the MCC package, which uses a single monolithic die. As shown in the slides above, the MCC design is used for chips up to 32 cores, which are the source of high-volume sales for Intel, while the XCC variants are used for the halo chips between 36 and 60 cores.

"Intel has faced another crop of design issues related to Sapphire Rapids MCC, the highest volume version of Sapphire Rapids. The 2-socket and 4-socket SKUs have paused shipments due to a timing issue since mid-June," Patel said.

Intel hasn't confirmed that the issue is confined to dual- and quad-socket SKUs, instead classifying this issue as limited to a 'subset' of the SKUs, and hasn't stated when the pause in shipments began. Intel also hasn't confirmed Patel's assertions that the bug is timing-related, or given us any clarification on the nature of the issue.

A timing issue could consist of any number of possibilities ranging from UPI interconnect to instruction timing issues, so the true nature of the bug remains nebulous for now. We do know that Intel can correct the issue with a firmware fix that apparently remains in validation for now, so the issue will not require a redesign or new revision/stepping to fix. Additionally, since new firmware is an adequate fix, Intel might not be required to replace any processors already in the field — although it could pose a validation headache for its customers.

Intel has earned plenty of criticism not only for its missteps on process node tech for the oft-delayed Sapphire Rapids, but also for the issues in its design and validation methodology that led to further delays and numerous new steppings (a typically minor redesign that requires a new version of silicon to correct an issue). Intel's Sapphire Rapids has been plagued with rumors that its design/verification missteps led to 12 steppings for some configs (an unnaturally large number — most chips see three steppings at most). Naturally, that led to severe production delays and missed launch dates.

The company has since communicated that it plans to take a different approach to its design, simulation, and validation flow that will correct those issues. Intel says those adjustments will kick in fully in the next generation of Emerald Rapids Xeon processors.

Intel says this new Sapphire Rapids bug wasn't encountered while "running commercially available software" (perhaps this was a hyperscaler's custom application), and it obviously wasn't caught during validation. This type of situation isn't entirely unheard of; nearly all complex chips have both known and unknown errata and bugs that are addressed with firmware, driver, and software workarounds that can reduce or eliminate those issues, and they ship that way — that's the very nature of modern semiconductor design and production.

For example, Intel's Skylake generation of processors shipped with 53 known errata, and six months later, Intel listed another 40 errata. Another example is the recent discovery that AMD's EPYC Rome chips crash after 1,044 days of uptime. Some bugs are simply left unfixed, as they aren't deemed critical enough to fix, or they are fixed with a combination of firmware and software. The most critical bugs sometimes require a new stepping to correct, which is the worst-case scenario. Luckily for Intel, that doesn't seem to be the case here.

However, while bugs aren't uncommon, it is uncommon for those types of bugs to lead to a halt in shipments, implying that this is more than a garden-variety errata. Intel hasn't clarified when it plans to resume shipments for its Sapphire Rapids MCC chips, but we'll update our coverage as we learn more.

Right now on Amazon US, there are some highly attractive deals on the AMD Radeon flagship graphics cards. An AMD Radeon RX 7900 XTX can be had for as little as $881.99 using an instant discount coupon. This graphics card, based on the AMD RDNA 3 architecture, ranks highly in our list of the best graphics cards and sits very close to the top of our GPU benchmarks hierarchy.

The hard to resist headlining offer is on the XFX Speedster MERC310 AMD Radeon RX 7900XTX Black. If you click through, you can see this model is listed at 5% off, nominally $979.99 on the XFX store on Amazon. However, a quick click of the $98 discount coupon (you may need to click on the list of available options) brings down the price to $881.99. Customers get free delivery and can apply for the free Resident Evil 4 code as well.

If you want other options, a little more cash can instead get you a Sapphire model. On the Sapphire Technology Store on Amazon you can see listed a Sapphire 11322-02-20G Pulse AMD Radeon RX 7900 XTX. The base price reads $999.99, but a quick coupon click delivers a $100 discount, bringing the total to just under $900. That's $20 more than the XFX card, but both of these may sell out.

The Sapphire card also qualifies for Prime shipping, which means you can get it in one or two days. The XFX seems to only allow slower shipping, with an ETA of next week. There's also a free code for Resident Evil 4 with the purchase of this GPU.

Both of the graphics cards above come with a factory overclock, though you can tweak them further if you like. The XFX GPU has a default boost clock of 2,615 MHz, while the Sapphire has a boost clock of is 2,525 MHz. The reference boost clock for AMD's GPU is 2,500 MHz.

Bang-for-the-Buck Considerations

We maintain a performance and street pricing table in our Best GPUs for Gaming feature. However, discount coupons are tricky to keep an eye on, as they come and go. Let's consider the average 4K FPS figures achieved by some top-end GPUs vs their street price for this exercise.

• AMD Radeon RX 7900 XTX: 4K FPS average of 56.3FPS, best online price of $882, FPS/$ is 0.0638.
  
• Nvidia GeForce RTX 4090: 4K FPS average of 84.8FPS, best Amazon price of $1,599, FPS/$ is 0.0530.
  
• Nvidia GeForce RTX 4070 Ti: 4K FPS average of 50.0FPS, best Amazon price of $799, FPS/$ is 0.0626.

While there's news swirling around Nvidia's upcoming RTX 4060 (non-Ti) debuting at some silly pricing in Europe — as high as €499 ($550) for an Asus ROG Strix model — we're seeing established top-tier cards with impressive discounts right now. Moreover, previous gen gems like the Radeon RX 6800 XT can now be found for as little as $467 in the US.

The best values will inevitably be on lower priced GPUs, but with the current discounts, the RX 7900 XTX moves up the ladder. If you've been looking at AMD's halo 7000-series GPU, the latest discounts might be the best we'll see in the near term.

It would seem that Raptor Lake Refresh isn't the only refresh that Intel has in mind. Reputable hardware leaker ECSM has revealed that Sapphire Rapids Refresh is also in the works while leaking the potential launch dates for Meteor Lake and Arrow Lake processors.

Starting on the consumer side, Raptor Lake Refresh is now the big talk. Intel recently downgraded the premium packaging on the Core i9-13900K and Core i9-13900KS. It may not sound like a significant move, but Intel typically does that when it's on the brink of launching a new series of processors. According to ECSM, Intel will unleash Raptor Lake Refresh in two waves. As usual, the K-series SKUs will arrive first, and the leaker claims they'll be on the shelves in October. Meanwhile, the non-K-series will likely launch between November and December, just in time for the holidays.

Raptor Lake Refresh won't come with a substantial performance uplift as with any refresh. The clock speeds will be higher, of course, and there's a rumor that the refreshed chips may have an improved integrated memory controller (IMC). However, the leaker believes that Intel will seemingly introduce a new SKU with an 8P + 12E configuration that would effectively slot between a Core i7 (8P + 8E) and a Core i9 (8P + 16E). Raptor Lake Refresh won't bring any new chipset or socket. The chips will slot fine into LGA1700 motherboards regardless if they're on the 600-or 700-series chipset.

ECSM stands firm that desktop Meteor Lake's cancellation remains unchanged. The 7nm processors will stay mobile-exclusive. We've already started seeing Meteor Lake-powered laptops at CES 2023, so the new processors, which will sport Intel's revamped branding, will be here in the year's fourth quarter.

Intel CPU Roadmap 2023 - 2025*

*Dates are unconfirmed by Intel.

Assuming that desktop Meteor Lake stays terminated, the desktop Raptor Lake's successor is Arrow Lake. Arrow Lake will continue to feature a multi-tile design, and this time around, the CPU tile will be a product of Intel's 20A process node. Unlike previous generations, Arrow Lake requires a new socket, like the LGA1851 one. According to ECSM, Arrow Lake might land between the fourth quarter of 2024 and the first quarter of 2025.

Intel's 4th Generation Sapphire Rapids Xeon processors haven't been on the market for a year. Yet, there are already rumors of a refresh. However, the refresh is strictly for the Sapphire Rapid WS series for workstations, the current W-2400 and W-3400 lineups. If the leaked information is accurate, Sapphire Rapids Refresh could arrive in early 2024 in the form of the W-2500 and W-3500 series. The new workstation processors will retain support for W790 motherboards. They'll rival AMD's upcoming Threadripper 7000 chips that wield Zen 4 cores.

Intel has shared a slew of new benchmarks of its fourth-gen Xeon Scalable Sapphire Rapids CPUs going head-to-head with AMD's fourth-gen EPYC Genoa processors, claiming up to 7 times more performance in AI workloads when comparing two 32-core chips. Intel also touts higher performance under certain conditions, like when Sapphire Rapids' in-built accelerators are brought into play, in a spate of standard general-purpose workloads. Intel's 56-core Xeon Max, the first x86 data center CPU with HBM memory, also takes on AMD's 96-core flagship in several HPC workloads, matching or exceeding AMD's bulkier chip.

Intel's performance comparisons come well after the company's launch of its Sapphire Rapids Xeons back in January of this year, but the company says its benchmark comparisons were delayed due to difficulties procuring AMD's competing EPYC Genoa chips, which launched in November of last year. The benchmarks come a day before AMD's AI and Data Center event that we're flying out to cover, so we'll attempt to get AMD's feedback about Intel's benchmarks while we're at the event.

With a few shipping OEM systems powered by AMD's Genoa in hand, Intel has conducted a wide range of benchmarks in multiple types of workloads spanning AI, HPC, and general-purpose workloads, to present its view of the competitive landscape. However, as with all vendor-provided benchmarks, these should be approached with caution. Intel claims it enabled all rational optimizations for both its and AMD's silicon for these tests, but be aware that the comparisons can be a bit lopsided, which we'll call out where we see it. The price of the chips used for comparison are also lopsided, too. We've included Intel's full test notes for the tested configurations in the relevant image albums below. With that, let's take a closer look at Intel's results. 

AI Workloads: Intel Sapphire Rapids Xeon vs AMD EPYC Genoa

For nearly every large organization, the question is no longer "if" or "when" they should deploy AI-driven applications into their deployments — the question is where and how. Yes, AI training remains the land of GPUs and various flavors of custom silicon, and we can expect Large Language Models (LLMs) to continue to rely upon those types of accelerators for the foreseeable future, but the majority of AI inference workloads still tend to run on CPUs. Given the quickening pace of AI infusion in the data center, the CPUs' performance in various types of inference will only become more important in the years to come.

Intel has had its eyes on accelerating AI workloads since the debut of its DL (Deep Learning) Boost suite with its second-gen Cascade Lake Xeon Scalable chips in 2019, which it claimed made them the first CPUs specifically optimized for AI workloads. Those chips came with support for new VNNI (Vector Neural Network Instructions) that optimized instructions for the smaller data types prized in AI applications. One of Intel's bedrock principles behind its AI strategy has been to use AVX-512 to vastly improve Xeon's performance and power efficiency in AI workloads by using VNNI and BF16. Intel's focus on AI acceleration features, including software optimizations, has expanded over the years to now include purpose-built AI acceleration engines on its Sapphire Rapids chips — if you're willing to pay the extra fee.

But a more important development lurks in the Sapphire Rapids silicon — Intel has now progressed to its new Advanced Matrix Extensions (AMX) x86 instructions, which deliver tremendous performance uplift in AI workloads by using a new set of two-dimensional registers called tiles. The Tile Matrix Multiply Unit (TMUL) that powers AMX is native to the Sapphire Rapids chips — you don't have to pay extra to use it like you do the dedicated AI accelerator engine — and leverages BF16 and INT8 to perform matrix multiply operations that can vastly enhance AI performance.

The benchmarks above leverage Intel's AMX, and not the optional in-built AI accelerator engine. Intel claims a 7X advantage over EPYC Genoa in ResNet34, a 34-layer object detection CNN model, using INT8 instructions at a batch size of 1 to measure latency — in this case, with an SLA of sub-100ms. Intel also claims a ~5.5X advantage in this same workload with a batched test. This model is trained in PyTorch but converted to the ONNX format.

Intel claims a ~3.3X advantage over AMD in ResNet50 (INT8 BS1) image classification with a sub-15ms SLA, and a 3X advantage in DLRM, a Deep Learning Recommendation Model, with PyTorch BF16 and INT8 in a batched workload.

We also see a ~5.5X advantage in BertLarge natural language processing with BF16, but that is versus Genoa with FP32, so it isn't an apples-to-apples test. Intel notes that BF16 datatypes were not supported with AMD's ZenDNN (Zen Deep Neural Network) library with TensorFlow at the time of testing, which leads to a data type mismatch in the BertLarge test. The remainder of the benchmarks used the same data types for both the Intel and AMD systems, but the test notes at the end of the above image album show some core-count-per-instance variations between the two tested configs -- we've followed up with Intel for more detail [EDIT: Intel responded that they swept across the various ratios to find the sweet spot of performance for both types of chips].

Overall, Intel claims that its AMX acceleration provides huge speedups in performance with industry-standard frameworks, but it's also important to call out the efficiency claims. The second yellow bar for each benchmark quantifies Intel's performance-per-watt claims, an incredibly important metric in today's power-constrained data centers — particularly with rising power costs in some climes, like the EU. Intel claims AMX delivers massive efficiency advantages when comparing two chips with similar core counts, which is surprising in light of Genoa's more advanced process node that tends to give it an efficiency advantage. Yes, dedicated silicon, as we see for AVX-512 and AMX, can be costly in terms of die area, and, thus, overall cost, but the advantages are huge if the applications can leverage the accelerators appropriately. 

Intel's 32-core Xeon Platinum 8462Y+ chip squares off with AMD's 32-core EPYC Genoa 9354, but be aware that while these are iso-core-count comparisons, Intel lists the 8462Y+ for $5,945 while AMD lists the 9354 for $3,420, so the Intel chip costs 74% more. That said, the list pricing from both vendors is usually not reflective of what customers (particularly Tier 1 customers) actually pay, so take the pricing as a fuzzy guideline.

AMD does tend to offer higher core count chips at any given price point in the product stack and has a higher peak core count of 96 compared to Intel's 56 cores. Intel stuck with a 32-core vs 32-core comparison here, with per-core software licensing fees being the company's rationale for why these remain comparable. In fairness, software licensing, and other server BOM costs, like DDR5 memory and GPUs/accelerators, do have an outsized impact on solution-level pricing.

Given the size of Intel’s AI performance advantage with the 32-core parts, it’s reasonably safe to assume that it can compete with higher core count EPYC chips in these AI workloads, even if saddled with a reasonable amount of subpar workload/power scaling as they move to their own higher core count parts. Much of this performance can be chalked up to Intel's efforts on software enablement.

Intel doesn’t have any benchmark comparisons with LLMs (of the smaller variety), largely due to the raw and rapidly-changing nature of the LLM landscape. However, the company does say it is seeing impressive results with bandwidth-hungry LLMs on its HBM-equipped Xeon Max models (more below), which could be interesting given GPU shortages. We’re told that LLM benchmarks with Xeon Max will come in the future.

General Workloads: Intel Sapphire Rapids Xeon vs AMD EPYC Genoa

Here we see a standoff between the same two chips in a broader spate of general workloads. You'll note that Intel shows itself lagging Genoa in SPECint and SPECfp by large margins in efficiency and smaller margins in performance, but the company has recently begun de-emphasizing those industry-standard SPEC benchmarks because it feels the aging benchmark no longer represents real-world customer workloads -- that's actually not an uncommon viewpoint in the industry. Intel's take on SPEC is evidenced by the slideware they shared during the Sapphire Rapids briefings (second slide in above album). Intel says it is working on helping define the next generation of the venerable SPEC to make it more representative. For SPEC, Intel used the ICC compiler for Sapphire Rapids and AOCC for EPYC. The remainder of the benchmarks use 'mostly' the GCC compiler for both types of chips.

Intel also trails in the stream triad memory benchmark, which isn't surprising given that Sapphire Rapids has eight memory channels while Genoa has 12 (due to its higher channel count, AMD has 50% more memory capacity at its disposal in all of these benchmarks). That disparity also gives AMD an advantage in several other memory-bound workloads, like HPCG and Ansyst Fluent/Mechanical (not shown). It's clear that Intel is avoiding memory-bound workloads in these comparisons, though we do see those types of workloads in the HPC section where the company employs Xeon Max for comparison.

Intel largely claims slight performance gains in these workloads but notes that some of the larger gains come from employing its on-chip accelerator engines that require an extra fee to unlock. Intel used these engines in several of the workloads, marked in bronze on the horizontal axis, like SQL HammerDB, Gromacs, LAMMPs, NAMD, and others, to highlight the advantages of using the engines in tandem with applications that are tuned to exploit the advantages.

We haven't seen many benchmarks with these engines fully employed via software support, so it's clear that Intel is moving forward on the enablement front, and to good effect — at least according to its own benchmarks. Just be mindful that these accelerators, which speed AI, encryption/decryption, compression, data movement, and analytics, aren't available consistently across the Sapphire Rapids product stack and also cost extra.

There's also the matter of DPUs, too. These discrete accelerators can perform many of the same functions, such as encryption/decryption, compression, and data movement, at higher performance levels while also offloading the CPU, managing network traffic, and providing a separate control plane. Advanced DPU-augmented data center architectures aren't as widespread as the data center architectures that we consider 'traditional,' but that means the relative value of Intel's on-die accelerator engines will vary by deployment.

HPC Workloads : Intel Xeon Max vs AMD EPYC Genoa

Here we see Intel's 56-core Xeon Max, the first x86 data center chip with on-package HBM to be pushed into full production, face off with AMD's 96-core EPYC Genoa flagship. Intel claims the onboard memory gives its smaller-but-not-less-expensive chip the advantage over the EPYC processor in a range of HPC-centric workloads, but without on-board accelerators enabled (oddly, only the data streaming accelerators are available on the Xeon Max models). 

Naturally, these benchmarks won't land without plenty of dispute - there's a whole slew of architecture-specific optimizations possible for AMD's chips that could extract more performance, particularly in heavily-optimized HPC environments - but the key takeaway here is that Intel claims its HBM-equipped CPUs can offer compelling advantages in HPC that rival AMD's core-heavy flagships.

It is worth noting that both power scaling and chip fabrics can become a constraining factor with higher core-count chips in these types of workloads. As such, it's conceivable that some of these workloads could shift to a more favorable outcome for AMD with a slightly less prodigious model, like its own 64- or 56-core Genoa’s. Also, Intel's chips here only use the HBM2e memory -- there is no DDR5 attached. As such, AMD's Genoa has a tremendous memory capacity advantage (1.5TB of DDR5 capacity vs 128GB of HBM2e capacity for Xeon Max).

AMD has taken somewhat of a different path of augmenting its chips with extra memory by 3D-stacking the L3 cache for its Milan-X processors, and we expect the company to announce its newer Genoa-X chips tomorrow at its event tomorrow. However, AMD gears Milan-X for certain technical workloads, as opposed to positioning it for the broader HPC market as Intel does for Xeon Max.

TCO: Intel Sapphire Rapids Xeon vs AMD EPYC Genoa

Every vendor has its own take on how to calculate the Total Cost of Ownership (TCO), but they're all typically questionable -- overall TCO will vary so widely by deployment that it's truly hard to derive solid metrics that are indicative of broader trends.

Intel's take is that using the SPEC benchmark for deriving TCO values isn't indicative of real-world use cases, which we touched on briefly in the general workloads section above. As such, Intel provides a range of examples here for different types of deployments, and the associated savings of using Intel chips over AMD's EPYC. Some of these comparisons, like the QAT backup, aren't much better than using SPEC, so take them with a big grain of salt and be sure to check the configuration slides we've included in the album.

Intel also points to an accelerating pace of deployment, with 200 designs of 450+ Sapphire Rapids design wins already shipping. Intel also points out that it already has cloud instances in general availability at Google Cloud (notably paired with its Mount Evans IPU/DPU), and preview instances already available at several other cloud providers, like AWS. In contrast, AMD has yet to have any publicly available Genoa instances in the US, preview or otherwise, at cloud providers (we are aware of public instances available in China). We'll inquire about the status of AMD's cloud efforts at its event tomorrow and update as necessary. 

Finally, Intel reiterates that its recently-revised data center roadmap remains on track, which you can read about here. We have a few questions still pending with Intel about the above benchmark test configurations, and will add more info as it becomes available.

Sapphire added a new Radeon RX 7900 XT, one of the best graphics cards, to the company's lineup. The Radeon RX 7900 XT 20G Ultra Platinum L arrives with a somewhat revamped exterior and cooling system.

The Radeon RX 7900 XT 20G Ultra Platinum L features a similar design to the Radeon RX 7900 XTX Nitro+ Vapor-X available in the U.S. market. However, there are some significant changes with the former. The Radeon RX 7900 XT 20G Ultra Platinum L lacks the Vapor-X moniker, meaning the graphics card doesn't leverage Sapphire's vapor chamber to cool the GPU and memory. Instead, the Radeon RX 7900 XT 20G Ultra Platinum L utilizes a standard copper base plate. As a result, it doesn't provide the same thermal performance, but it's cheaper, so Sapphire can reduce the overall product cost.

The different cooling system doesn't affect the dimensions of the Radeon RX 7900 XT 20G Ultra Platinum L. It's still a 3.5-slot monster that measures 320 mm long, just like the flagship Radeon RX 7900 XTX Nitro+ Vapor-X.

Sapphire has also cut some corners with the RGB design. For example, the Radeon RX 7900 XT 20G Ultra Platinum L only has one RGB strip on the shroud, whereas the Radeon RX 7900 XTX Nitro+ Vapor-X has two. However, it doesn't affect the appeal of the graphics card since the Radeon RX 7900 XT 20G Ultra Platinum L still looks very attractive, even without the extra RGB lighting.

The Radeon RX 7900 XT 20G Ultra Platinum L has the same 2,450 MHz boost clock speed as the Pulse Radeon RX 7900 XT. The new cut-down variant offers marginally lower performance than the Radeon RX 7900 XTX Nitro+ Vapor-X. The margin isn't that significant, though, since the Radeon RX 7900 XT 20G Ultra Platinum L only has a 4% lower boost clock than the Radeon RX 7900 XTX Nitro+ Vapor-X.

The variation in the cooler and clock speeds are the only aspects that separate the Radeon RX 7900 XT 20G Ultra Platinum L from Sapphire's other models. The graphics card still depends on three 8-pin PCIe power connectors for external power and offers three DisplayPort 2.1 outputs and one HDMI 2.1 port for connecting displays.

Sapphire's cost-cutting measures allow the vendor to bring the Radeon RX 7900 XT 20G Ultra Platinum L's pricing down to the $899 MSRP. Chinese news outlet Expreview spotted the Radeon RX 7900 XT 20G Ultra Platinum L for $898.88 on JD.com. The Radeon RX 7900 XTX Nitro+ Vapor-X sells for $970.23 in China. That's a $71.35 difference, which essentially tells us that it's what the extra RGB strip and vapor chamber cost.

Authentic-looking product renders of a Sapphire-branded Radeon RX 7600 have leaked. Meanwhile, a French tech site claims to know the Euro pricing for AMD's upcoming RDNA 3 architecture mainstream contender.

VideoCardz shared Sapphire product photos, showcasing box art and some views of one of its Pulse series Radeon RX 7600 graphics cards. France's CowCotLand followed up with a tasty price rumor, asserting that the Radeon RX 7600 will land on the continent at 349 Euros ($379) on May 25.

From the box images, it is clear that the Sapphire Radeon RX 7600 comes with 8GB of VRAM, which aligns with previous leaks and rumors. Zooming into the picture of the box wrap, we also note that this model will feature "32 AMD RDNA 3 Compute Units" and 32MB of AMD infinity Cache to help compensate for the stingy memory bus. The section of the box art looks like it might be the source for yesterday's leaked RX 7600 specs news.

Elsewhere in the images shared by VideoCardz, we can see that the Sapphire graphics card is a traditional dual-fan-cooled GPU with a single 8-pin power connector. This should be ample for a GPU expected to pull somewhere in the region of 130W, much like its predecessor, the Radeon RX 6600.

Besides the power consumption of the desktop Radeon RX 7600 and the resulting higher base/boost clocks, we expected specs to coincide with the RX 7600M GPU, which went official back at CES 2023 in January. However, it now looks like the desktop RX 7600 has 32 CUs and 2,048 CUs, which echo the spec of the Radeon RX 7600M XT. 

In other words, the desktop version of the RX 7600 has about 15% more shader processors than the mobile version. Moreover, the move from RDNA 2 to RDNA 3 architectures is trumpeted as providing a roughly 25% performance uplift.

Euro Pricing Rumors

RX 7600 desktop graphics card listings were spotted three days earlier in Singapore. The local price converted to approximately $410. CotCowLand claims that the Radeon RX 7600 will be priced at €349 in France. The site doesn't provide any documentary, link, or screenshot evidence of this pricing, so it's best to add a lump of salt. 

A €349 price converts to $380 at today's exchange rates. Taking away French VAT (20%) gives us a $304. This could indicate that the AMD Radeon RX 7600 launch price will be set at $299 stateside.

Finally, this latest pricing rumor and the Singaporean one point the finger at a launch/release date of Thursday, May 25 or thereabouts.

When AMD's chief executive Lisa Su said last week that the company was on track to release its mainstream Radeon RX 7000-series graphics cards this quarter, she never said exactly when this would happen. But it appears that AMD's Radeon RX 7600-series graphics cards may be closer than we think as boards are already being listed in Asia. Soon, the RX 7600 will face off against the best graphics cards.

Right now, someone in Singapore is selling Sapphire's Radeon RX 7600 Pulse 8GB GDDR6 graphics card for S$545 (around $410, though it is unclear whether the price includes taxes) on Carousell, an online marketplace, as discovered by @Komachi_Ensaka. The advertisement claims that the stock arrives on May 26, 2023, which suggests that the seller expects to have multiple boards in a little more than a couple of weeks from now.

VideoCardz has received images of Sapphire's Radeon RX 7600 Pulse 8GB GDDR6 graphics board on a store shelf. The package reveals that the GPU features 32 RDNA 3 compute units (2048 stream processors) and 32MB of Infinity Cache. There is no word about the availability date of the product, but the card will reportedly cost $249, which contradicts to information from Carousell — but that could just be an attempt to capitalize on early demand.

In any case, the picture itself points to the fact that Sapphire, which is one of AMD's closest partners, has already started to ship its boards to retailers and it is only a matter of time before these parts are available worldwide.

AMD's Radeon RX 7600-series graphics cards are expected to be based on the Navi 33 GPU. It's hard to estimate performance of the vanilla Radeon RX 7600 and compare it to that of Nvidia's yet-to-be-released GeForce RTX 4060-series graphics boards, but we certainly expect these product families to compete against each other. Based on what we've seen from the RX 7900 XT versus the RX 6900 XT, if the RX 7600 sees a similar improvement in performance, it would only be around 25% faster than the existing RX 6600. We'll find out in the next month or so it seems.

We're dealing with unofficial information about Radeon RX 7600 availability date and contradictory information when it comes to its pricing, so obviously take all of this with a grain of salt. Still, it's evident that at least some of Sapphire's retail partners have already gotten their Radeon RX 7600 products, so it's logical to expect boards to hit the market rather sooner than later.

Noctua, a leading supplier in air cooling, is one of a few makers of CPU cooling systems with no closed-loop cooling systems in its lineup. And perhaps the company doesn't need them, as its latest air cooler, the NH-U14S DX-4677, is capable of cooling Intel's 56-core Xeon W9-3495X — which consumes 700W of power. 

These days, most enthusiast-grade desktops and workstations use all-in-one (AIO) liquid cooling systems, which are more efficient than air coolers and better able to handle modern CPUs (which can consume well above their rated thermal design power when working in turbo boost mode, albeit for short periods of time). But while liquid coolers are efficient, a large air cooler can also do the job pretty well, as Noctua's demonstration shows. The company's new NH-U14S DX-4677 is able to cool down Intel's 56-core Xeon W9-3495X under load when it consumes over 700W and does not throttle. 

"Would you say 700W on air cooling is impossible," the Noctua tweet reads. "Here's our NH-U14S DX-4677 cooling the Intel Xeon w9-3495X at a continuous load of more than 700W!" 

The company did not disclose whether the CPU was overclocked (or at what clock speed it was running), but in the video the temperatures remain below 100°C. We suspect its power limits were increased in UEFI BIOS to demonstrate the capabilities of the cooler. 

The Noctua NH-U14S DX-4677 is a large, aluminum tower cooler featuring a nickel-plated copper base along with six thick, U-shaped, nickel-plated heat pipes. The cooler measures 6.5 x 5.9 x 4.4 inches (165 x 150 x 111mm) and weighs 2.5lbs (1.14kg), including two Noctua NF-A15 HS-PWM fans, so it may not fit into compact PC cases. The heatsink fins are slightly curved to maximize surface area and optimize airflow. 

The cooler features a total fin surface area of more than 6000cm², which makes it one of the largest air coolers available. Noctua says that the two 140mm fans generate airflow of around 140,2 m³/h and produce up to 24.6 dbA noise. 

The Noctua NH-U14S DX-4677 was specifically designed for Intel's Xeon processors, and it offers both impressive cooling performance and reliability with a six-year warranty (and an MTTF of 150,000 hours on the fans).

The Radeon RX 7900 XTX and XT launch wasn't quite enough to claim the pole position in our GPU benchmarks hierarchy, and now we've learned a few batches of reference 7900 XTX cards shipped with vapor chamber issues. The Sapphire RX 7900 XTX Nitro+ Vapor-X has no such problems, promising the usual bump in performance thanks to higher clocks and a better cooler — and slightly higher power draw as well. How does it fare when compared with the best graphics cards, and can you find cards in stock anywhere?

As with other third-party AIB (add-in board) partner cards, the only real changes relative to the reference designs will come in the form of cooling and aesthetics, potentially with a bump in clock speeds and power limits. Sapphire's RX 7900 XTX Nitro+ follows that pattern to a T: 

Sapphire prices its Vapor-X model $100 above the reference card, with a $1,099 price point. MSRPs mean next to nothing, unfortunately, and Newegg has a listing for $1,522.97 — plus $49.99 in shipping, to add insult to injury. (It's from a third party seller, naturally.) Not that the reference 7900 XTX is any better, currently listed for $1,439.99 from a different third party seller.

Hopefully the supply and pricing will improve in the coming months, but right now the 7900 XTX cards are all quite overpriced. The 7900 XTX typically beats the RTX 4080 in rasterization performance while trailing by a significant amount in complex ray tracing games, but RTX 4080 cards start at $1,269.99. That's more than we'd like but also about $200 less than the 7900 XTX for the time being.

If you want to learn more about AMD's RDNA 3 architecture, that article's a good place to start. Sapphire's design doesn't radically change anything, so we mostly know what to expect, but let's go ahead and check out the card itself and take it apart. 

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

I made fun of the box art on Sapphire's RX 6950 XT Nitro+ Pure, and I'm happy to say the blue and black artwork on the 7900 XTX is less ugly. Not that it really matters once you unpack the card.

There's plenty of padding around the card to protect it during shipping, plus a support bracket to help prevent card sag and an ARGB cable that can be used to let your motherboard control the card's lighting.

A quick word on the support bracket: It's a bit finicky to install, since it has overlaps on at least one of the screws that holds the graphics card in place. It has three positions for screws, but the width of the 7900 XTX Vapor-X means you'll only use one of them (and perhaps one of the empty IO slots below the card). The good news is that it won't slide around when you're moving the case, but for someone like me that regularly swaps GPUs, it's less convenient.

The card incidentally is quite large. It measures 321 x 134 x 72 mm and weighs 1948g. That's lighter and slightly smaller than most of the third-party RTX 4090 cards we've reviewed of late (see: Asus ROG Strix 4090, Colorful 4090 Vulcan, Gigabyte 4090 Gaming OC, and MSI 4090 Suprim Liquid-X), but not by much — it's also slightly longer and thicker than the 4090 Founders Edition. You'll need 3.25-slots of clearance, though at least there's a triple-slot IO bracket to help support the card.

Sapphire uses a fairly common triple-fan cooling setup with custom 95mm fans. The blades on these fans are also a new innovation, with an angular design that we're told helps improve airflow and reduce noise levels. The fans also include an integrated rim to improve static pressure and airflow, and finally the center fan spins clockwise while the two outer fans rotate counterclockwise, which reduces turbulence and noise.

Sapphire takes a slightly different approach on the display outputs, with dual HDMI 2.1 and dual DisplayPort 2.1 (54 Gbps). The DP2.1 outputs technically support a slightly higher bandwidth… but only if you have a DP2.1 capable monitor, which are hard to come by. Even our Samsung Odyssey Neo G8 32 with its 4K 240 Hz support still uses DP1.4 (with Display Stream Compression for 240 Hz).

Sapphire does use a triple 8-pin power connector design, and part of that is because the Vapor-X is rated for 410W — 55W more than the reference 7900 XTX. As usual, you get greatly diminishing returns for the increased power. We'll hit the numbers in a bit, but the Vapor-X consistently uses about 50W more power, a 15% increase, while performance typically only improves by up to 2%.

Sapphire RX 7900 XTX Nitro+ Vapor-X Teardown

Dismantling the Sapphire Vapor-X isn't particularly difficult, though there are a few extra steps. There's a reinforcing frame around the cooling, and the best order of disassembly isn't immediately obvious. I took it apart one way, but realized in the process that it would have been better to take a slightly different approach. The gallery below shows, more or less, how I proceeded, but I'll note where things should diverge.

The first thing I did was to remove the ten Phillips head screws securing the backplate, but you're perhaps better off removing the top of the metal shroud with the three fans first. That's images five through nine in the above gallery. You'll need an Allen wrench set to do that, I'm not sure of the exact size I used, but it's relatively small, like in the 2mm range.

Being able to get easy access to the fans does have an advantage, as those are one of the most common failure points on graphics cards. Take out the four screws as shown and the top should come off, at which point you'll also want to disconnect the fan and RGB LED cables.

You can proceed with the rest of the teardown from here, starting with the backplate. There's another LED cable connecting the plate to the PCB, so be careful to disconnect that before pulling the plate away.

Next, there are four more (smaller) Phillips head screws on the IO bracket that you'll want to remove. Then there are four more screws that hold on the back pressure plate, two of which have stickers that may or may not void your warranty. Once all of these are out of the way, you should be able to pry the heatsink off the GPU and GDDR6 memory, and the extra heatsink/frame should also come off.

That's about it for the teardown, and you can see the GPU, inductors, VRMs, GDDR6, and other components in the rest of the pictures. Our particular sample uses twelve SK hynix 16Gb (2GB) memory packages labeled "H56G42AS8D X014" that corresponds with the "10.0 GHz" memory listed at the website.

There are 20 relatively large VRMs present, seven of which are LR12 modules and the remaining 13 are LR15. Sapphire appears to be using a 17+3 phase VRM design (though feel free to correct me if I got that wrong).

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Sapphire RX 7900 XTX Nitro+ Vapor-X Overclocking

Overclocking of AMD's new RX 7900 cards is a bit unusual, in that the maximum boost clock is often set far higher than anything you'll ever achieve in practice. The reference 7900 XTX for example has a 3115 MHz boost clock, but typically runs in the 2.5–2.6 GHz range. Increasing the boost clock to 3300 MHz (or more) does basically nothing, as the GPU is smart enough that it never even tries for those higher clocks.

With the Sapphire RX 7900 XTX Vapor-X, the factory stock boost clock is actually slightly lower than on AMD's reference card: 2970 MHz. You can increase that to something higher, but it mostly goes unused. Increasing the power limit can do more for performance than adjusting the boost clock.

We used our normal trial and error process to attempt to dial in a stable overclock. Since the GPU clock didn't matter much, we just maxed out the power limit at 115% and set the GPU for 3100 MHz, and then attempted to find a stable memory overclock. The chips are officially rated for 20 Gbps ("2498 MHz" in Afterburner), but there's usually a fair amount of headroom, and we ultimately settled on a "2750 MHz" setting, which equates to a 22 Gbps memory speed.

Peak power draw spiked at over 600W during a FurMark stress test at these settings, while gaming showed a maximum power use of up to 550W. Average power use was quite a bit lower, just under 500W, but this is basically within spitting distance of the RTX 4090. At stock, both peak and average power use dropped by around 65W. In short, you don't want to run this sort of card on an insufficient PSU, and 1000W or higher models (perhaps one of the best power supplies) are a good idea considering the high transients.

Fine tuning of voltages using AMD's Radeon Settings could result in better figures, but for these tests we stopped here and set about benchmarking at 4K ultra in our test suite. You'll find the overclocked numbers from the 7900 series cards in the charts in green.

Sapphire RX 7900 XTX Test Setup

We updated our GPU test PC and gaming suite in early 2022, but with the RTX 40-series launch, we found more and more games were becoming CPU limited at anything below 4K. As such, we've upgraded our GPU test system again.

We're including the performance results for the RTX 4070 Ti in the charts, along with the reference 7900 XTX/XT, RTX 4080/4090, and the previous generation RTX 3090 Ti and RX 6950 XT. We used the latest AMD and Nvidia drivers available at the time of testing: 22.12.2 for the 7900 series, 22.11.2 for other AMD GPUs; 527.62 for the Nvidia GPUs.

Since this is a third-party custom card, we're only testing at 1440p and 4K in our gaming suite. Results in professional applications and at 1080p can be found in our initial RX 7900 XTX and XT review. 

AMD and Nvidia both recommend either the AMD Ryzen 9 7950X or Intel Core i9-13900K to get the most out of their new graphics cards, and for the third party cards we're going to stick with the 13900K. (The Ryzen 9 7950X provided similar performance on the reference 7900 cards.) MSI provided the Z790 DDR5 motherboard, G.Skill got the nod on memory, and Sabrent was good enough to send over a beefy 4TB SSD — which we promptly filled to about half its total capacity. The be quiet! 1500W PSU is ATX 3.0 compliant and 80 Plus Titanium certified.

We also have Nvidia PCAT v2 (Power Capture and Analysis Tool) hardware that allows us to capture the real-time power use, GPU clocks, and more during all of our gaming benchmarks. We'll have the results on the power testing page.

For all of our testing, we've run the latest Windows 11 updates. Our gaming tests now consist of a standard suite of nine games without ray tracing enabled (even if the game supports it), and a separate ray tracing suite of six games that all use multiple RT effects.

We've tested the Sapphire and other GPUs at 4K and 1440p, using "ultra" settings — basically the highest supported preset if there is one, and in some cases maxing out all the other settings for good measure (except for MSAA or super sampling). We've also hooked our test PCs up to the Samsung Odyssey Neo G8 32, one of the best gaming monitors around, just so we could fully experience some of the higher frame rates that might be available — with FreeSync enabled.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

With a suggested price that's nearly the same as the RTX 4080, and with street prices currently hovering above the 4080 cards, 4K performance should be a major consideration for anyone looking at the 7900 XTX cards. We'll start there.

With the current seven fastest GPUs available, plus the Sapphire 7900 XTX, Sapphire's card ends up ranking second place overall in our rasterization benchmarks (not counting the manually overclocked cards). At the same time, it's barely any faster than the reference 7900 XTX — and we should note that our reference card sample doesn't seem to be one of those affected by the vapor chamber defect.

There's a bit of variability in our testing, and the Sapphire card was up to 7% faster than the reference model in Red Dead Redemption 2 and Total War: Warhammer 3, but it was also a few percent slower in Forza Horizon 5 and Horizon Zero Dawn. Bottom line was mostly a wash, with a 2.2% advantage across the nine tested games.

Overclocking is a different story. Where the gains from the factory overclock are almost negligible, the manual overclock improved performance by 7% overall. That's still not a massive change, but relative to the overclocked reference XTX card it was also a 4% gain. Those gains do come at the cost of power and efficiency of course, and we'd generally recommend sticking to the default clocks or even toying with underclocking or undervolting as a better option.

Meanwhile, looking at Nvidia, the Sapphire card leads the RTX 4080 by 6% overall, with larger wins in Borderlands 3, Far Cry 6, and Watch Dogs Legion. Nvidia's biggest win is in Total War: Warhammer 3 where it was 6% faster than the Sapphire card.

Ray tracing of course turns the tables, and the best Sapphire and AMD can do is fifth place, not counting the manually overclocked cards. The Sapphire card is also just 0.6% faster overall compared to the reference model, and the individual games show the two cards trading blows. The extra power and clocks ultimately don't show up in our factory stock testing.

Versus Nvidia, the Sapphire card is now 27% slower than the RTX 4080, and also trails the RTX 4070 Ti by 6%. It actually takes a slight lead over the 4070 in half of the games, but the deficits in Minecraft (-29%) and Cyberpunk 2077 (-11%) are quite a bit larger. And that's not accounting for DLSS, which most demanding DXR games support — though some also support FSR2, which at least partially closes the gap.

It's the usual story of AMD being a fine choice for rasterization, while Nvidia generally dominates in ray tracing performance. If you don't care at all about ray tracing, that's fine, and the visual difference in a lot of current games isn't even that big. Still, there are enough games with decent ray tracing support that we wouldn't completely write it off as "unnecessary."

Overclocking shows the same gains as in rasterization, incidentally: 7% faster compared to the default factory clocks, and 4% faster than the overclocked reference XTX card. The overclock is enough to just push past the (stock) Asus RTX 4070 Ti card in overall performance, though Minecraft still heavily favors the Nvidia GPU. 

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Overall performance at 1440p improves by 52% compared to 4K gaming, and quite a few games are exceeding 144 fps — great if you have one of the best gaming monitors. But while those gains are nice to see, Sapphire's relative performance lead over the reference XTX shrinks to just 1%. You definitely won't notice the performance difference while gaming.

The lead over the RTX 4080 in rasterization games also shrinks to just 4% overall, with Horizon Zero Dawn and Total War: Warhammer 3 still favoring Nvidia's GPU. In contrast, Borderlands 3, Far Cry 6, and Watch Dogs Legion continue to favor AMD by larger margins. CPU limitations are starting to come into play as well, so things like Nvidia's DLSS 3 are potentially more useful (though not the same as true rendered frames).

Ray tracing at 1440p doesn't show any real benefit to the Sapphire card over the reference 7900 XTX. Overall performance is down by 0.7% — tied, in other words — with Metro Exodus Enhanced registering a 6% drop while Bright Memory Infinite showed a 3% improvement.

The new RTX 4070 Ti beats AMD's best at ray tracing as well, by 6% this time, which indicates most of the games are GPU purely GPU limited. That may not be entirely true in all cases, as the CPU has to do a fair amount of work building the BVH (Bounding Volume Hierarchy) structure that gets passed to the GPU, but the RT calculations are so complex that any additional load on the CPU mostly doesn't show up in the final performance.

Two of the games (Bright Memory Infinite and Cyberpunk 2077) show a slightly larger gap between the 4070 Ti and the Sapphire card, while the gap narrows in two of the other games (Metro and Minecraft), with the remaining two being effectively tied relative to 4K performance.

We measure real-world power consumption using Powenetics testing hardware and software. We capture in-line GPU power consumption by collecting data while looping Metro Exodus (the original, not the enhanced version) and while running the FurMark stress test. We also check noise levels using an SPL meter. Our test PC remains the same old Core i9-9900K as we've used previously, to keep results consistent.

We also test on our newer 13900K PC using PCAT v2 hardware in all of our gaming benchmarks, which gives a wider view of power use and efficiency. We'll start with the gallery of our PCAT results — note that FPS/$ uses the official MSRPs for the RX 7000- and RTX 40-series cards, while FPS/W (ie, efficiency) uses the measured power consumption.

We have results for our 1440p and 4K testing, along with the manually overclocked 4K testing, except on the RTX 4080. That was originally reviewed on our previous i9-12900K test PC, and while we retested standard performance on the 13900K system, we didn't retest with overclocking, so those lines are blank (but we left them in place so the images are all the same dimensions).

There's a lot to parse in the above tables, but if we just focus on the Sapphire card for the time being, what you'll see is that performance per watt and performance per dollar both drop compared to the reference card. That's no surprise, since the Sapphire card consumes more power, costs more, and offers a barely measurable change in performance in most of our testing.

Looking at the bigger picture, the RTX 4080 delivers the best efficiency out of the cards we've tested so far, while the RTX 4070 Ti offers the best "value" in terms of FPS per dollar. That's not to say it's a great value, or that other cards like the RX 6650 XT wouldn't beat it (we haven't retested that on the new PC yet), but of the latest generation GPUs, the RTX 4070 Ti delivers slightly more FPS per dollar spent — and it would have done much better if it had inherited the RTX 3070 Ti price!

Of course, that's predicated on any of the cards being available at MSRP. Newegg does list a couple of RTX 4070 Ti cards at $799.99, in stock, and the same goes for the RX 7900 XT, with the XFX reference model actually priced $20 below MSRP right now. The RTX 4080 starts at $1,269.99 (there's a PNY model that's also backordered at MSRP). The RX 7900 XTX meanwhile starts over $400 above MSRP, making it far and away the worst value at present. 

Our previous power testing results are still valid, though they're limited to two scenarios: FurMark and Metro Exodus. Both test scenarios last about 10 minutes, though Metro does have a "break" between loops that allows the GPU to recover slightly.

Sapphire's RX 7900 XTX Nitro+ ends up eclipsing everything except for the RTX 4090 and RTX 3090 Ti when it comes to power use, and it's not that far behind those in our gaming test. Power draw averaged 425W in Metro, just 6W less than the 3090 Ti and 10W less than the 4090, and also 73W more than the reference 7900 XTX and 128W more than the RTX 4080. So much for those efficiency gains AMD was talking about with RDNA 3 — they clearly don't apply when you redline the GPU.

You can of course opt to save power by switching out of the default performance mode, which might not be a bad idea. Manual overclocking, as we showed above, would only make things worse.

The clock speeds help to explain the higher power draw, as, in FurMark, Sapphire's card averaged just over 2 GHz, compared to 1.66 GHz on the reference model. While gaming, it also averaged 2.8 GHz compared to 2.56 GHz on the reference card.

We don't usually dig too much into the voltages, but it's worth mentioning that in our Metro testing, the reference card had an average VDDC of 0.895V while under load, compared to the Sapphire card's 0.988V for the same test sequence. That extra 0.93V makes a rather large difference in power use when combined with the higher clock speeds.

The good news is that temperatures are basically the same on the Sapphire card and on the reference XTX, despite the higher power draw. Both top out in the 65C range, with the reference XTX running its fans at up to ~1650 RPM compared to just 1450 RPM for the Sapphire card. That impacts the noise level as well.

We check noise levels using an SPL (sound pressure level) meter placed 10cm from the card, with the mic aimed right at the center fan. This helps minimize the impact of other noise sources like the fans on the CPU cooler. The noise floor of our test environment and equipment is around 32 dB(A).

After running Metro for over 15 minutes, the Sapphire RX 7900 XTX settled in at a fan speed of 38% and a noise level of 42.2 dB(A). That's only slightly higher than some of the quietest cards we've tested, which land at around 39 dB(A). The reference card meanwhile measured 49.9 dB(A), which is clearly audible over the ambient room noise. We also tested with a static fan speed of 75%, where the Sapphire generated 56.6 dB(A) of noise, so there's plenty of room for additional airflow and cooling if the fans need to ramp up — not that you'd normally want that. 

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

The Sapphire RX 7900 XTX Nitro+ Vapor-X — because who doesn't like extra descriptors in a product name — ends up as a potentially interesting card, but only if prices come down. That goes for all RX 7900 XTX models, incidentally. While you can make the argument that the XTX outperforms the RTX 4080 in rasterization performance, it lacks some of the features that Nvidia's latest GPUs provide and certainly doesn't warrant the currently higher price.

Sapphire's top variant packs on a massive cooler that does its job well. It requires a large case and has some very bright RGB strips, which some people will love and others might not appreciate. If you don't want the RGB but also don't want the reference 7900 XTX, Sapphire makes a 7900 XTX Pulse, which omits all the RGB bling but changes the fans and cooler and has a lower boost clock of 2525 MHz (compared to 2680 MHz on the Vapor-X). But given what we've seen from performance so far, there's probably not much difference there.

There's plenty to like with the Vapor-X, but a lot of it ultimately feels like overkill. Power use was consistently 50W higher than the reference 7900 XTX, a 14% increase that only delivered about 2% better performance on average. You could run with the secondary BIOS that lowers clocks and power use (or use Sapphire's TriXX software to switch modes), but then why buy a premium model for reference performance?

Okay, you can do that for the lower noise levels and improved thermals. Buy it for the looks and features, in other words, and not for any potential improvement in performance. And when you're already looking at a graphics card that costs a grand, what's another ten percent to get a design you like more — or not?

Unfortunately for gamers everywhere, right now isn't a great time in the world of consumer GPUs, despite having five new generation cards launch in the past three months. The issue is that prices have shot up across the entire gamut, so rather than getting better performance than the previous generation at a lower price, the best we can do is better performance at a similar or higher price — and it's mostly higher prices.

Third-party AIB cards like the Sapphire Nitro+ Vapor-X aren't spared, as they cost even more than the reference cards or models that don't pack in quite as many extras, usually for minimal performance gains. If we look at generational upgrades, you get about 40% higher performance than last generation's RX 6950 XT (at 4K). However, while the official launch MSRPs might be similar, the cheapest RX 6950 XT currently costs $699, while the best we can find for the 7900 XTX is $1,299 for Sapphire's reference card. That's 86% more expensive for 40% higher performance, definitely not the direction we want to be heading.

There was a big outcry over Nvidia's RTX 4070 Ti launching at $799, and deservedly so. But the RX 7900 XTX is no better and actually ends up quite a bit worse at current prices. Again, hopefully things will settle down in the coming weeks and months, because this is nearly as bad as what we saw during the height of the cryptocurrency mining boom in 2021. Frankly, no one should be paying such prices for these latest generation GPUs.

If you can find the Sapphire Vapor-X for close to its official $1,099 MSRP, it's far more tolerable for the top performing AMD solution. Until and unless that happens, we recommend sitting tight with whatever GPU you currently have, or perhaps consider something like the RX 6700 XT or 6750 XT. Sure, that's only a bit more than half the performance, but for less than one third the price it's a far more sensible option. 

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

After years of delays, Intel formally launched its fourth-gen Xeon Scalable Sapphire Rapids CPUs, in both regular and HBM-infused Max flavors, and its "Ponte Vecchio" Data Center GPU Max Series today. Intel's expansive portfolio of 52 new CPUs will face off with AMD's EPYC Genoa lineup that debuted last year. The company also slipped in a low-key announcement of its last line of Optane Persistent Memory DIMMs.

While AMD's chips maintain the core count lead with a maximum of 96 cores on a single chip, Intel's Sapphire Rapids chips bring the company up to a maximum of 60 cores, a 50% improvement over its previous peak of 40 cores with the third-gen Ice Lake Xeons. Intel claims this will lead to a 53% improvement in general compute over its prior-gen chips, but largely avoided making direct comparisons to AMD's chips during its presentations. However, Intel has provided samples to the press for unrestricted third-party reviews, so it isn't shying away from the competition.

Sapphire Rapids leans heavily into new acceleration technologies that can either be purchased outright or bought through a new pay-as-you-go model. These new purpose-built accelerator regions of the chip are designed to radically boost performance in several types of work, like compression, encryption, data movement, and data analytics, that typically require discrete accelerators for maximum performance.

Despite having a clear core count lead, AMD doesn't have similar acceleration features for its Genoa processors. When employing the new accelerators, Intel claims an average 2.9X improvement in performance-per-watt over its own previous-gen models in some workloads. Intel also claims a 10X improvement in AI inference and training, and a 3X improvement in data analytics workloads.

Intel's Sapphire Rapids, which comes fabbed on the 'Intel 7' process, also brings a host of new connectivity technologies, like support for PCIe 5.0, DDR5 memory, and the CXL 1.1 interface (type 1 and 2 devices), giving the company a firmer footing against AMD's Genoa. We're hard at work benchmarking the chips for our full review that we will post in the coming days, but in the interim, here's a brief overview of the new lineup.

Intel 4th-Gen Xeon Scalable Sapphire Rapids Pricing and Specifications

Intel's Sapphire Rapids product stack spans 52 models carved into 'performance' and 'mainstream' dual-socket chips for general-purpose models. There's also specialized models for liquid-cooled, single-socket, networking, cloud, HPC, and storage/HCI systems. As a result, it feels like there's a specialized chip for nearly every workload, creating a confusing product stack.  

Those chips are then carved into various Max, Platinum, Gold, Silver, and Bronze sub-tiers, each denoting various levels of socket scalability, support for Optane persistent memory, RAS features, SGX enclave capacities, and the like.

The Sapphire Rapids chips also now come with a varying number of enabled accelerator devices onboard. For now, it's important to know that each chip can have a variable number of accelerator 'devices' enabled (listed in the spec sheet above —think of the number of 'devices' as akin to accelerator 'cores').

You can buy chips that are fully enabled with four devices for all accelerators, or you can opt for less expensive chip models with a lower number of enabled devices. If the chip isn't fully enabled, you can activate the accelerators later via a new pay-as-you-go mechanism called Intel on Demand. The '+' models have at least one accelerator of each type enabled by default. However, there are two classes of chips with two different allocations of accelerators. We'll dive into those details, and the different types of accelerators, below.

The new processors all support AVX-512, Deep Leaning Boost (DLBoost), and the new Advanced Matrix Extensions (AMX) instructions, with the latter delivering explosive performance uplift in AI workloads by using a new set of two-dimensional registers called tiles. Intel's AMX implementation will primarily be used to boost performance in AI training and inference operations.

As before, Intel's 4th-Gen Xeon Scalable platform supports 1-, 2-, 4-, and 8-socket configurations, whereas AMD's Genoa only scales to two sockets. AMD leads in PCIe connectivity options, with up to 128 PCIe 5.0 lanes on offer, while Sapphire Rapids peaks at 80 PCIe 5.0 lanes.

Sapphire Rapids also supports up to 1.5TB of DDR5-4800 memory spread across eight channels per socket, while AMD's Genoa supports up to 6TB of DDR5-4800 memory spread across 12 channels. Intel has spec'd its 2DPC (DIMMs per Channel) configuration at DDR5-4400, whereas AMD has not finished qualifying its 2DPC transfer rates (the company expects to release the 2DPC spec this quarter).

The Sapphire Rapids processors span from eight-core models to 60 cores, with pricing beginning at $415 and peaking at $17,000 for the flagship Xeon Scalable Platinum 8490H. The 8490H has 60 cores and 120 threads, with all four accelerator types fully enabled. The chip also has 112.5 MB of L3 cache and a 350W TDP rating.

The Sapphire Rapids TDP envelopes span from 120W to 350W. The 350W rating is significantly higher than the 280W peak with Intel's previous-gen Ice Lake Xeon series, but the inexorable push for more performance has the industry at large pushing to higher limits. For instance, AMD's Genoa tops out at a similar 360W TDP, albeit for a 96-core model, and can even be configured as high as 400W. 

The 8490H is the lone 60-core model, and it is only available with all acceleration engines enabled. Stepping back to the 56-core Platinum 8480+ will cost you $10,710, but that comes with only one of each type of acceleration device active. This processor has a 3.8 GHz boost clock, 350W TDP, and 105MB of L3 cache. 

Intel Xeon Sapphire Rapids Accelerators

Intel's new on-die accelerators are a key new component for its Sapphire Rapids processors. As mentioned above, you can either purchase chips with all of the accelerator options activated, or you can opt for less expensive models and purchase accelerator licenses as needed through the Intel On Demand service. Not all chips have the same accelerator options, which we'll cover below.

Intel hasn't provided a pricing guide for the accelerators yet, but the licenses will be provided through server OEMs and are activated via software and a licensing API. Instead of buying a full license outright, you can also opt for a pay-as-you-go feature with usage metering to measure how much of a service you use. This feature will likely be popular among CSPs.

The idea behind the Intel On Demand service is to allow customers to activate and pay for only the features they need, while also providing a future upgrade path that doesn't require buying new servers or processors. Instead, customers could opt to employ acceleration engines to boost performance. This also allows Intel and its partners to carve multiple types of SKUs from the same functional silicon, thus simplifying supply chains and reducing costs.

These functions represent Intel's continuation of its long history of bringing fixed-function accelerators onto the processor die. Still, the powerful units on Sapphire Rapids will require software support the extract the full performance capabilities. Intel is already working with several software providers to enable support in a broad range of applications, many of which you can see in the album above.

Intel has four types of accelerators available with Sapphire Rapids. The Data Streaming Accelerator (DSA) improves data movement by offloading the CPU of data-copy and data-transformation operations. The Dynamic Load Balancer (DLB) accelerator steps in to provide packet prioritization and dynamically balance network traffic across the CPU cores as the system load fluctuates.

Intel also has an In-Memory Analytics Accelerator (IAA) that accelerates analytics performance and offloads the CPU cores, thus improving database query throughput and other functions.

Intel has also brought its Quick Assist Technology (QAT) accelerators onboard the CPU. This function used to reside on the chipset. This hardware offload accelerator augments cryptography and compression/decompression performance. Intel has employed QAT accelerators for quite some time, so this technology already enjoys broad software support.

Unfortunately, the chips have varying acceleration capabilities — you can't buy four 'devices' on all models. The Sapphire Rapids processors are comprised of two types of designs (Die Chops), as listed in the SKU table. The XCC chips are comprised of four total die, and each die has one of each accelerator (IAA, QAT, DSA, DLB). That means you can activate a maximum of four accelerators of each type on these chips (for example, 4 IAA, 4 QAT, 4 DSA, 4 DLB).

In contrast, some of the chips use a single MCC die, so they only have one IAA and DSA accelerator and two each of the QAT and DLB accelerators (2 QAT, 2 DLB, 1 IAA, 1 DSA). 

Intel Max CPU Series and Ponte Vecchio Max GPU Series

Intel recently announced details about its forthcoming Xeon Max Series of CPUs and the Intel Data Center GPU Max Series (Ponte Vecchio). Today marks the formal launch.

Intel's HBM2e-equipped Max CPU models come to market with 32 to 56 cores and are based on the standard Sapphire Rapids design. These chips are the first x86 processors to employ HBM2e memory on-package, thus providing a larger 64GB pool of local memory for the processor. The HBM memory will help with memory-bound workloads that aren't as sensitive to core counts, so the Max models come with fewer cores than standard models. Target workloads include computational fluid dynamics, climate and weather forecasting, AI training and inference, big data analytics, in-memory databases, and storage applications.

The Max CPUs can operate in a multitude of various configurations, such as with the HBM memory used for all memory operations (HBM only - no DDR5 memory required), an HBM 'Flat Mode' that presents the HBM as a separate memory region (this requires extensive software support), or in an HBM 'Caching Mode' that uses the HBM2e as a DRAM-backed cache. The latter requires no code changes and will likely be the most frequently used mode of operation.

The Xeon Max CPUs will square off with AMD's EPYC Milan-X processors, which come with a 3D-stacked L3 cache called 3D V-Cache. The Milan-X models have up to 768MB of total L3 cache per chip that delivers an incredible amount of bandwidth, but it doesn't provide as much capacity as Intel's approach with HBM2e. Both approaches have their respective strengths and weaknesses, so we're eager to put the Xeon Max processors to the test.

Notably, Fujitsu's A64FX Arm processor uses a similar HBM technique. The HBM-equipped A64FX processors power the Fugaku supercomputer, which was the fastest in the world for several years (until the AMD-powered exascale-class Frontier took over last year). Fugaku still maintains the second spot on the Top500.

Intel also launched its Max GPU Series, previously code-named Ponte Vecchio. Intel had previously unveiled the three different GPU models, which come in both standard PCIe and OAM form factors. You can read more about the Max GPU Series here. 

Intel Optane Persistent Memory (PMem) 300

As part of its Sapphire Rapids launch, Intel quietly introduced the final series of Optane Persistent Memory DIMMs. The final generation, codenamed Crow's Pass but officially known as the Intel Optane Persistent Memory 300, will come in 128, 256, and 512 GB capacities and operate at DDR5-4400. That's a big improvement over the previous peak of DDR4-3200, but it also means that Sapphire Rapids systems will have to downclock the standard memory to DDR5-4400 from the supported DDR5-4800 if they plan on employing Optane.

Intel claims that the 300-series offers 56% more sequential bandwidth and 214% more bandwidth in random workloads, along with support for up to 4TB of Optane per socket, or 6TB total for a system. Just like the previous-gen Optane 200 series, the DIMMs operate at 15W. However, they now step up to a DDR-T2 interface and AES-XTS 256-bit encryption.

At its debut in 2015, Intel and partner Micron touted the underlying tech, 3D XPoint, as delivering 1000x the performance and endurance of NAND storage in tandem with 10x the density of DRAM, but the technology is now winding its way to an end. Intel has already stopped producing its Optane storage products for client PCs, which makes sense because it is selling its NAND business to SK Hynix.

However, Intel has retained its memory business for the data center, including its persistent memory DIMMs that can function as an adjunct to main memory — a capability only Intel offers. Those products will also not see any future generations after the 300-series modules.

Intel cites an industry shift to CXL-based architectures as a reason for winding down the Optane business, mirroring Intel's ex-partner Micron's sentiments when it exited the business last year. Sapphire Rapids supports both Optane DIMMs and the CXL interface, but this will be one of the last times the two are seen together — CXL will be the industry's preferred method of connecting exotic memories to chips in the future.

We are currently underway with our testing for our Sapphire Rapids review, so stay tuned for the full performance breakdown and architectural details in the coming days. 

Intel, late on Tuesday, officially confirmed that its Sapphire Rapids processors are coming to workstations in a teaser video published on Twitter. This is the first official confirmation that the company intends to address this market with its upcoming Sapphire Rapids CPUs. However, Intel is still tight-lipped about the specifications of these upcoming products.

"This thing is ridiculous, it used to take whole room full of computers to do what this is doing on its own, there go my 30-minute rendering coffee breaks," says a lady in the video. "I was on a meeting the other day and our final test runs are coming back really good, so I'd say pretty soon. You should probably take that coffee break while you still can.  

Based on the latest leaks, Intel will market its Sapphire Rapids-WS CPUs (at least, this is how they are called for now, unofficially) under the Xeon W 3400-series name. These processors will use Intel's all-new W790 platform and offer up to 56 cores, eight DDR5 memory channels, and 112 PCIe lanes. In addition, the CPUs are set to be based on high-performance Golden Cove-derived cores with AVX-512 and AMX instructions enabled.

Many Sapphire Rapids-WS SKUs — including the flagship Xeon W9-3495X — will come with an unlocked multiplier and thus be overclockable. The 56-core range-topping model will be the industry's first overclockable CPU for powerful workstations released in years. Intel and AMD essentially abandoned this market and focused on more traditional workstations that never get overclocked.

A rumor is that Intel plans to release its Sapphire Rapids-WS processors next Spring after it first ships Xeon Scalable 'Sapphire Rapids' products for servers. So far, Intel has not confirmed any timeframes for its workstation-grade Xeon W 3400-series products, but at least it clearly stated that Sapphire Rapids is coming to workstations.

Intel's plans for the workstation market with its Sapphire Rapids-WS are taking shape as a well-known hardware leaker published preliminary specifications for the new CPUs. Intel's lineup of next-generation Xeon products for workstations and high-end desktops will include overclockable CPUs with up to 56 cores, eight memory channels, and 112 PCIe lanes if the information revealed by reputable hardware leaker Enthusiastic Citizen (ECSM_Official) is correct.

The New Family of Workstation CPUs from Intel

Intel's family of next-generation Xeon W processors for W790-based workstations will reportedly consist of two families of products that will offer slightly different capabilities. The Xeon W 3400-series CPUs will be derived from a multi-chiplet Sapphire Rapids design and will feature up to 56 cores, eight DDR5 memory channels, and 112 PCIe lanes. In addition, CPU cores used by these processors will be Golden Cove-derived cores with AVX-512 and AMX instructions enabled. By contrast, the Xeon W-2400-series processors will use a single-die design with up to 24 cores, four DDR5 memory channels, and 64 PCIe lanes. 

Intel's Xeon W-2400 and W3400-series processors are expected to come in LGA4677 packaging and use W790-based workstation motherboards. One of the first W790 mainboards leaked last week, which suggests that some of Intel's partners are getting ready to ship these products sooner rather than later. Meanwhile, a rumor suggests that Intel only intends to roll out its W790 platform next April, so it is too early to ship appropriate motherboards. Then again, Intel has never officially confirmed the launch timeframe for its W790 platform and only confirmed that this one is designed for workstations.

Intel Xeon W-3400: Up to 56 Overclockable Cores

Intel's Xeon W-3400-series lineup will allegedly include nine models, four of which will be overclockable. Even the flagship Xeon W9-3495X is expected to come with an unlocked multiplier making for overclocking support. 

Linux boot logs unearthed earlier this year essentially confirm the existence of Intel's Xeon W-3400-series CPUs (which come with AVX-512 and AMX enabled). Still, they also mention the Xeon W9-3495 (non-X) CPU clocked at 1.80 GHz base, which Enthusiastic Citizen does not list. We have no idea whether Intel changed its plans concerning its Sapphire Rapids-WS lineup since July, but we are dealing with preliminary information, so some details may be inaccurate. 

Intel's Xeon W-3400-series relies on Sapphire Rapids silicon, which will offer AVX-512 support and AMX instructions for artificial intelligence and machine learning applications. Advanced Matrix Extensions is a tiled matrix multiplication accelerator, a grid of fused multiply-add units supporting BF16 and INT8 input types that can be programmed using only 12 instructions and perform up to 1024 TMUL BF16 or 2048 TMUL INT8 operations per cycle per core.

Currently, there are no workstation-grade CPUs featuring up to 56 cores and AVX-512 instructions, so this will be a tangible advantage of Intel's X-3400-series processors over existing AMD's Ryzen Threadripper Pro 5000WX-series products. Meanwhile, AMD is working on the next-generation Zen 4-based Ryzen Threadripper Pro 7000WX family codenamed Storm Peak that will also support AVX-512, but we have no idea when these CPUs are set to become available. 

What will truly set Intel's Xeon W-3400 apart from the competition is AMX support since these instructions will make it extremely competitive in AI/ML and other types of matrix multiplication workloads. One may argue that workstations hardly run such applications, but someone has to develop and try AMX-supporting applications on something, and they will likely opt for a Xeon W-3400-based machine. Furthermore, once workstation programs learn how to use AMX properly, Sapphire Rapids-WS-based devices will have an advantage over the competition for some time.

Intel Xeon W-2400: Up to 24 Overclockable Cores

Intel's Xeon W-2400-series family will reportedly comprise eight SKUs, four of which will be overclockable. While we cannot state this for sure, but we suspect that the W2000-series will use Intel's 34-core Raptor Lake-S silicon. Meanwhile, even the range-topping Xeon W7-2495X features 24 cores with Hyper-Threading. Perhaps, Intel yet has to disclose additional Xeon W-2400-series models with higher core counts to partners. Or maybe the company decided to sacrifice eight cores for additional yields, clocks, or perhaps decent clocks for AVX-512. 

Three interesting SKUs that Intel allegedly has in the Xeon W-2400 lineup are the eight-core W5-2435 as well as the six-core W3-2425 and W3-2423 processors. Assuming that these chips can be overclocked by raising BCLK frequency (we are not sure this will work on a workstation-grade W790 platform, but who knows?), they will certainly attract the attention of the world's top professional overclockers. Meanwhile, we can only wonder whether these parts will be successful commercially or share the fate of Intel's Kaby Lake-X processors for the X299 platform aimed at enthusiasts. 

It is noteworthy that the Xeon W4-2423 processor does not support Hyper-Threading, according to Enthusiastic Citizen, which is a big surprise for a 2023 workstation processor. Interestingly, there are also Xeon W5-3433 and W5-3423 processors in the 3400-series family, but there is no word whether they support simultaneous multithreading.

Worthy Contender? Maybe

Intel's Xeon W-3000 and W-2000-series processors could give AMD's Ryzen Threadripper Pro a good run for their money if the information is correct. Yet, while Enthusiastic Citizen tends to be very accurate and therefore has a good reputation, we are still dealing with unofficial information, so take it with a grain of salt. 

Sapphire teased new Radeon RX 7900 XT and XTX aftermarket graphics cards on its Twitter page just a few hours ago. But these aren't your ordinary GPUs; these GPUs will be carrying the Vapor-X sub-brand, which we haven't seen from Sapphire in eight years.

Vapor was one of Sapphire's higher-up GPU lineups in 2014 and earlier, being the runner-up to its Toxic sub-brand of flagship cards. As the name implies, the Sapphires Vapor and Vapor-X Radeon GPUs featured a vapor chamber cooler design, which at the time was very rare and could be considered an exotic trait among GPU coolers.

Models like the Vapor-X R9 290X packed huge triple fan cooler designs beyond two slots in thickness, which was massive for GPU coolers at the time. It also packed massive power delivery systems and power targets that could be adjusted by a whopping 50% beyond the card's default power limit (which was probably already boosted beyond the reference design, to begin with).

It'll be exciting to see how Sapphire treats the Vapor-X brand now that it's been taken out of retirement for 2022 and beyond. But, of course, a lot has changed over the past 8 to 10 years, including that many cards now use vapor chamber coolers - including some of Nvidia's own Founder Edition cooler designs.

If Sapphire wants Vapor-X to stand out, it'll have to do much more than just feature a vapor chamber cooler. We suspect these cards will be massive, some of Sapphire's biggest cooler designs to date, and will probably have an enormous power delivery system with dual BIOS chips to match.

But this is just speculation on our end; all we know is that the RX 7900 XTX and RX 7900 XT will be getting Vapor-X models, and we don't know if the non-X versions are returning. One thing that does stand out in the teaser video from Sapphire is that the memory and the GPU core are covered in a pure copper baseplate that appears to be a completely separate piece from the graphics card's heatsink. This suggests the card will feature excellent cooling characteristics, mainly because the baseplate is made out of copper - which is more thermally conductive than other metals.

Another interesting note is that the card also features triple 8-pin power connectors, one step up from the dual 8-pin power connectors we saw on the reference 7900 XTX. Unfortunately, this might confirm that AMD's Radeon 7000 series GPUs are incompatible with the 16-pin power connector since triple 8-pins are far bulkier than a single 16-pin.

Pricing, availability, and specifications are unknown, but we should expect more details to arrive soon, with the RX 7900 XTX and XT launch date fast approaching.

A Canadian retailer has listed Supermicro's yet-to-be-announced X13SWA-TF motherboard based on the Intel W790 chipset. It is designed for workstations featuring Intel's Xeon processors in LGA4677 packaging, which points to Sapphire Rapids CPUs.

The Supermicro X13SWA-TF motherboard (listed at Atic.ca and spotted by @momomo_us) comes in an E-ATX form factor, which is used for workstations, desktops, and tower servers. The motherboard is priced at CAD$1290  ($965) with a discount if you pay cash, but there is no mention of its availability timeframe.

Earlier this year, Intel confirmed that its W790 is a workstation-grade chipset (not an entry-level workstation-grade chipset) but never disclosed which processors it supports. But since the motherboard has an LGA4677 socket, it should support Intel's Xeon processors codenamed Sapphire Rapids. 

We can speculate that we might be talking about a special version of Sapphire Rapids CPUs tailored for workstations (i.e., featuring fewer than 60 cores but with higher clocks) or Intel's 34-core Raptor Lake-S processors spotted by our own Paul Alcorn earlier this year. Meanwhile, both Sapphire Rapids and Raptor Lake-S feature high-performance cores based on the Golden Cove microarchitecture and have several things in common. Therefore, it is plausible to assume that Intel's 34-core Raptor Lake processors (also known as RPLS-34C) will be marketed under the Xeon brand, though Intel has never discussed it publicly. 

The Supermicro X13SWA-TF platform has six PCIe 4.0 x16 slots, two 10GbE ports controlled by Intel's X550 chip, and Intelligent Platform Management Interface (IPMI) for remote management, which is handy both for workstations and servers. The basic description does not mention PCIe 5.0 support or M.2 slots (indispensable for desktop PCs), but since this is an early representation of the X13SWA-TF's capabilities, this is not surprising. 

Rumors about a Sapphire Rapids-based platform for high-end desktops and workstations have been floating around for years. However, Intel has never publicly confirmed plans to address the client PC market with its SPR processors as it is focused on launching Sapphire Rapids for servers first. The listing is essentially the first confirmation that Intel's W790 platform supports the company's processors in LGA4677 packaging, which points to Intel Xeon 'Sapphire Rapids' CPUs. Meanwhile, Intel's 34-core Raptor Lake-S processors are perhaps better candidates for mainstream workstations and high-end desktops.

AMD and its partners plan to begin sales of the company's flagship Radeon RX 7900 XT and Radeon RX 7900 XTX graphics cards on December 13. Based on a media report, most RDNA 3 products will be reference designs on that date. Still, some custom-built models will also make it to the market.

Earlier this week, a post over Board Channel forums suggested that only reference design Radeon RX 7900-series graphics cards will be available at the launch date on December 13. Custom designs from add-in-board (AIB) makers will only hit the market one or two weeks later, according to the post, which means that such cards might miss the lucrative holiday season in the U.S. and Europe. But this is not entirely correct.

As it turns out, custom-designed Radeon RX 7900 XT and Radeon RX 7900 XTX graphics cards from at least two manufacturers will be available starting December 13, according to information obtained by ComputerBase.de. Meanwhile, the website stresses that some other manufacturers admitted that they will only offer reference design models on December 13 and will follow up with custom boards later.

Enough Motivation, Not Enough Time

The Radeon RX 7900-series launch schedule was tight for makers of graphics cards, based on information from ComputerBase, so they did not have enough time to develop custom versions of the Radeon RX 7900 XT as well as the Radeon RX 7900 XTX.

It should be noted that all the information we have about the status of custom-designed Radeon RX 7900-series graphics cards comes from third parties, so we cannot draw an objective conclusion here. But we can speculate for a bit.

Makers of graphics cards develop custom versions of their products to differentiate from their rivals in terms of performance and features and sell their parts at a higher-than-MSRP price to maximize profitability. Some designs are created to lower costs, but this is hardly the case with the first wave of AMD's Radeon RX 7900-series products that will retail for $899 - $999 or higher.

To build performance-enhanced high-end graphics boards, companies need quite some time to experiment with GPUs and memory in their possession, their printed circuit board (PCB) configuration, and cooling system performance to ensure that the final products can operate at desired clocks stably for an extended amount of time without failures.

Since AMD's RDNA 3-based Navi 31 graphics processor features two separate clock domains (for stream processors and everything else), there are many things for graphics card makers to experiment with, which takes time. In addition, since there are many moving parts to play with, AIB makers may opt to expand their high-end lineups with more SKUs to have enough time to test hardware in different configurations but, at the same time, roll out some SKUs ahead of the others. This is perhaps why Asus might be prepping as many as eight ROG Strix Radeon RX 7900-series (XT and XTX models) graphics boards.

Three AIB Makers

Meanwhile, it looks like initially, Asus will release TUF Gaming-branded Radeon RX 7900 XT and Radeon RX 7900 XTX boards. Those cards will feature an enhanced-PCB design with three eight-pin auxiliary PCIe power connectors to provide them with some additional overclocking potential and therefore enable higher performance for those who do not want to settle with reference designs. Yet, Asus positions its TUF Gaming-badged products below its ROG Strix boards, so the latter is poised to offer higher performance.

In addition to Asus, only two AIB manufacturers — PowerColor and Sapphire — have teased their custom Radeon RX 7900-series graphics cards. Meanwhile, the board makers committed to releasing AMD's Radeon RX 7900 products, including ASRock, Biostar, Gigabyte, MSI, Vastarmor, XFX, and Yeston.

PowerColor demonstrated elements of its top-of-the-range Radeon RX 7900-series Red Devil offering and Radeon RX 7900-series Hellhound product that will sit below the mighty range-topping board. Given that the Hellhound-branded card has only two eight-pin auxiliary PCIe power connectors, we would suggest that this device is not aimed at overclockers who tend to squeeze every last bit of performance from their graphics cards.

By contrast, Sapphire showed design elements of its Nitro-series Radeon RX 7900 graphics board. The AIB features a modern and cleanly designed cooling system. Still, one has to remember that Sapphire's Nitro family usually comprises moderately factory-overclocked products that are better than reference designs, but this is about it. Sapphire's range-topping offerings belong to the company's Toxic lineup. They are equipped with exotic cooling systems, come with industry-leading clocks, and offer tangibly higher performance than reference designs. Yet, Sapphire has not even teased a Toxic Radeon RX 7900-series product.

Some Thoughts

So far, we have seen only one announcement of custom Radeon RX 7900-series graphics cards: the TUF Gaming Radeon RX 7900 XT and XTX from Asus. Considering that the company was the first to reveal its custom RDNA 3 boards, we expect it to start selling them sooner rather than later. We do not know whether Asus intends to make them available on December 13.

PowerColor and Sapphire have only teased their upcoming Radeon RX 7900-series products so far, but neither company has made any announcements. We are not going to speculate when we could expect PowerColor's Red Devil/Hellhound Radeon RX 7900-series or Sapphire's Nitro Radeon RX 7900-series to arrive. Still, we can presume that the companies are optimistic about their launch date and believe it is time for a tease.

As for other makers, they still have 19 days to show their custom designs and if they are ready to ship those boards, ship them by air to markets they want to address. Meanwhile, since AIB makers reportedly complain about the tight schedule for custom Radeon RX 7900-series boards, we would instead assume that it will take a little longer for custom Radeon RX 7900-series cards to hit the market.

Sapphire has shared what looks like its first Radeon RX 7900 XTX / 7900 XT teaser on social media. The AMD-exclusive partner published a brief video clip on Twitter featuring an alluring pink LED-lit silver and black graphics card. In related news, discussions on China’s Bobantang forums indicate that only AMD reference design cards will be available at launch (December 13). Gamers, PC DIYers and enthusiasts must wait another one or two weeks for custom models, like the teased Sapphire cards, to materialize.

You can see the full Sapphire video teaser embedded above; however, not a lot is revealed. All we see is a triple fan graphics card running with pink LED edging. Like most triple fan designs, this one uses an alternative spinning layout, with only the center fan rotating clockwise. The design looks very premium from what we can see, with this impression making the upcoming Sapphire graphics card look like an MSI Suprim model.

AMD Reference Models First?

On the Chinese tech forum Bobantang, there are some interesting rumors regarding the launch schedule for the upcoming AMD RDNA 3 flagship pair. AMD has already officially established that the Radeon RX 7900 XTX and 7900 XT will become available on December 13. However, the China tech forum rumors suggest that customers will have their choice limited to the AMD reference designs. According to the tech forum rumor, this restriction will lift one or two weeks after the initial December 13 launch.

Another detail from the Bobantang post is probably more worrying. It suggests that AMD partner custom designs will be more expensive than the reference models. From the leaks we have seen, AMD reference designs look like some of the more modest upcoming RDNA 3 graphics cards. On the other hand, partner designs look almost as bulky as the latest GeForce RTX 4090 and RTX 4080 graphics cards in some instances.

There are plenty of Black Friday PC hardware deals going on right now, and PC gamers will probably be most interested in the Black Friday graphics card deals. This one might fly a bit under the radar, but we've received the Sapphire RX 6700 10GB Pulse graphics card for review and have finished most of our testing.

It's an interesting card in its own right, taking the Navi 22 GPU used in the RX 6750 XT and RX 6700 XT but trimming down a few of the specs. It's also on sale right now and that may not last, so we thought we'd give you a quick performance preview.

Overall, the RX 6700 10GB does quite nicely at 1080p ultra, coming in 12% behind the RX 6700 XT while shaving $50 off the price. It's also 25% faster than the Nvidia RTX 3060 and currently costs 11% less, and it's just a few percent slower than the RTX 3060 Ti at a price that's 21% lower. In terms of overall value, about the only GPUs that currently rate higher than the 6700 10GB are the RX 6650 XT and the RX 6600.

Sapphire's card isn't substantially faster, but the extra VRAM, bandwidth, and compute does give it a 9% lead on the 6650 XT and a 30% advantage over the vanilla 6600. It costs about 20% more than the cheapest 6650 XT and 58% more than the 6600. It's also 11% faster than the Arc A770 16GB and 24% faster than the Arc A750, at least in our standard (non-ray tracing) test suite.

Ray tracing doesn't do AMD's RDNA 2 architecture any favors, as usual. The gap relative to other AMD cards remains similar, though it does extend its lead a bit over the Navi 23-based RX 66xx-series parts. It's also only 10% slower than the RTX 3060, though it still beats the RTX 3050 by a healthy 28% margin. Intel's Arc A770 meanwhile delivered 34% higher performance and the A750 also beat the 6700 10GB by 19% — all of these DXR results are at 1080p medium, which is about as high as you'll want to go with ray tracing on AMD's midrange parts.

Overall, there's plenty to like with AMD's RX 6700 10GB card. We can't help but wonder what took so long, but so many things were wonky in the world of GPUs last year that we'll likely never know for sure. If you're looking for a good card that can mostly handle 1440p in standard games and cruises along at more than 60 fps with maxed out settings at 1080p, give the RX 6700 non-XT some thought.

Just days before Supercomputing 22 kicks off, Intel introduced its next-generation Xeon Max CPU, previously codenamed Sapphire Rapids HBM, and Data Center GPU Max Series compute GPUs, known as Ponte Vecchio. The new products cater to different types of high-performance computing workloads or work together to solve the most complex supercomputing tasks.

The Xeon Max CPU: Sapphire Rapids Gets 64GB of HBM2E 

General-purpose x86 processors have been used for virtually all kinds of technical computing for decades and therefore support many applications. However, while the performance of general-purpose CPU cores has scaled rather rapidly for years, today's processors have two significant limitations regarding performance in artificial intelligence and HPC workloads: parallelization and memory bandwidth. Intel's Xeon Max 'Sapphire Rapids HBM' processors promise to remove both boundaries.

Intel's Xeon Max processor features up to 56 high-performance Golden Cove cores (spread over four chiplets interconnected using Intel's EMIB technology) further enhanced with multiple accelerator engines for AI and HPC workloads and 64GB of on-package HBM2E memory. Like other Sapphire Rapids CPUs, the Xeon Max will still support eight channels of DDR5 memory and PCIe Gen 5 interface with the CXL 1.1 protocol on top, so it will be able to all those CXL-enabled accelerators when it makes sense.

In addition to vector AVX-512 and Deep Learning Boost (AVX512_VNNI and AVX512_BF16) accelerators support, the new cores also bring Advanced Matrix Extensions (AMX) tiled matrix multiplication accelerator, which is essentially a grid of fused multiply-add units supporting BF16 and INT8 input types that can be programmed using only 12 instructions and perform up to 1024 TMUL BF16 or 2048 TMUL INT8 operations per cycle per core. Also, the new CPU supports Data Streaming Accelerator (DSA), which offloads data copy and transformation workloads from the CPU.

64GB of on-package HBM2E memory (four stacks of 16GB) provides a peak bandwidth of around 1TB/s, which translates to ~1.14GB of HBM2E per core at 18.28 GB/s per core. To put the numbers into context, a 56-core Sapphire Rapids processor equipped with eight DDR5-4800 modules gets up to 307.2 GB/s of bandwidth, which means 5.485 GB/s per core. Meanwhile, Xeon Max can use its HBM2E memory in different ways: use it as system memory, which requires no code change; use it as a high-performance cache for DDR5 memory subsystem, which does not require change code; use it as a part of a unified memory pool (HBM flat mode), which involves software optimizations.

Depending on the workload, Intel's AMX-enabled Xeon Max processor can provide a 3X – 5.3X performance improvement over the currently available Xeon Scalable 8380 processor that uses conventional FP32 processing for the same workloads. Meanwhile, in applications like model development for molecular dynamics, the new HBM2E-equipped CPUs are up to 2.8X times faster than AMD's EPYC 7773X, which features 3D V-Cache.

But HBM2E has another important implication for Intel as it somewhat reduces data movement overhead between CPU and GPU, which is essential for various HPC workloads. It brings us to the second of today's announcements: the Data Center GPU Max Series compute GPUs.

The Data Center GPU Max: The Pinnacle of Intel's Datacenter Innovations

Intel's Data Center GPU Max compute GPU series will employ the company's codenamed Ponte Vecchio architecture, first introduced in 2019 and then detailed in 2020 ~ 2021. Intel's Ponte Vecchio is the most complex processor ever created, as it packs over 100 billion transistors (not including memory) over 47 tiles (including 8 HBM2E tiles). In addition, the product extensively uses Intel's advanced packaging technologies (e.g., EMIB) as different tiles are made by other manufacturers using different process technologies.

Intel's Data Center GPU Max compute GPUs will rely on the company's Xe-HPC architecture tailored explicitly for AI and HPC workloads and therefore support appropriate data formats and instructions as well as 512-bit vector and 4096-bit matrix (tensor) engines.

Compared to Xe-HPG, Xe-HPC has considerably more sophisticated memory and caching subsystems, differently configured Xe cores (each Xe-HPG core features 16 256-bit vector and 16 1024-bit matrix engines, whereas each Xe-HPC core sports eight 512-bit vector and eight 4096-bit vector engines). Furthermore, Xe-HPC GPUs do not feature texturing units or render back ends, so they cannot render graphics using traditional methods. Meanwhile, Xe-HPG surprisingly supports ray tracing for supercomputer visualization.

One of the most important ingredients of Xe-HPC is Intel's Xe Matrix Extensions (XMX) that enable rather formidable tensor/matrix performance of Intel's Data Center GPU Max 1550 (see the table below) — up to 419 TF32 TFLOPS and up to 1678 INT8 TOPS, according to Intel. Of course, peak performance numbers provided by compute GPU developers are important but may not reflect performance achievable on real-world supercomputers in real-world applications. Still, we cannot help but notice that Intel's range-topping Ponte Vecchio is significantly behind Nvidia's H100 in most cases and fails to deliver tangible advantages over AMD's Instinct MI250X across all cases except FP32 Tensor (TF32).

Meanwhile, Intel says that its Data Center GPU Max 1550 is 2.4x faster than Nvidia's A100 on Riskfuel credit option pricing and offers a 1.5x performance improvement over A100 for NekRS virtual reactor simulations.

Intel plans to offer three Ponte Vecchio products: the top-of-the-range Data Center GPU Max 1550 in OAM form-factor featuring 128 Xe-HPC cores, 128GB of HBM2E memory, and rated for up to 600W thermal design power; the cut-down Data Center GPU Max 1350 in OAM form-factor with 112 Xe-HPC cores, 96GB of memory, and a 450W TDP; and the entry-level Data Center GPU Max 1100 that comes in a dual-wide FLFH form-factor and carries a processor with 56 Xe-HPC cores, has 56GB of HBM2E memory and rated for a 300W TDP.

Meanwhile, to its supercomputer clients, Intel will offer Max Series Subsystems with four OAM modules on a carrier board rated for a 1,800W and 2,400W TDP.

Intel's Rialto Bridge: Enhancing the Max

In addition to formally unveiling its Data Center GPU Max compute GPUs, Intel today also gave a sneak peek at its next-generation Data Center GPU, codenamed Rialto Bridge which arrives in 2024. This AI and HPC compute GPU will be based on enhanced Xe-HPC cores, presumably with a slightly different architecture, but will maintain compatibility with Ponte Vecchi-based applications. Unfortunately, that additional complexity will increase the TDP of the next-generation flagship compute GPU to 800W, though there will be simpler and less power-hungry versions.

Availability

One of the first customers to get both Intel Xeon Max and Intel Data Center GPU Max products will be Argonne National Laboratory, which is building its >2 ExaFLOPS supercomputers based on over 10,000 blades using Xeon Max CPUs and Data Center GPU Max devices (two CPUs and six GPUs per blade). In addition, Intel and Argonne are finishing building Sunspot, Aurora's test development system consisting of 128 production blades that will be available to interested parties in late 2022. The Aurora supercomputer should come online in 2023.

Intel's partners, among server makers, will launch machines based on Xeon Max CPUs and Data Center GPU Max devices in January 2023.

This week, Intel revealed additional details about its Software Defined Silicon (SDSi) capability of its next-generation Xeon Scalable processors and the official brand name of this technology. The tech will unassumingly sport the 'Intel On Demand' moniker and allow system administrators to pay extra to enable special-purpose accelerators integrated into Intel's 4th Generation Xeon Scalable 'Sapphire Rapids' processors.

Intel this week released updates to SDSi patches merged in Linux 5.18 that reveal more details about this feature than the patches themselves, reports Phoronix. The software that will enable specific built-in acceleration capabilities of Intel's 4th Generation Xeon Scalable Sapphire Rapids CPUs (and probably their successors) will be called Intel On Demand, and it will do the following:

• Discover which features are physically present on a particular CPU.
• Offer administrators to activate them.
• Enable administrators to assess how often the feature is used.

The main intrigue about the Intel On Demand mechanism remains a mystery — we do not know which features Intel will allow for turning on post-purchase. However, we understand that Intel's Sapphire Rapids has several specific acceleration technologies. The list includes Advanced Matrix Extensions (AMX), Dynamic Load Balancer (DLB), Intel Data Streaming Accelerator (DSA), Intel In-Memory Analytics Accelerator (IAA), and Intel QuickAssist Technology (QAT) to accelerate specific workloads.

From previously uncovered beans about Intel's SDSi, or Intel On Demand if you prefer, we already know that the program enables access to the interface in the CPU to allow silicon features with an Authentication Key Certificate (AKC) and Capability Activation Payload (CAP) license. We also know that the program allows us to enable the specific feature on a particular CPU socket, not across all processors in the system or the data center itself.

Meanwhile, the fact that the software will need to discover which capabilities are physically supported by a processor and hide those not supported means that not all Intel Xeon Scalable 'Sapphire Rapids' processors will be created equal. Some CPU models may not gain support for certain features even by using Intel On Demand software.

Not all users will require AMX, DLB, DSA, IAA, and QAT at once. But which will be enabled by default on all SKUs and which will have to be activated using the IOD software is something that Intel will probably reveal on January 10, when it launches its next-generation Xeon Scalable CPUs officially.

Intel plans to formally introduce its 4th Generation Xeon Scalable processors codenamed Sapphire Rapids on January 10, 2023, the company revealed via Twitter this week. Some of Intel's next generation datacenter CPUs are already shipping to select customers, but in January actual machines based on these processors will be available for purchase. 

Intel's 4th Generation Xeon Scalable represents a major evolution of the company's datacenter platforms. Sapphire Rapids introduces a brand-new microarchitecture to the datacenter and increases a core count of up to 60 cores. It also brings in support for Advanced Matrix Extensions (AMX), Dynamic Load Balancer (DLB), Intel Data Streaming Accelerator (DSA), Intel In-Memory Analytics Accelerator (IAA), and Intel QuickAssist Technology (QAT) to accelerate specific workloads. On the Eagle Stream platform level, Sapphire Rapids brings PCIe Gen 5 support with the CXL 1.1 protocol on top, DDR5 and HBM2E memory support to an x86 CPU. 

To a large degree, Intel's next generation Xeon Scalable is a combination of high core count and special-purpose accelerators, something the datacenter world has not seen before at this scale, as traditionally servers use general-purpose CPUs and special-purpose accelerators. 

Adding so many features increased the number of hardware bugs and Intel had to fix around 500 of them, according to media reports. Since Intel had to do 12 respins of its Sapphire Rapids CPUs, it had to delay its CPU launch from 2021 to 2023. 

Apparently, Intel's Sapphire Rapids is now on its home straight and Intel can release these processors in high volumes and for all kinds of its clients. It remains to be seen how quickly Intel and its partners can ramp up volume production of servers based on the 4th Generation Xeon Scalable processors, but keeping in mind that server makers have been playing with the Eagle Stream platform for a while, the ramp could be accelerated

Another interesting thing to note is that Intel's Eagle Stream will also support Intel's 5th Generation Xeon Scalable codenamed Emerald Rapids processors that are also due in 2023, according to the chip giant. But will they arrive on time?

UPDATE 9/15 07:30 ET: Due to a misunderstanding of the context, we incorrectly said that the 52-core Xeon Platinum 8472C featured 64GB of on-package HBM2E memory. In a Twitter post @Yuuki_AnS said that none of the Sapphire Rapids processors in his leaked test results carried HBM2e memory. We have removed mentions of the 64GB of on-package HBM2E DRAM and corrected our comments.

Corrected Story

A well-known blogger has shared perhaps the first performance numbers of Intel’s upcoming Xeon Platinum Sapphire Rapids processors with 52 and 60 cores in the V-Ray benchmark. Apparently, neither a 60-core Intel CPU nor the 52-core Xeon CPU working at high frequencies can beat AMD’s high-end EPYCs available today.

Some of Intel's customers already get PRQ versions of select 4th Generation Xeon Scalable Sapphire Rapids processors. Still, the launch of high-volume models has suffered a delay to late 2022, with a full-scale ramp now scheduled for 2023 as the company had to correct certain bugs in silicon and then re-qualify those CPUs with partners. Intel keeps specifications of its upcoming products under wraps. Since there are many Sapphire Rapids processors in the wild already, not only do we know some of their alleged specifications, but their performance numbers have surfaced ahead of launch.

YuuKi_AnS, a prominent hardware leaker with a good track record and access to unreleased hardware, has laid his hands on Intel's 60-core Xeon Platinum 8490H CPU and a 52-core Xeon Platinum 8472C processor. The hardware blogger not only disclosed the specifications of the said products but also published their benchmark result. All the information is proven by screenshots (see the gallery below), so everything looks somewhat credible, while there may be some bumps. Meanwhile, since we are dealing with pre-production hardware, take the results with a grain of salt.

*Specifications are unconfirmed.

Last spring, Intel only planned to release Sapphire Rapids CPUs with up to 56 cores, but now the company is testing 60-core Xeon Platinum 8490H processors running at 1.90 GHz – 3.50 GHz. Furthermore, the company looks to be testing 52-core Xeon Platinum 8472C CPUs running at considerably higher 2.50 GHz – 3.80 GHz.

Due to higher frequency, Intel's Xeon Platinum 8472C offers slightly higher single-thread performance than the company's Xeon Platinum 8490H CPU in the CPU-Z 2.02.0 benchmark. Yet the 60-core model scores marginally higher in multi-threaded workloads.

Rendering workloads of Chaos's V-Ray 4.10.03 benchmark can take advantage of both many cores and higher per-core performance (i.e., higher clocks). Apparently, clocks play quite a role here as we can notice based on results of AMD and Intel CPUs. Intel's 52-core Xeon Platinum 8472C is about 32% faster when compared to its 60-core Xeon Platinum 8490H sibling, possibly because the pre-production 60-core CPU cannot hit high clocks for prolonged periods. There could be some issues with BIOS/microcode too as we are not dealing with production hardware.

Interestingly, AMD's EPYC 7773X (Milan-X with 3D V-Cache) does not have any edge over the EPYC 7763 processor, which runs only a tad faster. Nonetheless, both already available CPUs are currently untouchable by Intel's offerings that have yet to come to market.

One thing that YuuKi_AnS pointed out is that new BIOS versions tend to add performance to Intel's Sapphire Rapids CPUs. So, assuming that there will be further microcode optimizations of these processors, production servers and workstations based on these CPUs will offer higher performance than current samples. But until those machines arrive, all we have left is to analyze results obtained on pre-production CPUs.

MLCommons, an industry group specializing in artificial intelligence performance evaluation and machine learning hardware, has added results of the latest artificial AI and ML accelerators to its database and essentially published the first performance numbers for Nvidia's H100 and Biren's BR104 compute GPUs obtained via an industry-standard set of tests. The results were compared against those obtained on Intel's Sapphire Rapids, Qualcomm's AI 100, and Sapeon's X220. 

MLCommons' MLPerf is a set of training and inference benchmarks that are recognized by tens of companies that back the organizations and submit test results of their hardware to the MLPerf database. The MLPerf Inference version 2.1 set of benchmarks includes datacenter and edge usage scenarios as well as such workloads as image classification (ResNet 50 v1.5), natural language processor (BERT Large), speech recognition (RNN-T), medical imaging (3D U-Net), object detection (RetinaNet), and recommendation (DLRM). 

The machines participating in these tests are evaluated in two modes: in server mode queries arrive in bursts, whereas in offline mode all the data is fed at once, so evidently in offline mode they perform better. Also, vendors can submit results obtained in two conditions: in closed category everyone has to run mathematically equivalent neural networks, whereas in open category they can modify them in a bid to optimize them for their hardware, reports IEEE Spectrum. 

Results obtained in MLPerf not only describe pure performance of accelerators (e.g., one H100, one A100, one Biren BR104, etc.), but also their scalability, and performance-per-watt to draw a more detailed picture. All the results can be viewed in the database, but Nvidia compiled results of per-accelerator performance based on submissions from itself and third parties. 

Nvidia's competitors have not submitted all of their results just yet, so the graph published by Nvidia is missing some results. Yet we can still make some quite interesting findings in the table released by Nvidia (yet still keep in mind that Nvidia is an interested party here, so everything should be taken with a grain of salt). 

Since Nvidia's H100 is the most complex and the most advanced AI/ML accelerator backed by very sophisticated software optimized for Nvidia's CUDA architecture, it is not particularly surprising that it is the fastest compute GPU today that is up to 4.5X faster than Nvidia's A100. 

Yet Biren Technology's BR104, which offers around half of performance set to be offered by the flagship BR100, shows quite some promise in image classification (ResNet-50) and natural language processing (BERT-Large) workloads. In fact, if the BR100 is twice as fast as the BR104, it will offer higher performance than Nvidia's H100 in image classification workloads, at least as far as per-accelerator performance is concerned. 

Sapeon's X220-Enterprise as well as Qualcomm's Cloud AI 100 cannot even touch Nvidia's A100 that was launched around two years ago. Intel's 4th Generation Xeon Scalable 'Sapphire Rapids' processor can run AI/ML workloads, though it does not look like the code has been optimized enough for this CPU, which is why its results are rather low. 

Nvidia fully expects its H100 to offer even higher performance in AI/ML workloads over time and widen its gap with A100 as engineers learn how to take advantage of the new architecture.

What remains to be seen is how significantly will compute accelerators like Biren's BR100/BR104, Sapeon's X220-Enterprise as well as Qualcomm's Cloud AI 100 improve their performance over time. Furthermore, the real competitor for Nvidia's H100 will be Intel's codenamed Ponte Vecchio compute GPU that is positioned both for supercomputing and AI/ML applications. Also, it will be interesting to see results of AMD's Instinct MI250 — which is arguably optimized primarily for supercomputers — in MLPerf, and whose omission from the current tests is somewhat odd. Yet at least for now, Nvidia holds the AI/ML performance crown.

A pair of prototype Intel Xeon Platinum 8468 Sapphire Rapids 48 core server processors were recently tested in Geekbench 5, as shared by @BenchLeaks on Twitter. They featured some impressive performance figures, matching AMD's best of the best twin EPYC 64-core configuration in the same benchmark - and these chips aren't even the flagship 56-core variants.

The Sapphire Rapids system shared was equipped with twin Xeon Platinum 8468 CPUs featuring a total of 96 cores and 192 threads. The two chips combined outputted a single-threaded score of 1,257 points, and a mlti-threaded score of 74,586 points.

One of the highest performing dual AMD EPYC processor configurations we could find on Geekbench 5 for comparison features twin AMD EPYC 7763 processors with a combined core count of 128 cores and 256 threads. The two chips hit 1249 points in the single-threaded test and 75,539 in the multi-threaded test.

This results in a difference of 1% in both the single and multi-core for both platforms, essentially tieing them with equal performance. But, if we look at the individual performance of each core during the multi-threaded workload, Intel is 23% faster per core compared to AMD.

This is important to note because single-threaded benchmarks only tell you the maximum performance of an individual core when all the other cores are idle, allowing that single core to use excess power and thermal reserves. As a result, individual core performance under load shows us what per core performance looks like under maximum load.

But it's worth noting that Geekbench 5 is not a particularly useful benchmark for real-world performance results. So take this data with a grain of salt. On the flip side, Geekbench can at least give us a general idea of Sapphire Rapids' performance.

A Quick Recap on Sapphire Rapids and Its Delays

Sapphire Rapids is Intel's upcoming server architecture, designed to replace Intel's current 14nm Cascade Lake Xeon lineup. As a result, Sapphire Rapids will feature some serious upgrades, including Intel's cutting-edge Golden Cove cores running on the Intel 7 (10nm Enhanced SuperFin) node - just like Intel's current Alder Lake parts.

Like Alder Lake, PCIe 5.0 and DDR5 will be supported (with some parts sporting HBM compatibility). However, unlike Alder Lake, it will not support efficiency cores. From what we know, Sapphire Rapids will sport a maximum of 56 cores. But thanks to multi-socketed motherboards, core counts can go up to 112 or more. Sapphire Rapids will also sport additional features like AMX, and Intel DSA, which will be exclusive to its server architecture.

Unfortunately, Sapphire Rapids has allegedly turned into a bug-infested nightmare for Intel, with multiple delays setting back the platform's release window. Originally, Sapphire Rapids was intended to face off against AMD's Zen 3 EPYC Milan processors (like the ones we showed off in this article), but thanks to a whopping 500 bugs found in the microarchitecture, the launch window has moved to early 2023.

This means Sapphire Rapids will need to contend with current Zen 3 parts and AMD's future Zen 4 EPYC processors codenamed Genoa in 2023. Thankfully, it appears Intel's future server architecture can at least hang with AMD's Zen 3 competitors and outperform it by a modest amount. But there's no telling how it will perform against AMD's future Zen 4 architecture, which is already releasing in the form of Ryzen 7000.

Twitter’s YuuKi_AnS has been lucky enough to be able to test the latest and greatest workstation and server processors from AMD and Intel. The tech enthusiast and digital creator has put both an AMD EPYC Genoa two-processor (2P) system and an Intel Xeon Platinum 8480+ Sapphire Rapids 2P system through the Cinebench R23 rendering benchmark. The result is an obvious win for the AMD Genoa Engineering Sample. However, before going on, please take the results with a pinch of salt, as YuuKi_AnS admits this type of shared data isn’t guaranteed to be accurate but is “for reference only.”

There is something of an anomaly in the way Cinebench either uses or reports processor cores. The rendering benchmark appears to be somehow limited to 125C/256T. It is evident from the system info section of the benchmarking app. Meanwhile, the Windows Task Manager correctly shows that the AMD’  Genoa’ machine under test is a 2P system using a pair of 96C/192T processors working together. So clearly, Cinebench R23 should report and use 192C/384T of processing available but may not be; it might be limited to the readout - 125C/256T for now.

With the above system setup explanations and provisos out of the way, let us look closer at the results through the medium of a table.

According to the test results, the AMD ‘Genoa’ ES 2P system is 38% faster than the Intel Sapphire Rapids 2P system. It is because the AMD system wields more cores, and Cinebench R23 responds well to core scaling. Cinebench says the AMD system has 14% more cores, though according to Windows, it has 71% more cores, and if the benchmark were utilizing them, the score would be even better.

While consumers are getting excited at the impending launch of Zen 4 architecture Ryzen 7000 processors towards the end of this month, AMD’s EPYC Genoa parts (based on the same microarchitecture) are also causing ripples. A week ago, YuuKi_AnS posted a photo of an EPYC 9654 ES 96-core CPU (one of the CPUs in the tested system above?). The Genoa launch could be as soon as next month, but we aren’t sure as AMD isn’t as upfront about its non-consumer-related business plans. These powerful EPYC chips are scheduled for H2 2022, and we are in that part of the calendar now.

Intel has delayed the release of its 4th Generation Xeon Scalable "Sapphire Rapids" processor for a number of times without disclosing its reasoning. Last week the company admitted that it had to change up Sapphire Rapids because of a security bug, but it appears that the problem is bigger than Intel says. According to Igor's Lab, Sapphire Rapids had about 500 bugs that required the company 12 steppings to fix them. 

Intel's fouth Gen Xeon Scalable Sapphire Rapids' processor will not only increase core count to up to 60, but will bring in numerous new features, including Advanced Matrix Extensions (AMX), Data Streaming Accelerator (DSA), CXL 1.1 protocol, DDR5 and HBM2E memory support, PCIe Gen 5 interface, and many more. But the host of additional features increase probability of hardware bugs, so Intel had to fix almost 500 of them, Igor's Lab reports. 

So far, Intel has released A0, A1, B0, C0, C1, C2, D0, E0, E2, E3, E4 and E5 steppings of Sapphire Rapids processor to fix nearly 500 bugs. Given that modern processors integrate tens of billions of transistors, it is inevitable that have a certain number of bugs. They are called erratas and are mitigated with microcode or even software updates. But 500 erratas seems overwhelming, as does 12 respins considering that a respin costs tens of millions of dollars. 

Although it is expensive to build new respins, the more pressing issue is that Intel has to delay release of its next-generation datacenter CPUs. Right now, Intel targets 2023 calendar week 6 to 9 (Feb. 6, 2023 to March 3, 2023) launch window for high-volume Sapphire Rapids processors. Meanwhile, some SPR products may launch on 2022 calendar week 42 and 2022 calendar week 45. 

For Intel, the Sapphire Rapids processor and the Eagle Stream platform are crucially important products. Not only they are expected to improve Intel's competitive positions on the datacenter market, but they will open doors to the company's following generation products — the codenamed Emerald Rapids processor due in 2023.

Intel disclosed during its earnings call on Thursday that it had to respin its 4th Generation Xeon Scalable 'Sapphire Rapids' processor before initiating high-volume production. This redesign and tapeout took time, which is why Intel is delaying the volume ramp of SPR CPUs to late 2022. Meanwhile, the compute tile of Intel's Xeon 'Granite Rapids' processor, due in 2024, is on track to be powered later in Q3. Intel has also inked a deal with AWS to develop custom datacenter solutions. 

Not So Brilliant for Sapphire Rapids

As it turns out, Intel's previous iteration of its next-generation Xeon Scalable 'Sapphire Rapids' processor had a security vulnerability that required mitigation in hardware. 

As a consequence, Intel had to respin the CPU, mend the issue, tape out the new stepping, and then proceed with regular testing and validation procedures. This naturally postponed the volume launch of the product. Some Intel customers (probably among operators of hyperscale datacenters) are already running initial Sapphire Rapids processors unaffected by the security issue (perhaps because they come with certain features turned off). 

"We were not shipping at the quality levels, the security levels that we needed to […] clearly we should not have had that bug in the product in the first place," said Pat Gelsinger, chief executive of Intel, at the company's earnings call. […] So, we did another [stepping], which I will say for the larger-volume SKUs, and those will be volume shipping in the second half of the year." 

Intel has hundreds of customers with different requirements for its Xeon processors. For example, operators of hyperscale cloud datacenters usually require some specific features as well as high performance, but at times they do not need all the capabilities that Xeons have to offer. As a result, Intel can ship them new CPUs before general availability without any negative consequences. But for general customers like traditional server makers, Intel has to offer CPU models that will satisfy many types of customers with different workloads and software. Therefore, a bug in a chip may not affect giants like AWS, Meta, or Microsoft Azure. But it might affect enterprises with different needs and software stacks. Thus, it is better to fix the hardware issue before shipping. 

But a new stepping (and consequent retesting/revalidation) automatically delays high-volume launch, which is why Intel will make its 4th Gen Xeon Scalable 'Sapphire Rapids' processors generally available only later this year. So while the production ramp of those CPUs will begin in 2022, it will progress well into 2023. 

"Sapphire Rapids [is] ramping later," said Gelsinger. "We have some SKUs out, which is good, but the main SKUs are not out, and they happen later in the year. Of course, they will contribute way more significantly to next year than they are going to contribute to this year." 

Intel admits that its Sapphire Rapids processors are late to the party, and they will be widely available only in 2023. What remains to be seen is how this affects the availability of Intel's 5th Generation 'Emerald Rapids' CPU that is socket compatible with Sapphire Rapids (and is made using the same Intel 7 aka 10nm Enhanced SuperFin process technology as SPR) and is set for a 2023 debut. However, for now, Intel has no plans to delay Emerald Rapids and claims that this CPU indeed looks 'healthy.' 

"Emerald [Rapids] goes into the Sapphire [Rapids] platform, so we are working very closely with our customers and the timing there, the product is looking very healthy, so we are nicely on track," said Intel's CEO. "So that will be a 2023 product and then Granite [Rapids] and Sierra Forest are the 2024 products."

Granite Rapids to Power-On This Quarter

But while there are issues with Intel's 4th Generation Xeon Scalable 'Sapphire Rapids' processor, the development of one of its successors — the codenamed Granite Rapids CPU due in 2024 — seems to be proceeding OK.  

"We have now taped in the first stepping of the Granite Rapids CPU and expect power-on this quarter," said the head of Intel. "In the second half of this year, we plan to tape in numerous internal and foundry customer test chips on various process nodes including Intel 3 and Intel 18A." 

Intel's codenamed Granite Rapids is an important product for Intel. The CPU should arrive in 2024. It will use a brand-new platform that will succeed the upcoming socket LGA4677-based platform that will support Intel's 4th Gen Xeon Scalable 'Sapphire Rapids'ands 5th Gen Xeon Scalable 'Emerald Rapids' processors. It could carry the 6th Gen Xeon Scalable processor name while increasing performance and bringing additional functionality, which will improve Intel's competitive position on the server market. 

The compute tile of Granite Rapids will be made using Intel 3 fabrication technology that promises to offer denser high-performance libraries, increased intrinsic drive current, and reduced via resistance — three features particularly beneficial for datacenter processors. Meanwhile, Intel says that its I3 node will bring an 18% performance/watt gain compared to its I4 node. Intel's I3 process node has never been listed for any upcoming client PC product by Intel (Meteor Lake with an I4 compute tile will be succeeded by Arrow Lake with an I20A compute tile), so we might speculate that this technology is indeed very well suited for heavy-duty datacenter applications rather than for regular client products.

Custom Datacenter Solutions for AWS

In addition, Intel announced during the call that it would develop custom datacenter solutions for Amazon Web Services. 

"In Q2, we agreed to expand our partnership with AWS to include the co-development of multigenerational data center solutions optimized for AWS infrastructure and Intel as a strategic customer for internal workloads including EDA," said Gelsinger. 

Intel's wording about its expanded partnership with AWS is vague at best. The company did not say what exactly it plans to co-develop with the cloud giant. However, given the fact that Intel's main prowess is chip development, we could speculate that Intel is talking about datacenter-grade x86 system-on-chips (SoCs) or multi-chiplet/multi-tile system-in-packages (SiPs) 

Intel's intention to build highly-custom datacenter-grade x86 processors for large clients like AWS or Meta has always been one of the critical parts of its IDM 2.0 strategy. AWS is perhaps a perfect client for Intel and its foundry division. AWS consumes loads of chips, can take advantage of almost every IP that Intel has to offer, can bring properly developed in-house IP explicitly designed for AWS-enabled services and platforms like Netflix, and is willing to pay significant sums for the hardware it needs.  

Strategically, AWS's sticking with x86 means that the company will continue to use this architecture in the foreseeable future. Of course, AWS is a company that uses all available platforms, including AMD, Nvidia as well as Arm (e.g., Ampere Altra/Altra Max-based platforms). Still, custom Xeons, of course, indicate that AWS is very serious about tailored x86-based solutions.

Sapphire's Pulse Radeon RX 6500 XT 8GB, which recently released, is now available for $229.99. The Navi 24-powered graphics card is only $20 more expensive than the vanilla Radeon RX 6500 XT, one of the best graphics cards around if you're on a serious budget.

One of the more questionable aspects of the Radeon RX 6500 XT was the 4GB of GDDR6 memory. In a blog post titled "Game Beyond 4GB," published in 2020, AMD argued that 4GB simple wasn't enough for modern gaming. Therefore, AMD's decision to incorporate only 4GB of memory into the Radeon RX 6500 XT still makes many scratch their heads. However, Sapphire's Pulse Radeon RX 6500 XT 8GB solves the Radeon RX 6500 XT's memory deficiency. Unfortunately, AMD doesn't officially list a Radeon RX 6500 XT 8GB on its website, so it would appear Sapphire produced the 8GB version out of its own volition, and we may not see other vendors follow suit.

The Pulse Radeon RX 6500 XT 8GB clocks in slightly faster than the Pulse Radeon RX 6500 XT 4GB. The 8GB model features a 2,695 MHz game clock and 2,855 MHz, whereas the 4GB variant comes with a 2,685 MHz game clock and a 2,825 MHz boost clock. That's less than a 2% difference. The most significant upgrade lies in the memory system.

Sapphire's new model comes with double the memory as the original. However, the memory still operates at 18 Gbps across a 64-bit memory interface. Therefore, the Pulse Radeon RX 6500 XT 8GB offers the same memory bandwidth of 143.9 GBps as any other Radeon RX 6500 XT.

The Radeon RX 6500 XT is a 107W graphics card. Sapphire rates the 4GB and 8GB models with a board power of 130W. However, the slightly faster clock speeds and the double memory forced the manufacturer to swap PCIe power connectors. As a result, the Pulse Radeon RX 6500 XT 8GB has an 8-pin PCIe power connector as opposed to the 6-pin one on the 4GB counterpart. Display outputs still come down to a single HDMI 2.1 port and DisplayPort 1.4 output.

The Sapphire Pulse Radeon RX 6500 XT 8GB retails for $229.99 on Newegg. That was the original price tag on the Sapphire Pulse Radeon RX 6500 XT 4GB. The 4GB model sells for $209.99 before discounts and promotions. You get double the memory for $20 (10%). The problem is that other custom Radeon RX 6500 XT 4GB models start at $185, so the Sapphire graphics card already carries a premium over its rival. To worsen the situation, the Radeon RX 6600 is available for as low as $259.99. For many consumers, it makes sense to go up to the Radeon RX 6600, which already sports 8GB and offers significantly higher performance.

Without any doubt, Intel's 4th Generation Xeon Scalable 'Sapphire Rapids' server processors are among the highly anticipated products in 2022. The CPUs were delayed several times, and their specifications remain a mystery. But as Intel's partners are testing samples of the new processors, their benchmark results inevitably emerge in various databases. This time around, someone posted results from Intel's 56-core Xeon Platinum 8480+ processor to Primate Labs's Geekbench 5 database. 

As the name suggests, Intel's 56-core Xeon Platinum 8480+ with 112MB L2 cache and 105MB L3 cache belongs to the company's crème-de-la-crème server CPUs with all cores and features (AMX, AVX-512, CXL, DSA, etc.) enabled. Considering Intel's 3rd Gen Xeon Scalable nomenclature, we may be dealing with the new top-of-the-range SKU.  

The Xeon Platinum 8480+ supports HyperThreading and has a base frequency of 2.0 GHz, according to the Geekbench 5 database entry. The relatively low clocks of the Xeon Platinum 8480+ (down from 2.30GHz in the case of the 40-core Xeon Platinum 8380 CPU) are a bit surprising because it is made using Intel's 10nm Enhanced SuperFin fabrication technology that features advanced power delivery and is generally designed to enable high frequencies on CPUs with a high core count. Perhaps the new Sapphire Rapids processors feature very high turbo clocks, but unfortunately, we have no idea how high the Xeon Platinum 8480+ CPU can go in turbo mode. 

Primate Labs's Geekbench 5 is hardly the best benchmark for server CPUs (especially modern ones with various special-purpose accelerators). Still, it gives some clues about the integer, floating point, and multi-thread performance. Still, considering that we are dealing with pre-production hardware and firmware (Tyan's S5652AGM3NRE-2T LGA4677 motherboard was used), take the benchmark results with a grain of salt.

The performance of Intel's Xeon Platinum 8480+ 'Sapphire Rapids' processor in Geekbench 5 is a mixed bag: it wins in certain cases but loses badly in others.

The CPU is considerably behind AMD's 64-core EPYC processors (that also feature rather low base clocks) in single-thread workloads. Intel's Golden Cove microarchitecture (used by Sapphire Rapids and Alder Lake CPUs) has very good single-thread performance, so perhaps the reason why Xeon Platinum 8480+ performs so poorly is because the pre-production chip does not boost its clocks (i.e., appropriate modes are disabled).

In multi-core workloads, Intel's Xeon Platinum 8480+ shows its potential and outperforms AMD's 64-core EPYC 7763 processor. Meanwhile, it cannot beat AMD's 64-core EPYC 7773X CPU, which is armed with 3D V-Cache, as well as Ryzen Threadripper Pro 5995WX, which has considerably higher clocks. 

In general, Intel's 56-core Xeon Platinum 8480+ 'Sapphire Rapids' processor looks rather promising for pre-production hardware. It looks like, for now, Intel hasn't enabled boost clocks on samples of SPR CPUs for certain reasons, but what the 56-core/112-thread processor can do in multi-thread workloads at 2.0 GHz looks pretty impressive.

Sapphire surprised us by announcing a new Radeon RX 6500 XT configuration with 8GB of memory. The formerly entry-level graphics card now comes with double the original model’s 4GB quota. Nothing else has materially changed, the GPU and memory performance specs are the basically same, but we expect to see a price bump with the additional VRAM. The RX 6500 XT does claim a spot among the best graphics cards, mostly thanks to its budget pricing, but we'd much rather have a faster GPU for a bit more money.

The AMD Radeon RX 6500 XT had a tough time when it was launched, as it was quite widely dismissed as underpowered and under-featured. However, it seems like both Intel and Nvidia have sought to make it look like a champ by releasing competing graphics cards like the Arc A380 and RTX 1630 in the interim period. AMD’s own Radeon RX 6400 also makes the RX 6500 XT seem like quite a muscular offering.

When we tested the RX 6500 XT, memory capacity wasn't our primary concern. We noted that the specs and features had been cut too far from the RX 6600 and above. There were indeed concerns with the memory configuration, but it was more the 64-bit bus, not just the 4GB. There were further concerns with using a PCIe x4 connection — not terrible with PCIe 4.0, but a particular problem on older gen systems with slower PCIe standards — plus the limited display connectivity (one HDMI and on DisplayPort), and finally the complete lack of hardware video encoding support. In brief, no one was asking for more VRAM, except maybe Sapphire's marketing dept.

Above you can see exactly what Sapphire has changed for its Pulse RX 6500 XT model. The GPU max boost clock speed is technically 1% faster, but that could reasonably be described as negligible.

Most people would you play modern AAA games on the 6500 XT with low-ish settings due to the entry-level GPU, which also lowers the VRAM requirements. The larger memory capacity can help in some games with certain settings (like texture and shadow resolution), but the GPU computational power and memory bandwidth will quickly become the bottleneck.

Sapphire tweeted that the new Pulse Radeon RX 6500 XT 8GB is on shelves now. We haven't been able to find one for sale, which means we don't have a price yet. That will be a critical factor, considering the RX 6500 XT 4GB starts at $189, while the substantially faster RX 6600 8GB starts at $279, and the lowest price on a Sapphire RX 6500 XT 4GB is $199.

For reference, in our GPU benchmarks, the RX 6600 was 80% faster than the RX 6500 XT at 1080p medium and over twice as fast at 1080p ultra, making it a far more capable GPU overall. That's because it has 75% more GPU cores and double the memory interface width, which are both important for real-world gaming performance. It also has an x8 PCIe interface, supports up to four monitors, and includes full hardware video encoding support.

VideoCardz has posted the potential specifications for Intel's upcoming Sapphire Rapids-WS processors. However, it's crucial to highlight that these core-heavy chips target the workstation demographic, unlike Sapphire Rapids-SP, which caters to data centers.

Sapphire Rapids utilizes the same Intel 7 (rebranded 10nm Enhanced SuperFin) process and Golden Cove cores as Intel's mainstream Alder Lake lineup. The original plan for Sapphire Rapids was to compete against AMD's EPYC 7003 (Milan) processors. However, Sapphire Rapids has suffered multiple delays and, by how things look, will likely end up taking on the Zen 4-powered EPYC 9003 (Genoa) lineup. On the other hand, Sapphire Rapids-WS targets Team Red's Ryzen Threadripper Pro 5000 WX-series (Chagall) army. The Zen 3-powered processors are on TSMC's 7nm FinFET process, while Sapphire Rapids is a product of a 10nm process node.

According to the document that VideoCardz obtained, which the publication didn't share, Sapphire Rapids-WS support DDR5-4800 memory natively. The 10nm chips only support 1S configurations. The document reportedly didn't confirm the number of memory slots or PCIe 5.0 lanes. However, Sapphire Rapid-WS allegedly comes with 8-channel support and 112 PCIe 5.0 lanes. If comparison, Chagall is still on DDR4 memory and offers 128 PCIe 4.0 lanes.

Intel Xeon Sapphire Rapids-WS Specifications*

*Specifications are unconfirmed.

Sapphire Rapids-WS retain the same hierarchy as Intel's mainstream chips, but logically, substituted the Core branding for the Xeon moniker. Therefore, the stack consists of the W9, W7, and W5 SKUs. As always, we should treat the specifications with caution since Sapphire Rapids-WS is unreleased hardware and we don't know they state of the workstation processors.

The Xeon W9-3495X, one of the top SKUs, wields 56 cores, 112 threads and 105MB of L3 cache. The chip reportedly has a 1.9 GHz clock speed with a 350W TDP. It's still a few steps behind Ryzen Threadripper, though. For comparison, the Ryzen Threadripper Pro 5995WX has 64 cores, 128 threads and 256MB of L3 cache. AMD's chip has a 280W TDP. which is lower than the Xeon W9-3495X.

The W7 tier include SKUs with core counts spanning from 20 to 28 Golden Cove cores. The L3 cache varies from 52.5MB to 75W with TDP ranges between 270W and 300W. The W5 category. for example, houses chips with 12 to 16 cores. The L3 cache is more modest with parts sporting 30MB up to 45MB. The TDP for the Xeon W5-series ranges from 200W to 270W.

It's unknown when Sapphire Rapids-WS processors will hit the retail market. AMD's latest desktop processor roadmap reveals the chipmaker's plans to bring its Zen 4 cores over to the Ryzen Threadripper family next year. Sapphire Rapids-WS will face some tough competition ahead.

Intel already disclosed that it had begun shipments of QVL 4th Generation Xeon Scalable 'Sapphire Rapids' processors to select customers and is sampling these processors with a broad set of clients. However, Intel has not confirmed that its next-generation Xeon W-3400-series CPUs are already in the wild and being tested by third parties. Yet, this appears to be the case, and people are test driving Intel's Xeon W9-3400-series 'Sapphire Rapids' CPUs 

Some openly available Linux boot logs already contain mentions of Intel's 56-core Xeon W9-3495 'Sapphire Rapids' processor, as noticed by @InstLatX64. This is hardly surprising as Intel tends to release Xeon W-series processors for workstations based on server silicon shortly after it launches its latest datacenter CPUs that use the same microarchitecture. 

Intel's Xeon W-series processors are designed for single-socket workstations (like AMD's Ryzen Threadripper Pro), yet they retain all features and capabilities supported by their server counterparts, including support for massive amounts of memory over eight memory channels. The CPU in question — Intel's Xeon W9-3495 — not only has 56 cores with simultaneous multi-threading clocked at 1.80 GHz and eight DDR5-4800 memory channels, but it also comes with AVX-512 and AMX instructions enabled. Of course, since the information does not come from an official source and involves pre-production hardware, take it with a pinch of salt. 

One of the things that strikes the eye with the 56-core Sapphire Rapids processor for desktops is Intel's new naming scheme for Xeon W-series. Previously the company just offered these CPUs with a particular model number without splitting them into classes like it does with desktop CPUs (e.g., i9, i7, i5, i3). However, starting with the next-generation Xeon W-series products, Intel will add classes to the chips, so there will be Xeon W9-3400-series CPUs, Xeon W7-3400-series products, and so on. It is unclear what might be the difference between Xeon W9 and Xeon W7 besides core count and perhaps supported memory configurations as well as the number of active PCIe lanes. 

While it is nice to know that Intel's next-generation Xeon W-3400-series 'Sapphire Rapids' workstation-grade CPUs with up to 56 cores are already in the wild, what PC enthusiasts are eager to know is how enthusiast-grade Sapphire Rapids for high-end desktops (HEDT) are doing. Unfortunately, we have no idea. However, since Intel is gradually expanding the availability of Sapphire Rapids samples, there are good chances that someone is already test-driving Sapphire Rapids for HEDTs. Unless, of course, Intel wants to get rid of its HEDT series and offer only workstation-grade CPUs in the future, as AMD did with its Ryzen Threadripper/Threadripper Pro offerings.

A renowned hardware blogger has gained access to engineering samples of Intel's 4th Generation Xeon Scalable Sapphire Rapids processor and Intel's 3rd Generation Optane DC Persistent Memory 'Crow Pass' modules. Unfortunately, engineering samples of upcoming hardware do not always perform like commercial parts, which seems to be the case here. But there are several important points to mention.

Hardware bloggers, leakers, and big Intel clients have been playing with engineering samples of Intel's 4th Generation Xeon Scalable Sapphire Rapids CPUs for some time. So it's not surprising since Intel has been shipping QVL samples of these processors for months now. Yuuki_Ans, a well-known hardware collector with access to rare parts, is the first to obtain Intel's next-generation Optane DC Persistent Memory modules codenamed Crow Pass that use a DDR5 interface. The fact that the blogger got these modules is important as it proves that Intel is sampling its 4th Generation Optane DC Persistent Memory modules with select clients. So there are good chances that the company does not have plans to drop the technology for now.

While even getting 4th Generation Optane DC Persistent Memory modules was a big deal, Yuuki_Ans tried to test them and even posted benchmark results on his Twitter.

Intel's Optane DC Persistent Memory modules are not the hottest products on the market. However, they cater to particular server workloads and increase near-CPU memory capacity significantly below the price of appropriate traditional memory modules. To that end, you will not find many reviews of Optane DC Persistent Memory modules on the Web. Yet, we compared LegitReview's results of Intel's 1st Generation Optane DC Persistent Memory 'Apache Pass' to Yuuki_Ans's results of Intel's 3rd Generation Optane DC Persistent Memory 'Crow Pass.'

Performance of Intel Optane DC PM Modules in CrystalDiskMark

As it turns out, Intel's 3rd Gen Optane DC PM modules working with Intel's Xeon Scalable Sapphire Rapids processor are slower than Intel's 1st Gen Optane DC PM modules paired with 2nd Gen Xeon Scalable Cascade Lake CPU.

Of course, we are dealing with results obtained in only one benchmark using pre-production hardware, so take them with a grain of salt. Furthermore, from time to time, companies supply their pre-production parts with firmware and microcodes that deliberately lower performance to allow rivals to evaluate the performance of future hardware before release.

In any case, the key takeaway is that Intel is sampling its 3rd Generation Optane DC Persistent Memory modules and plans to offer the technology with its next-generation Xeon Scalable 'Sapphire Rapids' platforms.

The AMD Radeon RX 6950 XT launched last month, along with the RX 6750 XT and RX 6650 XT, bringing 18Gbps GDDR6 memory along with boosted clocks and higher power requirements. The Sapphire Nitro+ Pure that we reviewed initially performed quite well. Today's MSI Gaming X Trio is just a hair faster, but in getting there MSI seems to have increased power draw substantially over the competition. The RX 6950 XT ranks as one of the best graphics cards, and even holds the top spot for 1080p and 1440p gaming in our GPU benchmarks hierarchy (but not with ray tracing), but MSI's variant wouldn't be our first pick.

Here's a look at the specs for AMD's reference 6950 XT alongside the MSI and Sapphire cards we've had in for review — AMD elected to not sample media with its reference design, helping to ensure slightly higher overall performance, just like Nvidia did with the RTX 3090 Ti. 

Interestingly, MSI lists a power consumption of just 340W for its RX 6950 XT Gaming X Trio, but that feels more like the TDP for the GPU rather than the full card power. We've actually tested two different cards (the first died after a couple of days for unknown reasons), and both consumed far more than 340W at their default settings. In fact, the second card used a few more watts than the initial card. We'll get to the actual power measurements later, but we saw closer to 430W out of MSI's card, and enabling Rage mode (with its 10% bump in power limits) pushed that up to 440W. Basically, it was right in the same ballpark as the RTX 3090 Ti.

Core specs are mostly identical among the three RX 6950 XT cards, with GPU boost clocks and TBP being the only real difference. AMD gives a reference clock of 2310 MHz, Sapphire pushes that to 2435 MHz, and MSI tacks on an additional 19 MHz for 2454 MHz. While previous-generation AMD GPUs often listed a theoretical boost clock that was higher than you'd typically see in practice, that changed with the RDNA 2 architecture and the RX 6000-series GPUs. At default settings, we saw average GPU clocks of more than 2.5GHz, and with Rage mode the MSI card actually broke 2.6GHz.

The big concern right now with the RX 6950 XT is that we expect AMD will launch an RX 7900 XT before the end of the year, possibly as soon as October. That's not too far away, and AMD has publicly declared that it expects RDNA 3 to deliver a 50% improvement in performance per watt. It will also be the first GPU to use chiplets, and while the topology of such a design isn't entirely clear, this could be the "biggest Navi" yet. Amid falling GPU prices and the pending next-gen architectures, it's difficult to recommend rushing out to buy a soon-to-be-outdated GPU for over $1,000 right now.

Speaking of which, it's interesting that the retail prices on the RX 6950 XT have been coming down fast. Officially the GPU has a $1,099 MSRP, and at launch we saw multiple companies pushing factory overclocked models for $1,249 or more. Right now, Newegg has multiple RX 6950 XT cards selling for MSRP, and we've seen some sales come and go that dropped prices as low as $1,030. The MSI Gaming X Trio is one of the cards selling for MSRP, which does put it $150 below the Sapphire Nitro+ Pure, but when the cards launched last month, both were targeting the same $1,250 price point — MSI just happened to correct faster.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

The MSI Radeon RX 6950 XT Gaming X Trio uses the familiar triple fan cooler design that we've seen from MSI for several years now. It's a slight step down from MSI's top-tier Suprim X design, which is only used on Nvidia GPUs, and gives additional RGB strips on the front of the card. There's also a Gaming Trio (no "X") that uses the exact same design but with a 2324 MHz boost clock.

The MSI card measures 325x141x55 mm, a 2.5-slot width that nevertheless fully blocks any adjacent expansion slot. It would give a bit more breathing room if you were to use the next slot after that, however, which is why many companies seem to be avoiding the full triple slot width. Compared to the Sapphire Nitro+ Pure, MSI's card is a hair longer, 1cm taller, but 1.6cm thinner. Overall, it's similarly bulky and will need a large and spacious case, and MSI's card also weighs 1594g, just 12g more than the Nitro+ Pure.

There's not a ton of RGB lighting on the Gaming X Tri, unlike with MSI's newer Suprim X design. There are six diagonal LEDs on the front of the card (which will usually end up facing the bottom of the PC case), the MSI logo and dragon shield icon on the top of the card, and then a longer LED strip that runs about half the length of the card and can easily be seen from the back and top views. It's enough RGB lighting for most people, but not quite in the same "burn your eyes out" realm as cards like the Sapphire RX 6950 XT Toxic.

Display connectivity consists of the ubiquitous triple DisplayPort and single HDMI 2.1 outputs . There might be a few people that would prefer dual HDMI ports, since those technically have higher resolution and bandwidth support than DP1.4, but all the ports can drive at least 4K at 60Hz — check our DisplayPort vs HDMI guide for more information.

MSI includes three 8-pin power connectors on its RX 6950 XT, which means up to 75W more power (while remaining in spec) compared to the Sapphire card. That's not necessarily a good thing. AMD specs the 6950 XT at 335W for the reference model, in which case dual 8-pin connectors plus the PCIe slot should be sufficient. As we'll see later, the Gaming X Trio blows well past the 400W mark, which means the extra power connector was required. However , its competition manages to get by with dual 8-pin and a single 6-pin, and uses about 70W less power. Perhaps the additional power headroom was required to eke out every last ounce of performance, but the 1–2% increase we'll see in a moment hardly warrants it.

For cooling, MSI has three of its 90mm Torx 4.0 fans with a partial outer rim that helps improve airflow and static pressure. I do wonder how much of a difference the removal of half the rim makes, but cooling at least didn't appear to be a problem. As with so many other GPUs, it's also interesting to note that the heatsink fins are oriented parallel to the IO bracket at the back, meaning that all the perforations hardly matter — most of the heat from the graphics card will be vented into your case, so you'll want at least one or two exhaust fans.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

We've updated our GPU test PC and gaming suite since early 2022, with a Core i9-12900K processor, MSI Pro Z690-A DDR4 WiFi motherboard, and DDR4-3600 memory (with XMP enabled). We also upgraded to Windows 11 to ensure we get the most out of Alder Lake. We'll probably continue using the same core hardware for at least another year, unless it turns out that Zen 4 or Raptor Lake can deliver a serious improvement in gaming performance (which we doubt will happen).

We are not showing professional application performance with the MSI card, as it was basically the same story as we saw with our initial MSI RX 6950 XT content creation results. Refer to those charts, and add 1% if you want to know how the MSI card compares to the Sapphire card. Newer drivers did not appear to make a difference.

Our gaming tests consist of a "standard" suite of eight games without ray tracing enabled (even if the game supports it), and a separate "ray tracing" suite of six games that all use multiple RT effects. For this review we'll be testing at 4K, 1440p, and 1080p at "ultra" settings — which generally means maxed out settings, except without SSAA if that's an option. Our GPU benchmarks hierarchy also shows 1080p medium performance, if you want to see that.

Each card gets tested multiple times at every resolution, to ensure that the performance we report is consistent. The first run gets discarded, and we use the better result of the two additional runs — after checking that there's no major variance between the two runs. There's one exception to this: Far Cry 6, which at 1080p typically has a very fast first run, a slightly slower second run, and third, fourth, etc. runs often continue to drop in performance. We use the second run at 1080p as a compromise.

We initially tested the RX 6950 XT using AMD's pre-launch drivers, which are basically the same as the 22.5.1 release. AMD then delivered a "DirectX 11 Preview Driver" that was supposed to help with DX11 gaming performance, and rolled that into its 22.5.2 update, so we retested the MSI card with the newer drivers. You'll see both results in the charts, but remember the Sapphire RX 6950 XT was tested with 22.5.1, and in general we saw about a 3% average uplift in performance for our test suite from 22.5.2. (And yes, AMD just released 22.6.1 drivers, but those are only for Windows 7 holdouts.)

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

4K ultra is a good place to start with extreme GPUs like the RX 6950 XT. In our standard test suite, the MSI card ends up nearly tied with the Sapphire card — a 0.3% lead is definitely within the margin of error. Updating to the newer 22.5.2 drivers improved performance by just 1.3% overall, with very slight gains in every game we tested. CPU bottlenecks shouldn't be a factor at 4K, which is probably why the gains are a bit lower here than we saw at some of the other resolutions.

Even with the updated drivers, the RTX 3090 Ti still held down the top position, beating the RX 6950 XT by 5%. A big part of that is the fact that even a 128MB Infinity Cache can't quite hold all the data that might be needed. AMD has said cache hit rates drop from around 75% at 1440p to about 60% at 4K, which means memory bandwidth becomes more of a factor and the RX 6950 XT just doesn't have enough to keep up with the 3090 Ti. Still, it's an impressive showing and AMD still takes second place overall.

The individual game charts show areas where AMD excelled, and others where it fell far short. Far Cry 6 and Forza Horizon 5 put AMD well ahead of Nvidia's best, but conversely Flight Simulator and Total War: Warhammer 3 had multiple Nvidia GPUs outpacing AMD's best. In general, though, AMD still delivered better performance per dollar for most of its high-end and extreme GPUs, at least for standard gaming performance.

As usual, everything falls apart when we max out settings at 4K with our ray tracing test suite. Even the RTX 3090 Ti only managed just over 31 fps on average, with half of the games falling below 30 fps. The RX 6950 XT ended up in sixth place, behind the 3090 Ti, 3090, 3080 Ti, 3080 12GB, and 3080 10GB. Also note that the newer 22.5.2 drivers barely affected performance for the MSI card, yielding a 1.5% average improvement that only works out to a 0.3 fps difference.

For AMD, the RX 6950 XT only managed more than 30 fps in one of our six DXR (DirectX Raytracing) games, Metro Exodus Enhanced Edition — and only barely, at 33 fps with the latest drivers. Four of the games couldn't even manage 20 fps, either, so they weren't even close to being playable at 4K. Nvidia has the advantage of DLSS support, which remains far more common in DXR games than FSR — only Cyberpunk 2077 has FSR 1.0 as an option right now among our test suite.

I was curious about whether AMD would add tensor core-like hardware to RDNA 3, but after speaking with Sam Naffziger, it sounds like AMD will be content to include machine learning matrix hardware only on its CDNA line. Even without matrix cores, AMD delivers a decent amount of FP16 compute, which might be enough for an alternative upscaling solution — or just stick with FSR 2.0, which only has about a 7% performance hit compared to no upscaling and can compete reasonably well with DLSS. Now we just need more games to support FSR 2.0. 

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

While the GPU companies might like to talk about 4K, 1440p ends up being a far better target for most people, especially since high refresh rate 1440p gaming monitors are reasonably affordable. AMD also gets better cache hit rates, with a resulting net 65% boost in performance compared to 4K. Nvidia's RTX 3090 Ti by contrast only got 51% more performance, which allows AMD to claim the overall performance crown at 1440p ultra, beating out the nearest competition by 4%.

Looking at the updated drivers, 22.5.2 still only improved the MSI card's performance by 1.3%. Interestingly, the MSI card was actually slower than the Sapphire card at 1440p when using the same pre-launch 22.5.1 drivers, though not by much (a 0.2% difference).

Turning to the individual game charts, AMD took top honors in five of the eight games, while Nvidia still claimed wins with the RTX 3090 Ti in Flight Simulator, Red Dead Redemption 2, and Total War: Warhammer 3. There's also DLSS support in three of our standard test suite games (Horizon Zero Dawn, Red Dead Redemption 2, and Watch Dogs Legion), and we've measured a 10%–20% improvement with DLSS Quality mode in those. None of the standard test suite games have FSR support, unfortunately, though AMD still gets the overall win at native rendering.

Ray tracing didn't do any favors for AMD, as before, where the MSI card with updated drivers again landed in sixth place. It's ahead of the RTX 3070 Ti and previous generation Titan RTX, but that's about all it could manage. The RTX 3080 was still 17% faster, while the RTX 3090 Ti had a 52% lead. Even if you look at the comparably priced RTX 3080 Ti, that's a 32% advantage, and the RTX 3080 12GB can now be had for under $800 and had a 28% lead.

The MSI RX 6950 XT could at least play all of the games, though Cyberpunk 2077 would still be borderline at 29 fps. It will be interesting to see if AMD puts some serious effort into improved ray tracing hardware and performance with RDNA 3, or if it will continue to cede that segment to Nvidia. Considering the latest gaming consoles have about half the ray tracing capacity as the 6950 XT, there's still an uphill battle in getting game developers to enable higher quality RT effects, but FSR 2.0 and DLSS do help make it more viable.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Wrapping up our performance testing, we have 1080p ultra, and the RX 6950 XT once more reigns supreme. We started to hit CPU bottlenecks at this resolution, and the 6950 XT was only 20% faster at 1080p than at 1440p. Some games might be able to push hundreds of frames per second, but many will still struggle to take advantage of even 144Hz refresh rates, never mind the 240Hz, 360Hz, and even 480Hz displays that are now being marketed to esports professionals.

MSI's card took the top spot, with the updated drivers again improving overall performance by a mere 1.3%. We've heard of much larger gains when using the Ryzen 9 5950X, and AMD's CPUs appear to have been a major target with the DX11 improvements. Using an Alder Lake Core i9-12900K, the biggest improvement we measured out of all the gaming tests was only 3.7%, in Total War: Warhammer 3 at 1080p medium settings.

Nvidia still held onto the lead in Flight Simulator, Forza Horizon 5, and Red Dead Redemption 2, but the remaining games all have AMD in the lead, with a 6% overall average lead. The individual games range from the MSI RX 6950 XT being anywhere from 7% slower than the Asus RTX 3090 Ti (Flight Simulator) to as much as 23% faster (Forza Horizon 5). We didn't include the 1080p medium charts here, but the overall standings were the same as at 1080p ultra, with AMD holding a 6% lead.

What about ray tracing? The story remained unchanged, with the MSI card still in sixth place overall, leading the Sapphire card by 0.2% if we're being pedantic. The updated drivers aren't necessarily targeted at ray tracing improvements, but we still saw a 1.5% improvement at 1080p ultra. DXR remains so demanding that the RTX 3090 Ti was also still 45% faster than the MSI 6950 XT, though that's slightly closer than at 1440p ultra.

Dropping the settings down to 1080p medium with 'medium' DXR effects, which is a bit of a crock since things can actually look better at 1080p ultra without DXR and tend to run quite a bit better, the RTX 3090 Ti's lead shrank to just 34%. Even with DXR, CPU bottlenecks are apparently a problem at 1080p. But for roughly the same price as the RX 6950 XT, Nvidia's RTX 3080 Ti remained 22% faster at 1080p ultra DXR, and 20% faster at 1080p medium DXR — and that's without factoring in DLSS. 

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

For our power, temperature, clock speed, and fan speed testing, we use Powenetics testing hardware and software. We capture in-line GPU power consumption by collecting data while looping Metro Exodus at 1440p ultra as well as while running the FurMark stress test at 1600x900. Our test PC remains the same old Core i9-9900K as we've used previously, to keep results consistent. We tested the MSI card in its default mode and also enabled AMD's Rage mode in the drivers, which applies a potential 10% power limit increase. This is where the MSI card unfortunately fails to impress.

In theory, MSI lists 340W as the power draw for the RX 6950 XT Gaming X Trio. It labels that as "Power Consumption," but with our in-line power measurements, we saw far more than 340W. In Metro Exodus, we saw average peak power use of 429W, and FurMark showed even higher power use of 438W. We weren't running any MSI utilities or other software that would affect power use, so this represented the out of box experience. AMD's Rage mode gives a potential 10% increase in power limits, but here the MSI card only used 3% more power in Metro Exodus, though it did consume 6% more while running FurMark.

It's pretty obvious that MSI pushed power limits a lot higher than other GPUs with the Gaming X Trio, all in pursuit of the last few ounces of performance. As we showed in our gaming benchmarks, when using the same 22.5.1 drivers, things were so close between the MSI and Sapphire cards that it's hardly worth mentioning. We saw a range of up to 1.4% higher performance and up to 0.9% lower performance, and a lot of that is just the normal margin of error for testing GPUs, with an overall 0.2% difference on average.

Looking at the GPU clock speeds, you would expect more of an advantage for MSI, but without Rage mode the gap was still only 3%, and we didn't measure GPU clocks across our entire test suite. It's very likely that in many of the games, average GPU clocks were only 1% higher at most. With Rage mode, the difference in clock speed shrank to 2% in Metro Exodus. FurMark wasn't any different, showing about 3% higher GPU clocks for the MSI card with and without Rage mode, but again, MSI's card used up to 16% more power in FurMark, and 11–13% more power in Metro.

Temperatures were pretty similar across most of our extreme GPU test suite. The 3070 Ti and 3080 Ti (Founders Editions) ran a bit hotter, though custom AIC partner models tend to do a lot better, and the MSI card peaked at around 74C in our gaming tests, and up to 76C in FurMark. Rage mode bumped up fan speeds, so peak temperatures were 1C–3C lower than at stock for the MSI card.

Due to its higher power use, it's no surprise that the MSI card needed higher fan speeds to stay cool. At stock, fans averaged 1800 RPM, which is higher than the Sapphire card in Rage mode for gaming. Somewhat curiously, fan speeds started higher with both the RX 6950 XT cards, and then dropped after a minute or so to lower values, almost like the cards were overcompensating for potential GPU heat than never fully arrived. We've seen a bit of that with some of the other RX 6000-series GPUs, like the reference 6900 XT, but the Nvidia GPUs tended to be a bit more consistent.

Checking noise levels, we used an SPL (sound pressure level) meter pointed at the center fan at a distance of 10cm, which is done to limit the impact of environmental noise. The ambient noise level in the room, running the test PC but with the GPU fans off (since there was no GPU load) was 32–33 dB(A). Under load, running Metro Exodus for over 30 minutes (not looping), fan speeds reached a stable 29% and we measured 42.1 dB(A) of noise. Enabling Rage mode caused the fans to always run, so ambient noise levels increased to 35.5 dB(A) and our gaming test hit 43.3 dB(A) with 32% fan speed. In both cases, the MSI card was about 5 dB louder than the Sapphire card.

We also tested with a static fan speed of 75%, which was far higher than anything we expect most people will ever encounter during normal use. The Gaming X Trio reached 58.5 dB(A) of noise, at a close distance, which is pretty similar to what we've seen with other high-end enthusiast cards. 

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Without a new architecture, there's only so much AMD's RX 6950 XT can do to take on its Nvidia rivals. AMD has proven it knows how to get GPUs running at much higher clocks with RDNA 2, and RDNA 3 might be the first consumer GPU to break 3 GHz if the current trend continues. That would be pretty awesome, even if it uses a lot of power. But in the here and now, Navi 21 can lay claim to the 1080p and 1440p standard gaming performance thrones while falling behind at 4K and trailing with a large deficit in ray tracing performance.

The problem is that MSI's top-tier variant of the RX 6950 XT pushes power use far higher than its rival, without any real gains in performance. Frankly, it's flabbergasting that Sapphire can use the same GPU and effectively match MSI's performance while using 70W less power. We're not sure if the credit goes to some great engineering on the Nitro+ Pure, or poor engineering on the MSI Gaming X Trio, but if you have a choice between these two cards, Sapphire walks away with an easy victory.

Sapphire's card looks better (in my opinion), runs just as fast, and uses less power. And if you don't want to go with the Sapphire card for whatever reason, there are multiple other Radeon RX 6950 XT options that basically can't be worse than this MSI model. Okay, maybe they have 100 MHz lower boost clocks, but for $200 less, that's a far more sensible choice.

That's assuming you even want to go with an RX 6950 XT, of course. If you're not looking for the top performing AMD GPU right now, the RX 6900 XT is only about 6% slower and can be found for as little as $800, with several other models going for $900. There's also the RX 6800 XT, which is supposed to start at $650 but currently costs nearly as much as the least expensive 6900 XT. For the same price as the RX 6900 XT, you can also pick up a GeForce RTX 3080 12GB, and while it might fall behind in some games, the superior ray tracing hardware and DLSS support certainly makes that a good alternative. Even the formerly overpriced RTX 3080 Ti is worth a look, starting at under $1,100.

AMD does try to sweeten the pot a bit with some gaming extras, via its current Raise the Game Bundle, which you won't get from Nvidia. If you buy an RX 6950 XT (or any GPU from the RX 6700 XT and up), from a participating retailer, you'll get codes for the Saints Row reboot, Sniper Elite 5, and Forspoken (when it comes out). Game bundles aren't the same as buying games at retail, though, since you don't get a choice and AMD pays a lot less per game than the retail price — call it a $50 add-in if you want. That's fine, but for a GPU that costs over a grand, it's not some amazing offer.

We should see AMD's next generation architecture launch within a few months, so even with GPUs now on shelves at better prices than we've seen in two years, we still recommend waiting. If you simply have to buy a new graphics card right now, we suggest you pick up something for closer to $300 like the Radeon RX 6600 that can tide you over until RDNA 3 and Ada Lovelace arrive, rather than purchasing an extreme GPU that will invariably feel outdated and slow before 2022 wraps up.

Regardless, MSI's inefficient take on the RX 6950 XT just doesn't make a lot of sense. If it used 15% more power and provided at least a 5% boost to performance, it would be easier to recommend. Using 15% more power for the same performance, though? No thanks. More power means more heat in your case, which impacts noise levels and your PC in general. Even if you're using your PC as a space heater while gaming, we'd still prefer a card that uses less power and delivers the same performance. 

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

A couple of weeks ago, we reported on some leaked Sapphire Technologies data sheets which appeared to reveal a new AMD graphics card called the "Radeon 6700," expressly lacking the RX at the front of the product name. A few hours after that story went live, Sapphire published its product page, spilling the full tech specs of these cards, as well as providing further concrete proof of their impending launch. Now, Sapphire's product pages have been updated and we find out the leaked cards were supposed to be "Radeon RX 6700" models all along. Coincidentally, the Radeon RX 6700 has now also appeared in AMD's consumer graphics card web pages.

Sapphire's use of Radeon non-RX nomenclature had raised some questions about whether these new SKUs had something missing, but it now appears it was just a typo. Nevertheless, as one of AMD's closest graphics card partners, it is surprising that the mistake perpetuated so far as appearing in Sapphire product data sheets and packaging renders.

The Sapphire URLs which delivered product pages about "Radeon 6700" graphics cards have been deleted, so the old URLs in our original story now lead to 'page not found' errors. Instead, Sapphire has hosted product pages with appropriate URLs, including the previously missing RX text.

Now AMD and Sapphire have both published specs for the Radeon RX 6700 (non-XT), we can see that the partner design is slightly overclocked compared to reference, and below is an updated specs table with both reference and Sapphire designs betwixt their nearest neighbors.

Pondering over the reference and Sapphire product pages, we spotted one other difference, in regards to power consumption. AMD says that the typical Board Power (Desktop) of the reference card is 175W. Meanwhile, Sapphire says their Pulse model consumes up to 220W. Both card designs are fitted with a single 8-pin power connector. We would have to test some product samples to see what is happening here, checking the actual power differences and resulting performance differences, when these cards are under stress.

We look forward to testing these new graphics cards when they are but we don't have any official launch date or pricing indicators yet. In the meantime, if you are on the hunt for a new GPU today, you can take a check out our  Best Graphics Cards for Gaming in 2022.

Intel said this week that it would slightly delay high-volume ramp of its 4th Generation Xeon Scalable Sapphire Rapids processor. The new CPUs will still be available this year, but it looks like the company needs some additional time to validate the platform with different designs, components, vendors, and clients. 

"With innovation comes a level of complexity, and we see all the customers, OEMs, ODMs, you know the hyperscalers strong demand in that platform validation with us and we delivered the initial SKUs in Q1 that we had planned," said Sandra Rivera, General Manager of Intel's Datacenter and AI Group, at the Bank of America Securities Global Technology Brokers Conference (transcript). "At this point we are building in more platform and product validation time, so we see Sapphire Rapids […] ramp being later in the year than what we had originally forecasted, but the demand is still very high." 

Intel's 4th Generation Xeon Scalable Sapphire Rapids processor was originally meant to ship in 2021, but Intel delayed it to early 2022 — eventually moving high-volume availability of the processor to mid-2022. After shipping the initial production release qualification (PRQ) Sapphire Rapids processor models to select customers in Q1, the company moved high-volume availability again, to the second half of this year (which could have been interpreted as July or August). Apparently they're now delayed further, which probably means fall given that Intel's volume ramp will continue into 2023. 

"Intel's [3rd Generation Xeon Scalable] Ice Lake for, certainly for 2022 will be the highest volume product as we ramp Sapphire later in this year and then of course throughout 2023," said Rivera. 

Select Intel customers will continue to get PRQ Sapphire Rapids processors in the coming months ahead of the high-volume launch. These clients — e.g., operators of hyperscale datacenters — tend to use a lot of CPUs, so Intel will still make quite some money selling its 4th gen Xeon Scalable platform in 2022.  

Rivera stressed that while Sapphire Rapids is delayed, its successor Emerald Rapids is still on track for late 2023 — Emerald Rapids will use the same socket as SPR, so its volume ramp should be smoother as Intel's customers will have experience with the platform. 

"We know that Emerald Rapids which follows Sapphire [in] the second half of next year, a nice performance bump in terms of the memory, the networking, and the overall performance."