We learned a lot about Intel’s upcoming plans for desktop CPUs at Computex 2026. In classic Intel fashion, we’ve already heard a lot about the company’s next-gen CPUs, codenamed Nova Lake, even while the recent Arrow Lake Refresh CPUs are still warm from the oven. But on the ground in Taipei, we heard not only more about Nova Lake and the Z990 platform it’s arriving on, but also how Intel intends to handle the rollout and how it will fill the gaps in its lineup with “Raptor Lake Next,” which is supposedly slated to launch next year.

Trade shows are the best opportunity to learn details about unreleased products before they show up in a press deck, and simultaneously the worst venue to do so. With jet-lagged representatives and reporters, thousands of people whizzing past, and the threat of Jensen Huang showing up to sign components and shut down a floor on a moment’s notice, it’s easy for things to get lost in the shuffle. So, we’re going to work through everything we learned about Intel’s upcoming plans in stages, starting with details that are confirmed, and working toward more speculative murmurs.

Intel has a fairly aggressive consumer roadmap, which the company itself would tell you – and the company told us as much at Computex, as a matter of fact. Both Nish Neelalojanan, senior director of client product management, and the recently joined Alex Katouzian, executive VP and GM of client, played up Intel’s roadmap to Tom’s Hardware, and for good reason.

Chronologically, Intel’s plans look something like this: We’ll see the first Nova Lake SKUs roll out at CES 2027. A few months later, we’ll see a refresh on the LGA 1700 socket with “Raptor Lake Next” CPUs, and come Computex next year, Intel will launch a 52-core flagship Nova Lake SKU. None of that is confirmed by Intel, and we have varying degrees of confidence in each step of the roadmap, so take it as speculation for now. We’ll dig more into the details we have and what’s simply rumored below.

What about AMD?

Before Intel, we should at least look at why we’re not talking about AMD’s next-gen desktop plans. Basically, we don’t have a ton of information on Zen 6 CPUs yet, and even less information about Olympic Ridge, the desktop consumer lineup of Zen 6 chips. Computex didn’t change that fact.

At Computex, AMD revealed the Ryzen 7 7700X3D, relaunched the Ryzen 7 5800X3D, and brought the RX 9070 GRE to the rest of the world. Unlike previous years, AMD didn’t hold a keynote, where we might’ve seen a more concrete tease of Olympic Ridge; AMD has already teased Zen 6 broadly several times. Bigger Zen 6 news is likely at the company’s Advancing AI event next month.

Although AMD hasn’t said when Olympic Ridge will launch, we originally expected it in late 2026. Now, 2027 is very likely. AMD has shifted the Zen 6 conversation toward its EPYC Venice chips, and confirmed production ramp on Venice in May. Although AMD traditionally leads with a consumer launch at the turn of a new microarchitecture, it’s unlikely that Olympic Ridge will launch before Venice. Demand for CPUs is spiking in the data center for agentic AI workloads, after all, and AMD is adjusting accordingly.

Olympic Ridge probably isn’t top of mind right now, from both AMD itself and its partners. AMD laid the groundwork for a unified CPU architecture generations back, and Intel’s approach has been a bit more disparate across client and data center (although that’s been changing with releases like Xeon 6 and Xeon 6+). We don’t know when Olympic Ridge news will arrive, but it almost certainly follows far greater detail about Zen 6 in the context of Venice.

What’s confirmed

Let’s start with the concrete details about Intel’s future CPU plans. These are things we have direct evidence for, be it photos, our own hands-on time, or sources we’re extremely confident in. At least two Z990 motherboards were at Computex, a third is rumored, and we saw (and held) what looked to be a near-production model in a closed-door meeting. And from that, we can already tell a lot about Nova Lake.

First, the LGA 1954 socket, which has now been pictured (we were told not to take pictures, but someone else did the dirty work, it seems). It’s the same size as the LGA 1851 socket, measuring 45 mm x 37.5 mm, and it retains compatibility with existing coolers, which we were able to confirm at Computex. It features more pins, as the name reveals, and uses the 2L-ILM, or two-lever Independent Loading Mechanism. The picture of the socket circulating matches what we saw at Computex.

The motherboard we saw featured dual 8-pin EPS connectors, along with an 8-pin PCIe connector near the bottom of the board, which is said to provide auxiliary power to the CPU. We’ve seen a leaked photo of the Z990 PCH now, which is said to draw more power due to broader PCIe 5.0 support. The Z990 board we saw, at least, had three PCIe 5.0 M.2 slots, along with three PCIe 5.0 expansion slots. Short of perhaps specialized designs with extra M.2 slots, we expect Z990 to support PCIe 5.0 across the board.

As for the chips themselves, all that is confirmed from Z990 motherboards is that Nova Lake can scale up to a high-end power design. We’ll speculate more on specific numbers later, but we’ve seen auxiliary power beyond two 8-pin EPS connectors on two Z990 motherboards now, and the motherboard we held had an extremely high-end VRM design; we can’t say more than that at this point.

An important caveat here is that we’re dealing with high-end motherboards and discussing how high the platform can scale, not how it will scale. Plenty of ink has been spilled about Nova Lake’s supposedly high power draw, but we really don’t have details about the chips themselves, rather just the tippy-top of the platform that will support them.

Outside of Z990 boards, Intel has confirmed that Nova Lake is “coming at the end of 2026.” That’s what CEO Lip-Bu Tan said at the company’s full-year 2025 earnings call back in January. What we were told by multiple vendors at Computex is Q1 2027, with a portion of those vendors specifically pointing to CES 2027. Similarly, with Z990 motherboards, some vendors said Q1 2027 while others said Q4 2026 (one even hinted at Q3). Believe it or not, these timelines actually all match up.

What’s lost in translation here is when the sale is happening. Before Nova Lake launches publicly, Intel and motherboard vendors will need to sell products into the channel, which, a few months later, will be available for sale at retailers for you to buy. What we’re likely looking at is sales into the channel in Q4, a public launch of Nova Lake at CES 2027, and retail sales in Q1. When Tan says Nova Lake is coming at the end of 2026 to a group of investors, he’s likely referring to selling into the channel, not the final retail sale.

What’s likely

Now, we’re getting into a bit more speculation. These are some of the details we heard about at Computex, or confirmations of previous rumors that we don’t have any concrete evidence for. Given the conversations we had at Computex, and a healthy dose of critical thinking, these are the details that are likely but not confirmed. There’s always a chance we’re just blind men touching an elephant on some of these points.

First, Nova Lake. For nearly a year now, it’s been rumored that the highest-end Nova Lake SKU will scale up to 52 cores. That’s the number we heard at Computex, as well, but not as a typical flagship. Rather, we heard that Intel plans to lead Nova Lake with a 28-core flagship, which will launch at CES 2027, and introduce a high-end 52-core model later in the year. The timeframe we heard was Computex 2027, but if anything is subject to change, it’s a release date that’s a year away. For now, let’s call it later in 2027.

The 52-core SKU will apparently come with 16 Coyote Cove P-cores, 32 Arctic Wolf E-cores, and a cluster of 4 LP-E cores; we didn’t hear that at Computex, nor anything to the contrary, but that’s what has been previously rumored. That model will reportedly come with two compute tiles, so the 28-core model with a single compute tile will likely look like an 8 + 16 + 4 split. That’s pure extrapolation at this point, however.

As for the 52-core model, we were told it comes with a PL1 of 175W and a PL4 of up to 700W. The PL1 number is what’s important here. Although that is a sizable increase over the 125W PL1 of both the 285K and 14900K, 52-core Nova Lake doesn’t sound like a direct replacement for those parts. Given the timing and extra power demands, it looks more like a spiritual successor to Intel Extreme Edition chips, targeting enthusiasts with deep pockets and the HEDT crowd.

Nova Lake is treaded ground at this point, however. Something new we learned about from Computex is “Raptor Lake Next.” After hearing the name, we asked Intel, which declined to comment on Raptor Lake Next at this time. Apparently, however, it will be the third refresh of Raptor Lake CPUs on the LGA 1700 socket, particularly targeting budget-conscious builders while Nova Lake satiates the enthusiast crowd.

There are some pieces of circumstantial evidence that point to a reintroduction of LGA 1700 CPUs. First, this has been previously rumored. In April, prolific leaker Jaykihn hinted at another Raptor Lake refresh coming in 2027. We’ve now heard that the range is called Raptor Lake Next from multiple sources, and it’s specifically coming in the first half of 2027, some months after the initial Nova Lake launch.

Additionally, multiple motherboard vendors told us that they’re ramping production of LGA 1700 motherboards, including DDR4 boards, though they didn’t say it was in relation to any new CPU releases. Intel itself has dropped a few hints, as well. Earlier in the year, Intel’s Robert Hallock said that Raptor Lake will be “abundantly available” in the market, and at Computex, Intel’s Nish Neelalojanan told Tom’s Hardware that Intel “will continue to make sure that there are products which can take care of older memory technologies.”

It would certainly make sense for Intel to refresh Raptor Lake a third time. Although data center demand is offsetting it, the decline in desktop sales from high memory prices hits Intel and AMD on the balance sheet as well. Just about everyone we spoke with at Computex talked about the state of memory prices, and Intel has a DDR4 platform that it’s still actively selling on the market. AMD, with a hard switch to DDR5 with Zen 4, has to reach back further to revitalize DDR4 options, but Intel already has a small ecosystem of DDR4 motherboards and CPUs available now, which it could easily bolster. We’ve heard that bolster is coming in the opening months of next year.

What that range looks like remains a mystery, however. It could be a proper refresh, or it could simply be an infusion of 14th-gen stock (and LGA 1700 motherboards) into the market along with new price points; both Raptor Lake generations have slowly crept up in price since the end of last year. The important thing here is that it seems Intel is targeting LGA 1700 for the lower end of the market, as Arrow Lake, with its underperformance and high price due to exclusively using DDR5, won’t provide the last-gen value bridge that previous generations have.

After Tom's Hardware originally broke the news about Raptor Lake Next, we followed up with Jaykihn, who provided a few specs.

*Naming unconfirmed by Intel, specifications rumored

The specs we've heard about are for the four SKUs above, which would comprise the main lineup of chips with integrated graphics enabled; apparently, Raptor Lake Next will include options with the iGPU disabled, as well as mobile chips. The final branding is unconfirmed, but we've heard that Intel intends to launch under the Core Ultra 200 name.

Out of the four SKUs, the 16-core Core 5 looks like Intel's breadwinner. Throughout 12th- to 14th-Gen, Intel topped out Core i5 models at 6 P-cores. You'd have to step up to a Core i7 for 8 P-cores. If these specs are correct, Intel is stepping down to an 8 P-core configuration a tier in branding, which will hopefully come with a cut to price.

What’s still up in the air

Some of the finer details of Nova Lake are still up in the air. That is, we don’t have any direct evidence for them, nor any corroboration from Computex. That’s not to say that the details here are false. Rather, we just need more information to say, for sure, that some of these details are a part of the Nova Lake lineup.

First and most obvious is bLLC, or big Last Level Cache. This is one of the earliest Nova Lake rumors that is still circulating, and for good reason. Intel hasn’t found an effective counter to AMD’s 3D V-Cache CPUs in more than four years. We’re closing in on half a decade where AMD has entirely owned the high-end of PC gaming, which has continually eaten away at Intel’s market share. bLCC is, apparently, Intel’s counter to 3D V-Cache, using its own Foveros 3D hybrid bonding to stack additional last-level cache.

Tom’s Hardware asked Intel CEO Lip-Bu Tan and a panel of executives at the company how it plans to address X3D CPUs, and Alex Katouzian, a 20-year Qualcomm veteran who recently joined Intel in a leadership role over the client group, said the following: “When I first came in and started reviewing road maps for the team, I was very pleasantly surprised. So, stay tuned, a very strong roadmap [is] coming, and we will be gunning for that section of the market as well. And so, please stay tuned.”

Context is important, but Katouzian is really only saying that Intel is gunning for high-end gamers with its roadmap, which, of course, it is. Otherwise, bLLC has entirely been a topic of the rumor mill. Intel has indirectly teased it with PR hits about its packaging capabilities, but that extends far beyond bLLC. Hybrid bonding, especially from a foundry perspective, has far greater legs in the data center.

Although Intel has the packaging and bonding capabilities, the scale of them for a mass-market product like Nova Lake is questionable. Intel would need to bond the SRAM to the logic tile with Forveros and package the chip with EMIB, creating the “EMIB 3.5D” combination that Intel has talked about previously. We first saw EMIB 3.5D on the Ponte Vecchio data center GPU, but most recently and relevantly on Clearwater Forest, Intel’s first foray into putting 18A in the data center. The capability is there, but if Intel can scale that up to a consumer range with more limited die space and higher per-core performance remains to be seen.

One advantage of Intel’s hybrid bonding and advanced packaging is that it can package dies from other foundries, not just those from Intel foundries. That brings us to the second finer point about Nova Lake, which is the node. Originally, the assumption was that Intel would use 18A for Nova Lake. We have 18A on mobile with Panther Lake, in the data center with Xeon 6+, but not on the desktop. Further, Intel has previously commented about reshoring its manufacturing for consumer chips after a brief stint with TSMC for logic tiles in both Lunar Lake and Arrow Lake.

Around this point last year, however, rumors started circulating that Intel is using TSMC’s N2 for Nova Lake. The source of the rumor is flimsy, however. Well-known reporter Charlie Demerjian of SemiAccurate reported in July 2025 that Intel taped out a major product. The report didn’t mention what product, what foundry, or even include “TSMC” anywhere on the page. Still, other outlets took the story, claiming that not only was Demerjian talking about Nova Lake, but also that he was talking about TSMC N2.

There are reasons Intel could use TSMC for the logic die. The company has reiterated that it’s shifting wafer capacity toward the data center, so if TSMC can fill additional capacity on the desktop, we could see TSMC on the main logic die. It’s also possible that TSMC is manufacturing other tiles on Nova Lake. Intel has consistently blended nodes in recent generations, so even if Intel were to confirm that it’s tapping TSMC for Nova Lake, that doesn’t necessarily mean the Taiwanese giant is manufacturing logic.

And, just as easily, Intel could absolutely be using TSMC for logic. That’s the point here; we really don’t know at this point, outside of vague reporting, getting swept up in the rumor mill, and taking on a life of its own. The Cinderella story for Intel would be Nova Lake on 18A, but given the struggles on 18A yields, it wouldn’t be surprising to see TSMC at the helm for Nova Lake once again.

Hurry up and wait

Intel needs a much more aggressive roadmap on the desktop than AMD, frankly, and that roadmap is starting to take shape. Although AMD and Intel compete on the finer points of performance, Team Red has almost exclusively taken market share away from Intel, quarter over quarter, for the past decade. There are only a handful of quarters in that time when AMD has lost market share, which it has always rebounded from in the quarter that follows.

Even if Intel still represents the majority of the desktop market — and it does based on the latest market research — the trend is abundantly clear. Add on top of that clear fumbles like Arrow Lake, and it’s obvious that AMD doesn’t need to move the needle much to continue swiping customers. Intel needs to make big moves to recover.

We should have more official details about those plans soon. Intel mostly sat Computex out on the consumer front, short of the Arc G3 range that, although exciting for gaming handhelds, is destined to be a niche product given the high prices of the devices those chips are going in.

For the past four years, Intel has held its Tech Tour event in the fall, taking the place of its previous Architecture Day, which took place in the late summer (most of those details have shifted to the Hot Chips conference in August). Intel has already told us that Hot Chips will have more details about Diamond Rapids, Intel’s next-gen P-core Xeons. That leaves Tech Tour for when we’ll likely get a full architectural deep dive on Nova Lake. Intel has yet to confirm Tech Tour 2026, but we have no reason to believe the company will sit out the rest of the year at this point. It also lines up with what we’re hearing about Nova Lake’s release — architectural details in the fall, a launch at CES 2027, and availability in Q1.

Regardless of when the exact dates fall, Computex made it clear that Intel is readying Nova Lake for a release soon. Multiple motherboard vendors brought Z990 motherboards to Computex and actively showed them to the press; I can’t imagine that was sanctioned by Intel.

As for Raptor Lake Next, Computex is the first quasi-confirmation we’ve heard of the range. That name apparently appears on Intel’s roadmap at some point in the first half of next year. With Nova Lake at the high-end and Raptor Lake Next in the midrange, Intel might have a one-two punch strategy to earn back some spots in the market, especially as AMD turns its Zen 6 focus toward the data center and prioritizes older architectures on desktop, given high DDR5 prices. Now, we just need to wait and see how those internal plans materialize as the rest of the year goes on.

Yesterday, we broke news of a third Raptor Lake refresh in the works from Intel, dubbed "Raptor Lake Next." Following that, new leaks have emerged from reliable tipster Jaykihn confirming the codename and detailing the upcoming family. Raptor Lake Next is slated for early 2027 and will land on the LGA 1700 socket as reported previously, but some outlets say it will also retain the existing Core 200 branding.

According to the rumors, Raptor Lake Next doesn't use Bartlett Lake silicon, which is Intel's special lineup for edge and embedded devices with only P-cores. That means a 12 P-core SKU has been ruled out for now, and that Raptor Lake Next will stick to the usual hybrid config. It's expected to scale across mobile and desktop with up to 125W parts and HX models as well, but it won't bring any new notable features.

Essentially, we're looking at rebranded 14th Gen CPUs that are likely being re-released because of their high yields. Intel spent a considerable amount of time and effort fixing the stability issues with 13th and 14th Gen families, so it makes sense the company wants to maximize its returns here. The Core-i9 14900K is still the most performant gaming CPU Intel has ever released, so calling this leftover silicon doesn't feel right either.

Jaykihn also provided further specs for the Raptor Lake Next, claiming it will only spread across Core 3, Core 5, and Core 7; no Core 9 SKUs in sight at the moment. At the top sits a Core 7 part with 8 P-cores and 12 E-cores, totaling out to 20 cores rated at 65W, same as the Core i7-14700. Then there's a 16-core 125W processor, which is identical to the Core i7-13700K, but it's counted as "Core 5" for Raptor Lake Next.

There's a special 10-core 65W part with 24MB of L3 cache because the previous 10-core silicon only had 20MB. It looks like the cache from disabled core clusters will still be accessible in certain Raptor Lake Next SKUs, and this might become a trend going forward. Lastly, we have a Core 3 CPU with 4 P-cores and no E-cores, running at 65W as well. Since all these chips are based on Raptor Lake silicon, they use Raptor Cove P-cores and Gracemont E-cores.

Keep in mind that RTL and RTL-R both came with Intel HD 700 series integrated graphics; the new Arc Xe architecture debuted with Arrow Lake on desktop. Nonetheless, Raptor Lake Next will work with DDR4 memory because it uses the LGA 1700 platform. That could help alleviate at least some of the burden of building a PC today as DDR5 RAM doesn't seem to be getting cheaper anytime soon.

Raptor Lake Next is apparently going into production in January 2027 so it could be available in Q1 2027, coexisting with Nova Lake as perhaps the value-focused option from the Blue Team. All rumors point toward Nova Lake being delayed to next year, but officially we still expect an announcement later this year. Regardless, the two families sure look like they'll overlap, which is why the Core 200 vs Core 400 distinction would be important for consumers.

Mozilla has successfully addressed a critical bug related to Firefox that caused the web browser to crash on desktop systems powered by Intel’s Raptor Lake CPUs. With the latest stable release of Firefox version 151.01, the company has managed to patch the issue that has been under investigation for more than a year.

Mozilla engineers initially zeroed in on failures in a zlib-rs compression routine where certain dist values appeared incorrect, resulting in index out-of-bounds crashes. However, the root cause was tied to Intel’s Raptor Lake CPU instructions, specifically RPL050 and RPL060, which sometimes caused the CPU cores to read incorrect or outdated data.

Senior Staff Engineer Gabriele Svelto first flagged the issue last year, blaming Intel for its CPU instabilities and highlighting mass browser crash reports coming from systems powered by Intel Raptor Lake, specifically in locations suffering from heat waves.

“If you have an Intel Raptor Lake system and you’re in the northern hemisphere, chances are that your machine is crashing more often because of the summer heat. I know because I can literally see which EU countries have been affected by heat waves by looking at the locales of Firefox crash reports coming from Raptor Lake systems,” said Svelto on Mastodon.

He also noted that while Intel’s newer 0x12c microcode update significantly reduced the number of crashes, the bugs came back with the release of version 0x12F.

Intel’s Raptor Lake CPU instability issues first began surfacing in late 2022 before exploding the following year with users reporting widespread game crashes, browser instability, and system failures on 13th-gen and 14th-gen processors. Intel eventually confirmed after several months that the root cause was tied to a physical degradation issue caused by prolonged exposure to excessive voltage and heat. While the company rolled out several microcode patches, including 0x125, 0x129, 0x12B, and, more recently, 0x12F, these updates were only designed to mitigate the conditions triggering the degradation rather than reverse existing damage. Eventually, Intel announced an extended warranty for customers facing the issue from three to five years.

If you have been facing Firefox crashes on your desktop PC running Intel’s 13th-gen or 14th-gen Raptor Lake CPUs, it is recommended to update the browser to its latest stable version by heading to the official page here.

A few days ago, we covered the modding efforts of kryptonfly, a community member who managed to get a Bartlett Lake CPU running on an Asus ZZ90 motherboard. At the time, the Core 9 273QPE used for the experiment successfully booted but didn't go beyond the POST screen. Since then, several users have been hard at work over on the Overclock.net forums, and today, their efforts have finally paid off. Bartlett Lake can not only boot inside Windows now, but it's actually stable enough to run benchmarks.

Thanks to tipster HXL, we found out about one of the Cinebench R23 scores on the forum, but there are at least three more with varying degrees of performance. First up is CarSalesman, who achieved 33,111 points in the multi-core test while their Core 9 273QPE sucked up 286W of power, and boosted to 5.4 GHz across all cores. That score puts it just above the Ryzen 9 9900X3D on average, and below Intel's own Core i7-14700 — impressive for an OEM-only chip that was never meant to run like this.

CarSalesman also had a much less exciting Cinebench run before that achieved only around 25,000 points. The CPU was consuming upwards of 320W in that test because of unlocked power limits, and Vdroop was causing it to throttle, limiting the clocks to 4.3 GHz. By locking the Vcore to 1.35V and using LLC6 in BIOS, they were able to force voltage to stay high under load, improving stability and allowing higher sustained clocks, which resulted in more consistent (and slightly lower) power draw.

All of this was made possible by kryptonfly's efforts, who was able to patch the setup BIOS, injecting code that forced it to recognize the Core 9 273QPE. Since this is a unique 12-core CPU, featuring all 12 P-cores with Hyperthreading and no E-cores like other Raptor Lake chips, the Z790 BIOS wasn't directly compatible with it. Bartlett Lake CPUs feature only (Raptor Cove) P-cores and aren't designed for consumer setups.

Therefore, the modder applied a patch that essentially told the BIOS it should allow 12 physical P-cores. Previously, as soon as the BIOS tried to access the 9th P-core, it would exceed the hardcoded limit of 8 P-cores found in consumer firmware and just crash to a black screen. This is because standard 13th and 14th Gen CPUs only had up to 8 P-cores with Hyperthreading, even if they have a higher total core count due to E-cores without Hyperthreading.

Another user by the name of Talon2016 was able to score 32,288 points, which puts them in Core i9-13900K territory. More importantly, though, Talon said they were able to actually play Battlefield 6 with Secure Boot enabled in the BIOS, achieving a similar 5.4 GHz boost clock across all cores. The test config consisted of an Asus ROG Maximus Z790 Apex motherboard kitted with 64 GB of DDR5 RAM running at 5,600 MT/s. The same user scored 33,818 points in a future run posted on the forum as well.

The thread is alive with a lot of enthusiasts kicking in with their questions, testing, and gratitude, all contributing to this project's progress in some way. Getting Bartlett Lake running on consumer hardware is a significant milestone, considering these chips were never meant to be distributed outside edge and industrial OEMs. They share the same LGA 1700 socket as other 13/14th Gen chips, so they were always physically compatible — it was only a matter of making them electrically compatible, too.

A YouTuber and computer technician went on their channel to complain about Corsair’s pre-built, as his customer went back to his shop three times in the past year just to RMA the Intel Core i9-14900K on their $5,000 gaming PC. According to Matthew Wieland of Matt’s Computer Services, he is certain the processor on the gaming PC is suffering from instability after he ran several tests on both the motherboard and CPU. This should have been an easy fix, especially since Intel has released the 0x12F update that fixed the Vmin instability issue once and for all. Turns out, Corsair hasn't updated its custom BIOS to include the fix.

The Intel instability issue first exploded in early 2024, with the company blaming the custom BIOS installed by motherboard manufacturers that disables thermal and power protections. Eventually, the chip maker acknowledged that there was an issue with the processor, but it took Team Blue around five months from when the first reports surfaced for it to find the root cause and release a patch that fixed the problem. However, it wasn’t until May of this year that Intel released a final fix for the issue, which is deployed via a firmware update.

Unfortunately, six months after Intel dropped 0x12F, Corsair has yet to release a BIOS update with this fix for the motherboard in question. The gaming PC sports an Asus Prime Z790-P WIFI motherboard, and the latest BIOS on the Corsair website only has the 0x12B update. When Matthew downloaded the latest BIOS version from the Asus website and tried installing it on his customer’s gaming PC, it refused to recognize the file. He said that this is likely because most pre-built systems have a locked-in BIOS to prevent customers from tinkering under the hood. So, he needs a BIOS with the 0x12F fix directly from Corsair, but it’s not yet available.

Corsair’s customer service was pretty helpful, but their hands were tied because Corsair’s engineers have yet to release a new BIOS. Thankfully, Intel extended the warranty for all chips affected by the instability, allowing the tech to RMA the damaged CPU. However, going back three times in a year to get the issue fixed is mighty inconvenient, and unless Corsair comes up with a BIOS update by the time he gets the replacement processor, he has no choice but to recommend a motherboard replacement to avoid destroying the new CPU as well.

Demand for Intel's processors for client and data center applications was on the rise in the third quarter of 2025, but the company could not fully capitalize on it as it faced numerous supply shortages across its business, which included shortages of its own production capacity and an industry-wide shortage of substrates, which will persist into 2026. The company is prioritizing the supply of data center CPUs, but going forward, it also plans to adjust the pricing of its products, which means increased prices of previous-generation client CPUs.

"Capacity constraints, especially on Intel 10 and Intel 7, limited our ability to fully meet demand in Q3 for both data center and client products," said David Zinsner, chief financial officer of Intel, during the company's conference call with analysts and investors.

Raptor Lake in tight supply

While the Intel 7 (formerly 10nm Enhanced SuperFin) process technology introduced in 2021 – 2022 looks ancient, it still packs quite a punch in terms of performance capability and Intel uses it to produce a host of CPUs, including 13^th and 14^th Generation Core 'Raptor Lake' CPUs as well as I/O die for Xeon 6 'Granite Rapids' CPUs and 5^th Generation Xeon Scalable 'Emerald Rapids' processors that are still in demand. Recently, prices of Intel's Raptor Lake processors rose amid supply constraints up as demand for these CPUs is still high three years after the introduction.

As the company has no plans to expand capacities for previous-generation nodes, but foresees demand for Intel 7 and Intel 10-based products to remain strong in the coming quarters, it expects these CPUs to be in short supply well into 2026. Furthermore, to capitalize on it, Intel intends to adjust pricing and produce more high-end SKUs.

"Given the current tight capacity environment, which we expect [to] persist into 2026, we are working closely with customers to maximize our available output, including adjusting pricing and mix," said Zinsner.

As for Intel 10 (formerly 10nm SuperFin), it is hard to tell which of Intel's broadly available CPUs use it, but Intel probably ships a boatload of long-life-cycle products made using this technology under long-term supply contracts.

Diverting wafers to data center CPUs

As Intel's Xeon 6 'Granite Rapids' uses an I/O die made on Intel 7 process technology, insufficient capacity hits the company's ability to ship enough expensive data center processors. On the one hand, this might be good news for Intel as demand for its server CPUs is back and it can sell them without a discount or even at a higher price. However, if it does not have enough I/O dies, it cannot ship processors at all. Therefore, it has to divert capacity to data center CPUs, thus cutting the supply of client CPUs, which is one of the reasons why Raptor Lake processors got more expensive recently and will likely gain price in the coming months.

"[In Q4] we expect [client computing group] to be down modestly and [data center and AI group] [to be] up strongly sequentially as we prioritize wafer capacity for server shipments over entry-level client parts," said Zinsner.

Intel's decision to sacrifice some lower-end client CPU volume to keep server CPU customers supplied — particularly hyperscalers and AI infrastructure buyers — has a great rationale as data center CPUs are sold at thousands of dollars a unit, whereas even the highest-end client CPU is hardly sold at $500 - $600. Given the cost of semiconductor fabs and limited production capacity at Intel's 7-capable lines, Intel's management is forced to triage output toward the most profitable products.

Shortages are here to last

In addition to insufficient Intel 7-capable capacity, Intel also blames the shortage of substrates used for CPU packages for the tight supply of its processors, which further complicates the company's supply chain.

" There is also shortages even beyond our specific challenges on the foundry side," said Intel CFO. "I think there's widely reported substrate shortages, for example. So, I think the demand, you know, there is a lot of caution coming into the year, I think across the board."

In general, Intel warns that shortages of its processors will persist through 2026. On the client side, this is conditioned by the slow 18A ramp and decent demand for Raptor Lake processors, whereas on the data center side, it will be driven by insufficient Intel 7 capacity amid the continuous ramp of Xeon 6-series products. The shortages are expected to worsen in Q1, but then there may be some relief.

 "We may actually be at our peak in terms of shortages in the first quarter because we have lived through the Q3 and Q4 with a little bit of inventory to help us and just cranking the output as much as we could with the factory," said Zinsner.

A new report claims that Intel is set to increase the price of its older Raptor Lake chips by as much as 10% in the face of the continued popularity of the lineup and customers shunning AI-equipped Lunar Lake models, according to Digitimes. While the report doesn't specify which models are set for an increase, it specifically highlights that chips launched in October 2022 (13th-gen chips), noting that prices will increase from $150-160 to $170-$180. This could encompass some popular i3 or i5 models, depending on the region and availability.

This increase will occur despite the company announcing that the integrated GPU drivers for these processors will be placed on legacy software support, effectively putting them on the back burner, despite some having launched as recently as 2023. Digitimes cites industry insiders who say "lukewarm" consumer response to AI PCs as a factor.

Intel's current crop of processors hasn't performed well in the market, with CEO Lip-Bu Tan commenting frankly that the chips aren't competitive in the market. That, coupled with customers seeking value options, has led many to opt for lower-cost previous-generation models instead.

The Raptor-Lake price jump aligns with the soaring costs of DDR4 memory chips, meaning PC makers will be hard-pressed to keep their prices low by opting to use the older memory in lieu of newer, faster DDR5, thereby squeezing the budget end of the PC building market. According to the report, some brands are even experiencing supply shortages, especially as the volatility of chip pricing is making them reluctant to place long-term orders. Due to this, PC and component prices may increase slightly as we approach the end of 2025.

Windows 10’s end-of-life this coming October might result in some corporate sales, especially for companies unwilling to pay for extended support. However, this is less of a concern for consumers, meaning it’s unlikely that retail sales will increase as a result. The upcoming holiday season may drive some sales figures, especially if companies release great deals.

We've pinged Intel for comment and will update as neccesary.

Mozilla has reportedly been swamped with crash reports for its Firefox browser, the majority of which are coming from Raptor Lake Intel CPUs that are known to suffer from instability that is further exacerbated by heat. Senior Staff Engineer Gabriele Svelto said on Mastodon that a mass of browser crash reports are coming from Intel Raptor Lake-powered systems located in areas that are suffering from heat waves.

“If you have an Intel Raptor Lake system and you’re in the northern hemisphere, chances are that your machine is crashing more often because of the summer heat,” says Svelto. “I know because I can literally see which EU countries have been affected by heat waves by looking at the locales of Firefox crash reports coming from Raptor Lake systems.” The engineer even added that it has gotten so bad that the team disabled the bot, which filed these crash reports automatically, especially since these events almost exclusively happened to Intel Raptor Lake PCs — specifically, the Intel Core i7-14700K model.

The instability issue exploded around the second and third quarters of last year, and it took several months for the company to find its root cause. Since this was a physical degradation problem, no amount of patches can reverse the instability — Intel’s microcode updates only mitigated it and prevent the conditions that triggered the instability from occurring. Just last month, Intel released microcode update 0x12F to address the Vmin shift that’s happening to Raptor Lake CPUs that have been running for several days in a row. However, Svelto says that this version also caused the bugs to “come back in full force”.

Thankfully, Intel extended the warranty for all the affected chips from three to five years. So, if you’re one of the millions of users affected by this issue, and you have a dead Raptor Lake CPU, you can RMA it and get a replacement.

These CPU instability issues have caused Intel to lose ground to AMD in the desktop PC market share, and its Arrow Lake chips performed poorly in sales, exacerbating the company’s financial problems. In comparison, AMD has been releasing excellent gaming processors, especially the AMD Ryzen 7 9800X3D chip, which eclipses even Intel’s latest offerings in several gaming benchmarks.

Intel demoed working Panther Lake Core Ultra 300 silicon for laptops, its first chips based on its crucial 18A process node, here at Computex 2025 in Taipei, Taiwan. Unlike the first public demoes at CES 2025 that merely showed the chips powered on, Intel put Panther Lake its paces in real-time rendering and AI applications, showing that the silicon is healthy and on-track for retail availability in early 2026. Intel also shared more information about its performance and power consumtpion expectations for the new chips. 

As you can see in the image above, Intel also had a Panther Lake chip on display, enabling us to see how the CPU, GPU, I/O tile, and SoC tile are arranged on the package. These chips are thought to come with Cougar Cove P-cores and Darkmont E-cores (you can see the unconfirmed leaked specifications of some of the chip models here). 

Intel says the Panther Lake chips blend the power efficiency of Lunar Lake with the performance of Arrow Lake-H, noting that while the chips will be in production in the second half of 2025, presumably launched at CES, full retail availability will not come until early 2026. Intel did tease that the chips will come with the next-gen integrated graphics with XMX graphics, but aside from saying the iGPU performance will be closer to Lunar Lake than Arrow Lake, the company didn't elaborate. These iGPUs are thought to be based on the Xe3 architecture.

Intel ran its Panther Lake benchmarks on two Reference Validation Platforms (RVP) that you can see in the above album. These platforms are used to validate the design and emulate real-world conditions. Both RVPs were equipped with a heatsink and fan, so they were presumably operating without thermal constraints. 

Intel demoed one system running the newly-resurrected Clippy as a large language model to demonstrate that the chips are running AI workloads. The presenter used the system to write game code in Python code. Intel didn't share performance metrics from the benchmark. 

Intel also demoed a system running Da Vinci to edit and manipulate video using local AI processing to process the video, enabling fast manipulation of the video clip, such as changing backgrounds, clothing colors, and adding flying text to the clip. 

Intel also displayed a running developer kit that 300+ developers with ISVs are using to enable software support for the coming chips. Intel demoed the system being used for image editing with auto-coloring and upscaling features, powered, of course, by AI. As you can see in the album, the developer system is quite compact. Intel also had a host of laptops on display from its OEM partners. 

Intel's Panther Lake appears to be on track for its launch schedule, which bodes well for the company's immensely important 18A process node. Intel teased that the next steps are to release concrete speeds and feeds along with more information about the various chip models. We expect those to come trickling our over the next several months. 

Most of my cooling reviews are designed to let you know what the best and most powerful cooling solutions are. But not everyone is looking to overclock the hottest Intel i9 CPU. Instead of focusing on the best possible performance, today’s review will be looking at Endorfy’s Fortis 5 air cooler, which won’t win any overclocking competitions, but runs quietly while delivering enough cooling performance to satisfy the needs of mainstream consumers and gamers.

Let’s take a look at the specifications and features of the cooler, then we’ll go over thermal performance so you’ll know how cool it will keep your computer’s processor. We’ll also look at noise measurements to let you know how silently this product operates.

Cooler specifications

Packing and included contents

The exterior of the box isn’t flashy, but it showcases the features of the product well.

Included with the box are the following:

• 140mm fan
• Thick single-tower heatsink
• Thermal paste
• Mounting accessories for modern AMD & Intel platforms

Features of Endorfy Fortis 5 ARGB

▶ Rigid backplate

The included backplate is extremely strong, and won’t bend even under pressure.

▶ AM Clearance

There’s no reason to worry about RAM clearances with Endorfy’s Fortis 5 ARGB. As you can see in the picture below, the fan doesn’t interfere or overhang DIMM slots, allowing full compatibility no matter how tall (or short) your RAM is.

▶ Six Direct Touch Heatpipes

It is generally rare to see coolers with more than four or five direct-touch copper heatpipes, but Endorfy’s Fortis 5 has six of them! In theory, direct touch heatpipes should allow for more efficient heat transfer. Is that really the case? Well, we’ll get to the benchmarks soon enough and you can decide for yourself based on our tests.

▶ Thick single tower heatsink

Most air coolers incorporate one or two smaller heatsink towers, but Endorfy’s Fortis 5 incorporates one larger, thick heatsink.

The backside of the cooler incorporates a canyon in the middle. While Endorfy didn’t specifically comment on this design, other companies with similar designs have claimed it helps increase to static pressure of airflow.

▶ 40mm Fluctus ARGB fan

There’s more to a cooler than just the heatsink or radiator. The bundled fans have a significant impact on cooling and noise levels, as well as how the cooler looks in your case. The included Fluctus ARGB fan has 9 blades, with teeth on the edges of each fan blade. I’m not sure if those teeth serve any real purpose, but they sure look menacing!

Many cheaper products include fans with MTTF ratings of only 40-60,000 hours of operation, but not Endorfy. They indicate confidence in the quality of their product with a rating of 100,000+ hours for the Fluctus 140 ARGB fan.

Real world testing configuration: Intel LGA1700 and AMD AM5 platform

My results may differ from others because I place an emphasis on having results that are comparable to real-world use.

This means that I test CPU coolers inside of a closed desktop case, which increases cooling difficulty compared to other testing methods. Open benches have lower ambient temperatures, which in turn makes weak coolers appear stronger than they really are. Some publications have also used generic thermal plates to test cooling solutions.

I reject both of these methods because they don’t accurately reflect the real-world conditions where a CPU cooler is used.

My previous reviews have also tested Intel’s latest platform, using the Core Ultra 9 285K Arrow Lake CPU. We’re going to be retiring this from our testing suite. Between BIOS updates and Windows updates, Arrow Lake’s thermal characteristics have changed in some scenarios, rendering much of our previous testing data useless.

With today’s review, we’re also testing AMD’s recently released Ryzen 9 9950X3D. This is a beast of a CPU, providing the best gaming and some of the best multithreaded performance on the market, and can prove quite challenging thermally when PBO is enabled for overclocking.

LGA 1700 and AMD AM5 Installation

The installation of this cooler is simple and should be easy to complete for most users. It’s also pretty much the same process for Intel and AMD systems.

1. You’ll first need to place the backplate against the rear of the motherboard on an Intel system, using the standoffs to secure it. AMD users will need to remove the default mounting mechanism and replace it with the included standoffs.

2. Next, place the mounting bracket on top of the standoffs and secure it with the included screws.

3. Apply the included thermal paste to your CPU. If you have any questions on how to do this properly, please refer to our handy guide on how to apply thermal paste.

4. Place the CPU block on top of the CPU and secure it with a screwdriver.

5. Attach the fan and then connect the PWM cable to your motherboard, and installation is complete.

CPU-only thermal results without power limits

Because the CPU reaches its peak temperature (TJ Max) with air coolers like the one we’re looking at with this review, I’ve measured the CPU package power to determine the maximum wattage cooled to best compare their performance. This is why you won’t see results from the best AIO coolers in this chart.

It’s important to note that thermal performance can scale differently depending on the CPU it’s being tested with. First, we’ll look at results with Intel’s i7-14700K.

Looking at the results with Intel’s Raptor Lake CPU, you might think this cooler isn’t very impressive. It’s the worst-performing result I have in this chart, outside of a compact SFF cooler. Let’s take a look at how it handles AMD’s Ryzen 9 9950X3D now.

This is a much more impressive result. With 219W cooled on average, Endorfy’s Fortis 5 is able to handle the full “stock” power of AMD’s flagship CPU (~200W) and has a little bit of room to spare to let PBO push the CPU beyond its typical performance.

Noise levels are where this cooler shines. While thermal performance on Intel’s hottest CPUs isn’t the greatest, this cooler isn’t designed to win overclocking competitions. It’s designed to provide quietly running, essential cooling performance. With a maximum noise level of 37.3 dBA, this cooler is the quietest product I’ve tested on Intel’s Z790 Platform.

CPU-only noise normalized thermal results

Finding the right balance between fan noise levels and cooling performance is important. While running fans at full speed can improve cooling capacity to some extent, the benefits are limited and many users prefer a quieter system.

For noise-normalized testing with Intel CPUs, I typically set the noise levels of the cooler to reach 38.9 dBA. But this presents a problem: Endorfy’s Fortis 5 doesn’t reach that level of noise! Because this issue prevents a proper comparison to the other results I’ve collected, I’ll be skipping noise-normalized tests on Intel’s Core i7-14700K.

Let’s take a look at my initial noise-normalized results using AMD’s flagship Ryzen 9 9950X3D instead. This chart is still lacking the amount of comparison results I’d like, as I’ve just started sting with this CPU. That said, let’s take a look at what I have recorded thus far.

This test was run a little differently than I have done with previous tests on Intel CPUs. I’ve lowered the noise threshold of the CPU cooler down to 37.3 dBA, as some readers have indicated they consider 38.9 dBA “too loud” for this type of test. The other change is that I’ve measured the thermal performance of the cooler while also running Furmark 2 on MSI’s RTX 4070 TI Super, increasing the heat inside the case and increasing cooling difficulty.

The results here are encouraging. With 195W cooled, that’s almost the full power budget of AMD’s Ryzen 9 9950X3D at “stock” settings while the system also handles the full heat of a GPU. It performed significantly better than Thermalright’s impressive Royal Knight 120 SE in the same scenario!

“Stock” 200W results

My recent reviews have focused more on tests with both the CPU and GPU being stressed, but many have indicated they would like to see more CPU-only tests. So let’s take a look at how Endorfy’s Fortis 5 handles the Ryzen 9950X3D with default power levels, which limits consumption to ~200W.

Going by temperature data alone, the results aren’t very impressive here. The CPU reaches a toasty 83.7 degrees. But that really doesn’t matter, because it keeps the CPU under its peak temperature and does so while running very quietly, at or below the lowest measurement my noise meter can measure, at 36.4 dBA. This means the true measurement of noise level might actually be lower.

150W CPU + 290W GPU results

Testing a CPU Cooler in isolation is great for synthetic benchmarks, but doesn’t tell the whole story of how it will perform. If your GPU is running a full load, that heat doesn’t just disappear – it makes it harder for your CPU cooler to do its job due to the increased heat within your computer case.

The CPU power limit of 150W was chosen based on the worst CPU power consumption I’ve seen reported in gaming with AMD’s Ryzen 9 9950X3D. In my personal experience, most games use much lower power, but this might be biased by the games I play.

Again, looking at thermal performance alone would give you the incorrect assumption that this cooler is not impressive. The important thing is that the CPU is operating well under its peak temperature, and the cooler is running whisper quiet – 36.4 dBA!

140W CPU results

The last round of test results we’ll look at in today’s review is a test with a 140W power limit imposed. This level of power is much easier for most CPU coolers to handle. But I’ve recorded lower power results for comparisons with SFF and air coolers for future reviews.

As with previous results, Endorfy’s Fortis 5 keeps the CPU within reasonable thermal values while running whisper quiet.

Conclusion

Endorfy’s Fortis 5 ARGB is a single tower air cooler with whisper-quiet noise levels, intended for users who value silent operation over maximum thermal performance. It’s powerful enough to handle AMD’s Ryzen 9 9950X3D at stock settings, handling 219W when power limits were removed. However, the performance on Intel CPUs was not as impressive. If you’re looking to compete in overclocking competitions or running a previous-generation Core i9, you might want to pass on this one. But for everyone else, this is the level of cooling I’d recommend for AMD’s Ryzen 9 9950X3D, so anything lesser should also run below thermal thresholds and extremely quiet with this cooler.

Nvidia and MediaTek's hotly anticipated AI PC chips may not see meaningful shipment volumes to the mass market until the second half of 2026, according to a new report. As reported by Digitimes, Nvidia and MediaTek are expected to jointly unveil their new 'N1' Arm chips for Windows PCs at Computex.

According to Digitimes, the joint chip will "likely debut under the Nvidia brand," with both N1X and N1 models planned, echoing previous reports. Digitimes says that both companies are well into production ramp-up, however, states "insiders believe meaningful shipment volumes won't emerge until the second half of 2026."

The chips, which we expect to be unveiled at Computex next week, will mark the second offspring of Nvidia and MediaTek's newly-minted partnership, inaugurated by the GB10 chip unveiled earlier this year.

As the report notes, early N1X benchmarks hint at performance that lags behind some Arm-based PC chips, noting "the results have raised industry concerns."

Perhaps more worryingly, Digitimes reiterates reports that there are "unresolved integration issues with endpoint devices." These manufacturing headaches have previously been reported elsewhere, and could explain the hefty lead time of 2H26 on these chips.

MSI has confirmed it will take the covers off a host of exciting new products at Computex 2025 later this month, including a brand new desktop AI supercomputer powered by Nvidia's DGX Spark platform.

The company confirmed in a press release that it will unveil its EdgeXpert MS-C931, a new desktop AI supercomputer built on the Nvidia DG Spark platform. The MS-C931 is powered by Nvidia's GB10 Grace Blackwell Superchip and is capable of 1,000 AI TOPS FP4 performance. The GB10 SoC features Nvidia Blackwell GPU architecture and fifth-generation Tensor Cores, as well as NVLink-C2C connection to Nvidia's Grace CPU, an Arm architecture core featuring 20 power-efficient chips.

It will also feature ConnectX 7 networking, 128GB unified memory, and LLM support, which Nvidia has previously promised can support up to 4TB of NVMe storage, and can run up to 200-billion-parameter LLMs, or 405-billion parameter models when running two linked chips.

Nvidia's DGX Spark platform promises compact and efficient performance, and comes pre-installed with Nvidia's AI software stack so that developers can run AI models from all the major players, including DeepSeek, Meta, and Google.

Alongside the MS-C931, MSI also says it will unveil a new lineup of Industrial Motherboards, as well as systems powered by Intel Twin Lake, Raptor Lake Refresh, Bartlett Lake, and Arrow Lake processor families.

Specifically, the company highlighted three new products. The MS-C926 is an ultra-slim fanless box PC with applications in smart retail and digital signage.

The Ms-927 is an ultra-compact box PC featuring Intel Core Ultra processors for high-performance edge computing.

Finally, the MS-CF20 is a new next-generation ATX motherboard featuring 16th Gen Intel Arrow Lake-S processors.

MSI says it will also have live demonstrations for solutions for smart retail and digital signage. There will also be a new fanless palm box for "space-constrained industrial environments, remote control management solutions utilizing SysLink, and edge AI innovations, including LLM and chatbot applications enabled by AI smartLink software.

While Jiushark isn’t well known in U.S. enthusiast and PC building circles, we’ve covered the company in the past, highlighting some of its unique products like the Jiushark M.2 Three, an M.2 SSD cooler that uses a dual heatpipe radiator and fan, much like a mini CPU cooler. We’ve also tested the JF13K Diamond low-profile CPU cooler.

Today, we’re looking at the new Jiushark JF800 Diamond dual-tower air cooler, available in white and black, with the company claiming cooling capabilities of up to 280W!

Will Jiushark’s latest make our list of best CPU coolers on the market? There’s no question that it deserves a shot because, spoiler, this is the strongest performing dual-tower cooler we’ve ever tested, on both Intel Raptor Lake and Arrow Lake CPUs. But it’s also pretty hard to come by in the U.S., unless you’re willing to order it directly from China.

Let’s take a look at the specifications and features of the cooler. Then we’ll go over thermal performance and noise levels.

Cooler specifications

Packing and included contents

Opening the compact box reveals the accessories, neatly packaged above the cooler.

Included in the box are the following:

• Two 120 mm fans
• Dual-tower heatsink
• Thermal paste
• Mounting accessories for modern AMD and Intel platforms
• Installation manual

Features of Jiushark’s JF800 Diamond

*️⃣ Six copper heatpipes

The Jiushark JF800 Diamond moves heat away from the CPU using six copper heatpipes – but as you’ll see in our benchmarks, it offers performance superior to similar air coolers we’ve tested.

*️⃣ Heatsink Design

The towers of the JF800 feature what appears to be a fairly basic fin design.

The tops of the heatsinks have a grille reminiscent of cattle guards, adding some nice aesthetics with the ARGB lighting shining through them.

*️⃣ Two 120mm fans with ARGB lighting

There’s more to a cooler than just the heatsink or radiator. The bundled fans significantly impact cooling and noise levels, as well as how the cooler looks in your case. The included fans here have nine blades, and there are models with or without ARGB lighting, according to your preference.

As you’ll see in the benchmarks below, the fans help enable superior thermal performance, but can be a little noisy.

Testing configuration – Intel LGA1700 and LGA1851 platform

There are many factors other than the CPU cooler that can influence your cooling performance, including the case you use and the fans installed in it. A system's motherboard can also influence this, especially if it suffers from bending, which results in poor cooler contact with the CPU.

To prevent bending from impacting our cooling results, we’ve installed Thermalright’s LGA 1700 contact frame into our testing rig. If your motherboard is affected by bending, your thermal results will be worse than those shown below.

Not all motherboards are affected equally by this issue. I tested Raptor Lake CPUs on two motherboards. And while one showed significant thermal improvements after installing Thermalright’s LGA1700 contact frame, the other motherboard showed no temperature diffrence! Check out our review of the contact frame for more information.

I’ve also tested this cooler with Intel’s latest platform, Arrow Lake and the LGA 1851 socket.

LGA 1700 and 1851 installation

The installation of this air cooler is easy with the included mounting hardware.

1. You’ll first need to place the backplate against the rear of the motherboard after assembling it.

2. Next, you’ll need to take the rubber standoffs and slide them against the poles of the backplate.

3. Then, you’ll take the mounting bars and place them on top of the rubber standoffs, securing them with the included screws.

4. Apply the included thermal paste to your CPU. If you have any questions on how to do this properly, please refer to our handy how to apply thermal paste guide.

5. Remove the middle fan from the cooler, then place the cooler on top of your CPU. Use a screwdriver to secure it using the screws in between the two towers, shown below.

6. Once complete, connect the PWM and ARGB cables to the corresponding motherboard headers. Installation is now complete.

CPU-only thermal results without power limits

Without power limits enforced on Intel’s Core Ultra 9 285K and i7-14700K CPUs, the CPU will hit its peak temperature (TJ Max) and thermally throttle with with almost all air coolers and even most liquid coolers on the market. For the best liquid coolers on the market, the results of this test will be shown using the CPU’s temperature. However, when the CPU reaches its peak temperature, I’ve measured the CPU package power to determine the maximum wattage cooled to best compare their performance. It’s important to note that thermal performance can scale differently depending on the CPU it’s being tested with.

We’ll start by looking at the performance of this cooler with Intel’s Arrow Lake Core Ultra 9 285K, as these are rather interesting!

Normally, when I’ve tested air coolers with Intel’s Core Ultra 9 285k, they run with the CPU operating at peak temperature and experience small amounts of thermal throttling as a result. Every air cooler I’ve tested thus far has demonstrated this behavior – until now! When I tested Jiushark’s JF800 did momentarily peak at maximum temperature, but operated without throttling – the first air cooler I’ve tested to accomplish this.

That said, Arrow Lake’s thermal characteristics have changed a bit since the launch due to both Windows and BIOS updates, which have altered its behavior. I’m going to need to retest coolers like Thermalright’s Peerless Assassin 140 as a result. It’s quite possible that other coolers can also manage this feat with Arrow Lake now.

Again, CPU coolers can scale (and perform) differently depending on the CPU they are paired with due to differences in manufacturing processes and the location of hotspots on the CPU. We’ve also tested Jiushark’s JF800 Diamond with the harder-to-cool Raptor Lake Core i7-14700K to see if its amazing performance would hold up with other CPUs.

With 254W cooled during the course of testing, Jiushark’s JF800 Diamond is again impressive! This is better cooling performance than I’ve seen from any other air cooler and competitive with many AIOs!

To achieve these levels of thermal performance, the fans of this cooler reach up to 48.9 dBA. This is much louder than it needs to be. I don’t understand why Jiushark allowed these fans to run so loudly, especially since – as the next section will show – they perform well even when limited to low noise levels.

CPU-only thermal results with noise normalized to 38.9 dBA

Finding the right balance between fan noise levels and cooling performance is important. While running fans at full speed can improve cooling capacity to some extent, the benefits are limited and many users prefer a quieter system.

With this noise-normalized test, I’ve set noise levels to 38.9 dBA using the i7-14700K system. This level of noise is low, but slightly audible to most people. Jiushark’s JF800 continued to offer the best performance we’ve seen from any air cooler here, with 236W cooled during this test.

253W results

My recent reviews have focused more on tests with both the CPU and GPU being stressed, but many of y’all have indicated that you would like to see more CPU-only tests. In response, I’ve started testing Intel’s “Arrow Lake” Core Ultra 9 285K with a 253W limit.

Testing with the default power limits of 253W shows very good performance, on par with Thermalright’s Peerless Assassin 140!

135W CPU + 290W GPU results

Testing a CPU Cooler in isolation is great for synthetic benchmarks, but doesn’t tell the whole story of how it will perform. I’ve incorporated two tests with a power limit imposed on the CPU, while also running a full load on MSI’s GeForce RTX 4070 Ti Super 16G VENTUS 3X.

The CPU power limit of 135W was chosen based on the worst CPU power consumption I observed in gaming with Intel’s Core Ultra 9 285K, which was in Rise of the Tomb Raider.

In this test, the JF800 Diamond continues to impress. With a temperature recorded of 66 degrees Celsius, it again takes the spot for the best air cooler, outperforming rival options from ID-Cooling, Thermalright, and Ocypus! Noise levels weren’t bad, but they were near the loudest of the results we’ve recorded thus far.

85W CPU + 290W GPU results

Our second round of CPU + GPU testing is also performed with Arrow Lake. The power limit of 85W was chosen based on typical power consumption in gaming scenarios using the Core Ultra 9 285K CPU. This should be fairly easy for most coolers. The main point of this test is to see how quietly (or loudly!) a cooler runs in low-intensity scenarios.

With a CPU temperature of 54 degrees C, Jiushark retains its lead over other air coolers. Its noise level was also relatively low in this test, averaging 38.2 dBA.

Conclusion

Jiushark’s JF800 Diamond is the best air cooler on the market that you can’t (easily) purchase. It outperforms its competitors in every thermal benchmark, even when the fans are noise-normalized. My only complaint is the lack of availability, at least in the U.S. If you want this amazing CPU cooler, you’ll need to import it from AliExpress or another Asian distributor, at least until there is better availability stateside.

Alienware is typically known for sci-fi focused, spaceship-inspired designs. But with its new Aurora laptops, the Dell subsidiary is going for a bit more subtlety.

In time for PAX East, Alienware is launching the Alienware 16 Aurora and Alienware 16X Aurora, systems that blend understated (well, for Alienware) designs with gaming power. It feels like the successor to last year's Alienware m16 R2, which dialed down the gaming looks and added a "stealth mode" for quiet performance and plain lighting – and that feature is coming back with the Auroras.

Both Auroras come in a dark blue "interstellar indigo" chassis that features the alien head logo on the lid, but little else that screams gaming notebook. They have curved edges designed to be comfortable to pick up and carry, and can also be opened with one hand.

Alienware has made these gaming laptops smaller than its classic designs by eliminating the "thermal shelf," a cooling solution that typically juts out the back of the laptop. Instead, there's a cooler on the bottom of the device, where a rear foot would usually be found. This sucks cool air in from the bottom of the laptop and exhausts it from the back and sides.

The 16X Aurora is a more premium version of the system, with higher-end components and parts. It goes up to an Intel Core Ultra 9 275HX, Nvidia GeForce RTX 5070, 64GB of RAM, 2TB display, and a 240 Hz, 300-nit screen. Additionally, it has a higher-resolution webcam than the standard 16, and has a metal lid and bottom (as opposed to just the metal lid on the 16 Aurora).

There are other quality differences: The 16X has single-zone RGB, while the regular 16 just has white backlighting. Additionally, one of the 16X's USB Type-C ports supports Thunderbolt 4, while that's not the case on the 16.

The new systems will start at $1,149 in the US. For the regular Alienware 16 Aurora, that will get you an Intel Core i7 14700H (Raptor Lake Refresh), 16GB of RAM, 1TB SSD, Nvidia RTX 4050, and a QHD, 120 Hz display.

At launch, the Alienware 16X Aurora will start at $1,949, though a cheaper base model will launch "soon after," an Alienware rep tells me. The $1,949 system has a Core Ultra 9 275HX, 32GB of RAM, 2TB of storage, an RTX 5060, and a QHD, 240 Hz screen.

The Aurora laptops follow the launch of the new Area-51 machines announced earlier this year and recently released, and are the first Aurora-branded notebooks from the company in almost 20 years. Alienware's revival of its Aurora and Area-51 laptop sub-brands come as its parent company, Dell, attempts to simplify its naming schemes.

In a bid to improve Arrow Lake sales, Intel has significantly discounted its mainstream 20-core Core Ultra 7 265K, which is readily selling at or below $300 at various retailers, including Amazon. That's a massive 25% cut over its original $400 recommended customer price set by Intel at launch. While the slightly more affordable 14-core Core Ultra 5 245K is also selling below its MSRP at $269, the Core Ultra 7 265K is arguably the better buy.

This is likely part of a small promotional sale, which is why we're using the term discount instead of permanent price cuts. Even so, they don't appear to be regional, as reports from ComputerBase indicate similar pricing trends in Germany. Intel's candidate is a great choice for promoting Arrow Lake sales, since the Core Ultra 7 265K wields an impressive 20 (8P+12E) core / 20 thread configuration, rivaling the i7-14700K from the previous generation, while offering better efficiency.

The CPU carries 66MB of total cache (36MB L2 + 30MB L3). Sticking to JEDEC-compliant speeds, Arrow Lake at stock can handle 6400 MT/s DDR5 kits (CUDIMM), going as fast as 8000 MT/s with Intel's warrantied boost profiles. As of writing, the Core Ultra 7 265K (and its KF variant) can be purchased for $294 at Amazon. The model with the integrated GPU is a no-brainer, given that its built-in Xe-LPG (Alchemist) iGPU offers QuickSync functionality with AV1 encoding.

Intel on Thursday posted its financial results for the first quarter of 2025. The company's earnings were flat year-over-year; however, its losses deepened, and its gross margin declined despite lower operating expenses. While sales of the company's data center grade products demonstrated signs of growth, sales of client CPUs declined compared to the same quarter a year ago. Perhaps more importantly, Intel gave a bleak outlook for the second quarter due to macro challenges. 

In the first quarter of 2025, Intel reported flat year-over-year revenue of $12.7 billion, with a net loss of $821 million, nearly twice the amount compared to the same quarter a year ago. The company's gross margin declined to 36.9%, pressured by a product mix, startup costs for the 18A ramp-up, and uncertainties (which Intel referred to as macroeconomic headwinds).

The company's operating expenses — including research and development (R&D) as well as management, general, and administrative costs (MG&A) — declined to $4.8 billion in Q1 2025 from $5.9 billion in Q1 2024. However, despite this decline, the company's losses increased. 

"The first quarter was a step in the right direction, but there are no quick fixes as we work to get back on a path to gaining market share and driving sustainable growth," said Lip-Bu Tan, Intel CEO. "I am taking swift actions to drive better execution and operational efficiency while empowering our engineers to create great products. We are going back to basics by listening to our customers and making the changes needed to build the new Intel." 

Perhaps the most alarming sign is that Intel's Client Computing Group (CCG) revenues fell 8% year-over-year to $7.6 billion. This drop was attributed to weaker-than-expected PC demand, particularly in the consumer segment, competitive pricing, and unfavorable product mix that includes a plethora of products made by TSMC. Interestingly, many of Intel's customers favored older-generation products like Raptor Lake over newer, higher-cost platforms such as Meteor Lake, Arrow Lake, and Lunar Lake. 

Despite these challenges, Intel noted demand for AI PCs from business customers, as well as enterprise fleet upgrades and Windows 10 end-of-service migrations, although this was insufficient to offset the broader softness in consumer sales. 

Intel's Data Center and AI (DCAI) business unit reported $4.1 billion in revenue in Q1 2025, achieving an 8% year-over-year increase, making it one of the few growth areas for the company. The performance was primarily driven by strong demand from hyperscalers for host CPUs in AI server deployments. However, the segment faced margin pressure due to competitive dynamics from AMD, product mix, and elevated demand for older-generation parts rather than newer offerings. 

Despite the revenue growth, Intel acknowledged macroeconomic uncertainty, potential spending pullbacks, and competition from AMD and Arm-based server solutions as risks that could affect DCAI performance in the coming quarters. The company remains focused on stabilizing market segment share and increasing average selling prices (ASPs) while preparing for the ramp of its next-generation server products. 

Intel Foundry reported $4.7 billion in revenue, reflecting a 7% year-over-year increase driven mostly by internal demand, particularly from Intel's own product groups for wafers and advanced packaging services. Despite revenue growth, the Foundry segment continued to operate at a significant loss, posting an operating loss of $2.3 billion, which remained roughly flat compared to the previous quarter. 

Intel's Lip-Bu Tan reiterated at the conference call that the company's Foundry success hinges not just on manufacturing capabilities but also on building customer trust, improving process design enablement, and supporting a broader range of customer flows. For now, the key mission of Intel Foundry is to ramp up production of Intel 18A-based Panther Lake and then Clearwater Forest processors in late 2025 – 2026 to prove that IF has a node that is competitive with TSMC's N2. 

Times are already tough for Intel, but now it turns out its new, heavily-promoted AI PC chips aren't selling as well as expected, thus creating a shortage of production capacity for its older chips. The news comes as the CEO announced looming layoffs and a poor financial report sent the company's stock tumbling.

Intel says its customers are buying less expensive previous-generation Raptor Lake chips instead of the new, and significantly more expensive, AI PC models like the Lunar Lake and Meteor Lake chips for laptops.

During the earnings call, Intel announced that it currently faces a shortage of production capacity for its 'Intel 7' process node, and the company expects this shortage to "persist for the foreseeable future." That's an unexpected shortage to have, as Intel's current-gen chips use newer process nodes from TSMC instead of Intel's older 'Intel 7' node. Intel is a master at production capacity planning, so its disclosure points to an unexpected surge in sales of the older 'Intel 7' products.

Intel explained that the shortage of its 7nm production capacity is due to an unexpected surge in demand for its "N-1 and N-2" products, a reference to its two prior-generation chip families. This trend is occurring in both the consumer and data center markets.

"What we're really seeing is much greater demand from our customers for n-1 and n-2 products so that they can continue to deliver system price points that consumers are really demanding," explained Intel's Michelle Johnston Holthaus. "As we've all talked about, the macroeconomic concerns and tariffs have everybody kind of hedging their bets and what they need to have from an inventory perspective. And Raptor Lake is a great part. Meteor Lake and Lunar Lake are great as well, but come with a much higher cost structure, not only for us, but at the system ASP price points for our OEMs as well."

Bernstein Research's Stacy Rasgon pressed Holtahaus about the implications for the company's upcoming Panther Lake chips, which are set to launch at the end of the year, especially given that the looming tariff disruptions have not yet occurred.

Holthaus said the Panther Lake launch remains on track and the company expects continued success in the commercial market, which she said typically precedes broader consumer adoption. Notably, she did not directly address the company's expected next-gen AI PC adoption for consumer laptops. Regardless, the company also continues its expansive work to promote and cultivate a growing developer ecosystem to unleash the power of its AI wares.

However, the fact is that AI still doesn't seem to have the 'killer app' that would send waves of customers to stores to purchase an expensive new laptop. Instead, most of the new features revolve around built-in features in existing applications, such as chat and productivity software, that are more nuanced and not quite flashy enough to spark a wave of adoption.

According to documents shared with Tom's Hardware by a source, Intel will announce a new "Intel 200S Boost" feature for its Arrow Lake processors tomorrow that's designed to boost gaming performance by providing official warranty coverage for a subset of overclocking features, including memory overclocking. As you can see below, we have put the new feature through a battery of tests before its official launch and found the gains generally match our expectations for memory overclocking, with an average improvement of 7.5% over the officially supported memory speeds.

It's no secret that Intel's Arrow Lake chips delivered disappointing gaming performance at launch — in fact, they are significantly slower than even Intel's own previous-gen models. The company has since corrected multiple launch-day issues, but that has not improved overall performance. The new approach aims to leverage several existing features and package them under the warranty protection umbrella, much like AMD introduced a 105W mode to boost performance for its underperforming 65W Ryzen 9000 models. However, Intel hasn't issued any official performance projections for the new feature yet.

The Intel 200S Boost feature enhances the performance of Arrow Lake K-series processors by enabling a few overclocking features in an easy-to-use one-click BIOS profile, but the new settings don't impact CPU clock speeds or power settings above current warranty limitations. Instead, the tweaks optimize specific memory and fabric speeds, marking the first time Intel has offered official warranty coverage for potential chip damage resulting from XMP memory overclocking profiles or adjusting fabric speeds.

There are, however, several caveats, and the tweaks are already well known to the enthusiast and overclocking community. Firstly, Intel now covers "up to" DDR5-8000 memory speeds within its warranty; however, not all chips will be able to reach that speed, and because the approach is still considered overclocking, Intel does not guarantee system stability with XMP profiles. As we demonstrate below, more affordable and easily supported DDR5-7200 kits offer nearly the same performance in most games and applications we tested.

The 200S Boost feature will be integrated into BIOS revisions from major motherboard vendors, with BIOS updates expected to arrive tomorrow from at least a few OEMs. The feature will only be implemented on Z-Series motherboards, which is a curious limitation given that Intel now supports memory overclocking on B-Series boards. It's also only available on K- and KF-series SKUs.

The 200S Boost profile also increases the speed of the Next Generation Uncore (NGU/SA Fabric), which enables communication between various chip elements, such as the CPU cores, memory controllers, and other components. This interface is upgraded from its standard 2.6 GHz speed to 3.2 GHz. Additionally, the Die-to-Die (D2D) communication fabric, which serves as a bridge between the Compute and SOC tiles or dies present inside the Arrow Lake chip, is increased from its stock 2.1 GHz to 3.2 GHz.

Intel is also obviously wary of motherboard vendors pushing the limits with their BIOS settings (which they have been known to do) and thus creating another potential chip reliability issue. As such, the company has also instituted several guardrails around the feature, with strict limitations that prevent motherboard makers from altering any other features, such as CPU clock speeds or power thresholds, as part of the 200S Boost settings. Intel also has voltage ceilings for the System Agent and memory that cannot be exceeded. You cannot use XMP kits that exceed the DIMM voltage ratings. The limits are listed in the table above.

The OEMs are allowed to tailor their voltage and speed settings within those constraints to optimize performance with their product. Any manual manipulation by the end user of clocks or other settings will automatically disable the 200S Boost profile, reverting you to manual overclocking. This feature also locks the overclocking mailbox to prevent OS-based overclocking. Finally, 200S Boost is entirely opt-in; it can't be enabled by default in the BIOS and requires users to turn it on.

Intel 200S Boost is separate from the Intel Performance Optimization (IPO) program, a China-specific collaboration between Intel and System Integrators (SIs) that facilitates more robust overclocking, including clock speeds and power settings. However, the SIs carry the warranty for that program, and Intel has no current plans to bring the IPO program to other regions.

Now on to the benchmarks.

Intel 200S Boost Gaming Performance

The 200S Boost feature is built on firmware with an MR1 or newer revision. We tested with the Core Ultra 9 285K on the MSI MEG Z890 ACE motherboard with a .1A53 BIOS revision that supports adding the feature, although it is not explicitly enabled yet. We merely recreated the correct settings for the feature in this BIOS, and sources close to the matter have confirmed that our results align with general expectations.

We tested six configurations and three memory speeds across 16 games at 1080p. The tested memory speeds include the stock DDR5-6400 with JEDEC timings, which was the previous limit for Intel's warranty, as well as the cost- and compatibility-friendly DDR5-7200 speed we use for our CPU reviews (32GB G.Skill Trident Z5 RGB DDR5-7200 kit for both). We also tested with a 32GB Patriot Extreme 5 DDR5-8000 kit to measure the peak supported XMP speeds.

The first slide displays the geometric mean of our gaming tests, which we conducted using an RTX 5090 Founders Edition. We tested the different memory speeds with the fabrics at stock settings and the listed memory speed (marked as stock in the chart), and then retested after increasing the fabric speeds to the higher 3.2 GHz threshold (marked as '200S Boost').

The largest increase in performance undoubtedly comes from memory overclocking. Moving from the stock DDR5-6400 configuration to the peak DDR5-8000 with fabric overclocking (200S Boost) yielded a 7.5% performance increase in our overall measurement. Naturally, performance increases vary by title, from as low as a 3.7% increase in A Plague Tale: Requiem to an 11.6% increase in Baldur's Gate 3. The per-game results generally hover in the 7% to 9% range. The titles that benefit most are simply the memory-sensitive titles, so there are no surprises here.

We, like many other outlets, test with a DDR5-7200 XMP profile as our default memory configuration. We chose this speed because it is widely supported by most chips (you may encounter issues with DDR5-8000 UDIMMs on some chips or motherboards) and it is more affordable — 32GB DDR5-8000 kits typically carry a $45 to $60 (43 to 57%) premium, yet deliver precious little extra performance.

We see that trend hold strong in our results, with the DDR5-8000 200S Boost configuration only being a mostly imperceptible 1.2% faster performance overall in 1080p gaming. Our advice for most enthusiasts remains the same — DDR5-7200 is the sweet spot for price and performance.

We have experimented with overclocking the fabric speeds in the past, but we found the increased performance to be largely unremarkable unless you are engaging in heavy overclocking of the CPU cores, which is not allowed in tandem with the 200S Boost feature.

To determine the impact of fabric speeds on the overall performance improvement, we toggled the fabrics between 200S Boost and Stock fabric settings for each memory speed. As you can see, it did give at least some boost in performance, but the gains undoubtedly fall into the imperceptible range. For instance, we measured a sub-1% increase in performance when using higher fabric clocks with the DDR5-6400 and DDR5-7200 configurations, and a 1.4% increase with the DDR5-8000 configuration.

It's possible that you could eke out higher gains from the fabric tweaks with lesser chips, like the Core Ultra 7 265K and the Core Ultra 5 245K, but you should keep your expectations in check.

We'll see how the fully-boosted Intel chips stack up against the competition further below.

In the interest of due diligence, we ran the chips through several productivity application benchmarks to assess the impact, but the results were entirely predictable — applications that benefit from memory overclocking saw small gains, while others saw none at all. In fact, the variations mostly fall within our expected run-to-run variance, so you shouldn't expect significant uplift in productivity applications.

The competitive landscape is basically unchanged

Here we've added the primary competitors for the Core Ultra 9 285K to the test results, and as you can see, the landscape remains largely unchanged from our most recent testing. Notably, we don't see as significant a gain because we test all processors with reasonable XMP settings applied as our stock configuration in reviews.

The fact that Arrow Lake couldn't match the gaming performance of Intel's own prior-gen Raptor Lake Refresh chips was one of the most disappointing aspects for enthusiasts. That still remains the case; the Core i9-14900K is now 6.5% faster than the 285K, whereas it was 9% faster in our prior testing (with both at DDR5-7200). That change isn't enough to drastically change the value equation between the Intel chips.

The situation also remains rough in comparison to AMD's competing models — the 285K is now about 3% slower than the Ryzen 9 9950X, roughly halving the distance between the two and bringing it closer to a draw, but the gaming-optimized X3D chips continue to dominate by almost absurd amounts. Here, the much less expensive Ryzen 7 9800X3D and its premium counterpart, the Ryzen 9 9950X3D, still hold a 30%+ lead in gaming, so it remains a no-contest if you're strictly focused on gaming performance.

Despite the addition of new fabric tweaks and support for up to DDR5-8000, we still recommend that most users stick with DDR5-7200 — stepping up to DDR5-8000 incurs a significant cost for only about 1% more performance.

Overall, the new 200S Boost feature doesn't alter the competitive landscape, but it does provide an easy-to-use option for less-advanced users to gain a few extra percentage points of performance. The addition of warranty coverage for damage associated with the limited XMP memory or fabric overclocking is nice, but moderate memory overclocking is typically fairly safe in either case.

Intel is expected to officially announce the 200S Boost feature tomorrow, we'll follow up with further details as warranted.

GPU Benchmarks Introduction

Tom's Hardware exhaustively benchmarks every GPU to find out which are worthy of our list of the best graphics cards. Our GPU benchmarks hierarchy ranks current and previous generation graphics cards by performance. Whether it's playing games, running artificial intelligence workloads, or doing professional video editing, your graphics card typically plays the biggest role in determining performance — even the best CPUs for gaming take a secondary role.

Our 2026 GPU Hierarchy testing spans three generations of Nvidia and AMD graphics cards, as well as Intel's Arc B-series GPUs.

Our testing has been made easier by the fact that no truly new gaming GPUs have been introduced in almost a year. If you haven't already upgraded your graphics card after the GeForce RTX 50-series and Radeon RX 9000-series launches in 2025, well, you're still looking at the exact same products now.

AMD did make its formerly China-only Radeon RX 9070 GRE available globally after Computex 2026, but in our review, we found that $549 product to be too expensive given the level of performance it delivers and the compromises made to hit its price point. Check out that coverage for all the details.

Most of the products we recommend remain at elevated prices compared to their MSRPs, but this is just life in mid-2026. It's admittedly cold comfort, but unless you're shopping for an RTX 5090, graphics card prices haven't risen much more than they already did earlier this year.

Compared to the doubling or tripling of prices we've seen for RAM kits and SSDs in 2026 versus last year, a GPU upgrade remains a relatively affordable (and self-contained) option, either as a boost for an existing PC or part of an all-new parts list.

Let's dive into our ranking of GPUs past and present so you can figure out how all those cards stack up.

Our full GPU hierarchy using traditional rendering (aka rasterization) comes first, and below that we have our separate ray tracing hierarchy. We've also mashed up these results into one overarching ranking for the complete pictures.

The results are all collected at native resolution, without enabling DLSS, FSR, or XeSS upscaling or frame generation.

As a brief refresher of the cards in this version of our hierarchy, Nvidia's cutting-edge Blackwell architecture and its DLSS 4 suite of tech upgrades the quality of both upscaling and ray reconstruction on RTX 50-series GPUs. It also adds Multi Frame Generation support. As of mid-2026, MFG can add anywhere from one to five AI-generated frames in between natively rendered ones.

Even if you're not into framegen, DLSS 4.5 upscaling and its transformer neural network architecture can offer a big boost in image quality at the same output resolution compared to earlier DLSS versions. That tech can benefit all GeForce RTX GPUs going back to the RTX 20-series, albeit at a heavier performance cost on older hardware compared to earlier DLSS versions.

Nvidia's Ada Lovelace architecture powers its prior-gen RTX 40-series cards. Ada introduced DLSS Frame Generation, which can double output frame rates in supported games. Ada cards don't benefit from MFG, though.

AMD's Radeon RX 9000 series cards, powered by its latest RDNA 4 architecture, get a big boost to both ray-tracing and AI capabilities with dedicated RT and matrix math accelerators. AMD uses those capabilities to enable its FSR 4 upscaler and its much-improved image quality in a small but growing range of titles, whether through native support or with driver-level overrides.

The FSR Redstone update last year brought ML Frame Generation to RX 9000-series cards. Like FSR 4.x upscaling, ML Frame Generation can be directly integrated in games or enabled through a driver override.

Meanwhile, the last-gen RDNA 3 architecture powers seven RX 7000-series seven desktop cards. Until very recently, RX 7000-series cards couldn't officially run FSR 4 upscaling, but an official version of that tech will be made available for those cards in July 2026.

Intel's Battlemage-powered Arc B580 and Arc B570 offer major improvements in performance and efficiency compared to its Alchemist architecture. Battlemage only serves the entry-level end of the graphics card market today. Intel introduced a larger Battlemage chip in the form of the Arc Pro B70 earlier this year, but that product is intended for AI and professional visualization, and it's priced like it.

On page two, you'll find our 2025-2026 test data for posterity. Page three has our 2024–2022 benchmark suite, which covers previous-generation GPUs running an older test suite and a Core i9-12900K. Page four has an even older 2020–2021 test suite with only raster games, running on a Core i9-9900K testbed. The legacy tables are no longer being actively updated. There's also the "Ancient Legacy GPU hierarchy" (without benchmarks, sorted by theoretical performance) for reference purposes.

The following tables sort everything solely by our performance-based GPU benchmarks, from fastest to slowest. Factors including real-world pricing, graphics card power consumption, overall efficiency, and features aren't factored into the rankings here. The latest results use an AMD Ryzen 7 9800X3D testbed. Here are the tables and benchmark results — rasterization games first, then ray tracing games, and finally the content creation results.

GPU Benchmarks Hierarchy 2026: The tests

For our latest GPU benchmarks, we test every card at a mix of high and ultra settings, depending on the game. We test across three resolutions: 1080p, 1440p, and 4K.

All the scores are expressed as percentages relative to the top-ranking 1080p ultra card, which is of course the GeForce RTX 5090.

Our 2026 test suite comprises the following games. 11 are raster titles, and of those, four have RT support. We test another four games that either require RT to run (DOOM: The Dark Ages and Indiana Jones and the Great Circle) or look best with RT (Assassin's Creed Shadows and Grand Theft Auto V Enhanced.)

• Black Myth Wukong (+RT)
• Alan Wake II (+RT)
• Apex Legends
• Counter-Strike 2
• Fortnite
• Arc Raiders
• Stalker 2
• DOOM: The Dark Ages (RT required)
• Grand Theft Auto V Enhanced (RT only)
• Marvel's Spider-Man 2 (+RT)
• Indiana Jones and the Great Circle (RT required)
• Marvel Rivals
• Assassin's Creed Shadows (RT only)
• Cyberpunk 2077 (+RT)
• Clair Obscur: Expedition 33

GPU Benchmarks Hierarchy 2026: Raster gaming

The FPS score is the geometric mean (equal weighting) of all 11 games. Note that the specifications column links directly to our original review for the various GPUs.

Rasterization GPU Benchmarks, Key Findings

• Unsurprisingly, the RTX 5090 takes the top spot across the board, but prices remain stratospheric in mid-2026. It's also difficult to fully engage the GB202 GPU for pure raster gaming unless you're playing at 4K.
• Among current cards, the RX 9060 XT 8GB and RTX 5060 duke it out for the best performance per dollar at 1080p.
• The Radeon RX 9070 is the 1440p raster value champ. The recently introduced RX 9070 GRE sits just behind, and the RTX 5070 is a somewhat distant third.
• At 4K, the RX 9070 XT is the cheapest way to get into 4K 60 FPS native gaming. But the RX 9070 and RTX 5070 are also strong options among current GPUs if you're willing to enable a dash of upscaling.

Overall, if you're only interested in native raster gaming, Radeons are a strong choice in 2026. But Nvidia offers superior DLSS 4.5 upscaling and Multi Frame Generation support across all RTX 50-series cards, which makes matching your GPU's performance to your monitor's refresh rate a snap.

FSR 4.x upscaling isn't on par with DLSS 4.5 yet, and FSR ML Frame Generation is limited to a 2x framerate boost where it's available.

Rasterization GPU Benchmarks Rankings 2026

Remember that we're not including any upscaling or framegen results in the above table. DLSS, FSR, and XeSS offer differing image quality, and we want to keep things directly comparable.

Don't buy any of the cards at the top of our hierarchy without a high-refresh-rate 2560x1440 or 4K monitor to match. Even one of our best gaming CPUs, like the Ryzen 7 9800X3D, can only do so much when a powerful card becomes CPU-bound.

In general, if you're still running a 60 Hz fixed-refresh-rate monitor, it's entirely possible that you're not seeing all the frames your graphics card can generate. Practically any current-gen graphics card from the RTX 5060 on up is good enough for high-refresh-rate gaming at 1080p in 2026 even without upscaling or framegen, and the RTX 5070 or RX 9070 will give you the same experience at 1440p.

In this era of upscaling and framegen, a high-resolution monitor is no longer an obstacle to the best gaming experience. Upscalers work best with higher-than-1080p output resolutions, and the advent of the DLSS 4.5 upscaling model especially means that you can get both smooth performance and crisp image quality from surprisingly modest graphics cards.

But if you don't have a high-refresh-rate 1440p or 4K monitor to begin with, you can't take full advantage of the free boost to both performance and image quality that AI-powered upscaling offers, nor can you enjoy the full smoothness boost of framegen. Check out our list of the best gaming monitors for a potential upgrade if you're still rocking a 1080p 60 Hz display from the 2010s.

On to our 2026 raster results. Among currently available graphics cards, Blackwell rules the top three spots. The RX 9070 XT ends up fourth, just behind the potent (but much more expensive, as of June 2026) RTX 5070 Ti.

Moving into the midrange, the RX 9070's strong raster performance gives it the edge over the RTX 5070, but it's close. And the 5070 has the full arsenal of DLSS 4.5 features at its disposal in virtually every modern game.

Given the image quality and smoothness advantage of DLSS 4.5 upscaling and MFG, the 5070 got the nod in our most recent round of best graphics card picks after our 2026 retests.

The recently introduced RX 9070 GRE lands between the 5070 and the lower-end 5060 Ti and 9060 XT, a position for which Nvidia has no current-gen answer (and no easy path to one.) But AMD may have priced the GRE too high at $549, and we're already seeing big price drops on that product that might make it uniquely appealing in a market where the RTX 5060 Ti is its direct price competition.

The formerly midrange RTX 5060 Ti 16GB is hopelessly overpriced for the performance it offers in mid-2026, and the formerly somewhat affordable 8GB model is also selling for elevated prices now, making the entire 5060 Ti family impossible to recommend.

The RX 9060 XT 16GB holds down the entry-midrange 16GB position by itself, and its $459 price tag isn't so absurd as to make it un-recommendable. But as RX 9070 GREs start going on sale for $500-ish, the step up is going to look mighty tempting.

As we move further down the stack, the $300-ish Arc B580 represents Intel's best card right now, but it remains very much a budget part in mid-2026, trading blows with the RTX 3060 Ti and RTX 4060. The only reason it lands as high as it does in our results is that its 12GB of VRAM prevents its performance from totally plummeting at 4K, as it does for those older 8GB cards.

The RTX 5060 is 24% faster than the Arc B580 at 1080p and 17% faster at 1440p, and its prices are proportionately higher. The scarce RX 9060 XT 8GB is a whole 30% faster at 1080p and 23% faster at 1440p, for similarly more money.

We don't think the potential longevity afforded by the B580's 12GB of VRAM is worth the massive performance tradeoff versus current-gen 8GB GPUs. Save up a few more bucks and grab an RTX 5060 or RX 9060 8GB instead (and enjoy the more widely adopted DLSS 4.5 or FSR 4 versus XeSS while you're at it).

The $250 Arc B570 is still a good value in mid-2026, and as the only modern, decent graphics card available for (well) under $300, it's still worth a look if you can't stretch to an RTX 5050. But recent software ills that took months to fix have led us to consider it more of a gamble than a product worthy of a recommendation, even though those issues did eventually get corrected.

Ray Tracing GPU Benchmarks 2026

For 2026, we're testing a range of RT titles that present a progressively more difficult workload to the GPUs under test.

Grand Theft Auto V Enhanced, DOOM: The Dark Ages, and Indiana Jones and the Great Circle are all relatively lightweight RT titles that will run on a wide range of RT-ready hardware. Spider-Man 2, Cyberpunk 2077, and Assassin's Creed Shadows represent greater challenges to compute resources, VRAM, or both. And Black Myth Wukong and Alan Wake II still bring even the most powerful graphics cards to their knees at native resolutions.

Our results for those more demanding games represent more of a jumping-off point for upscaling and framegen rather than native performance. But lighter RT titles mean that you can explore the differences in image quality that ray tracing offers without crushing your GPU flat.

In any case, graphics cards at the top of our RT tests are ready for practically any RT game. But as you'll see, that's still quite the high bar to clear in 2026.

Ray Tracing GPU Benchmarks, Key Findings and Notes

• The RTX 5090 really comes into its own with RT enabled, as expected.
• AMD's RX 9070 XT outperforms the RX 7900 XTX in our RT tests, showing that the ray-tracing improvements in RDNA 4 deliver.
• Even with those improvements, the RX 9070 XT can only manage a tie with the RTX 4070 Ti with RT in play, coming in eighth place overall.
• Maintaining 60 FPS in RT titles (as a good foundation for upscaling and framegen enhancements) is a very tall order. Among recent products, you'll want:
  • an RTX 5060 Ti 16GB, RTX 5070, or RX 9070 GRE to handle ~60 FPS RT at 1080p without running out of VRAM
  • An RTX 4070, an RTX 5070 or RX 9070 at 1440p, or an RX 9070 XT or RTX 5070 Ti for the best possible experience
  • An RTX 5080, RTX 4090, or RTX 5090 at 4K

Ray Tracing GPU Benchmarks Rankings 2026

The RTX 5090 and RTX 4090 lead our RT results. But if you want to enjoy high-end RT without a 4090- or 5090-sized bankroll, the RTX 5070 Ti proves itself as the last card with a reasonable price-to-performance ratio before things get crazy.

The 5070 Ti leads the RX 9070 XT by a wide margin in our RT tests, and it's anywhere from 10%-15% behind the absurdly overpriced RTX 5080 across the board. But if you want the absolute best RT performance possible without stepping up to a 4090 or 5090, the extra cash for a 5080 is your best — and only — option.

That said, you should really explore DLSS 4.5 upscaling and MFG on the 5070 Ti before spending big to step up to a 5080.

The RX 9070 XT remains AMD's fastest RT GPU, outpacing the 7900 XTX across all tested resolutions. It leads the RDNA 3 halo card by about 10% across the board, which is an impressive generational leap considering its 33% deficit in CUs versus the XTX.

But the 9070 XT can only trade blows with the RTX 4070 Ti and RTX 5070 with Nvidia competition in the picture. It beats out the 5070 at 1080p and 1440p, but DLSS 4.5 upscaling and MFG give Blackwell gamers more flexible options for boosting performance with only minimal impact to image quality versus both RDNA 4 (and Ada).

Of course, the RX 9070 XT has FSR 4 upscaling and framegen at its disposal in compatible titles, but support for those features isn't as widespread as DLSS.

Meanwhile, the RX 9070 comes in slightly behind the RTX 5070 at 1080p and 1440p for RT. Even though its 16GB of VRAM prevents the 9070's 4K RT performance from completely nosediving, the average frame rate at that resolution isn't high enough to make it a 4K RT choice.

At 1080p, the RX 9070 GRE presents an intriguing RT option for Radeon gamers, delivering a near-60-FPS average baseline that would make a great FSR 4 foundation. But its 12GB of VRAM quickly becomes a liability at 1440p versus the 9070, and 4K is out of the question.

Overall, our demanding 2026 RT test suite shows that the bar for entry to a good RT experience remains high, especially as elevated prices on the RTX 5060 Ti 16GB mean that the first "real" RT-ready Blackwell card at 1080p and up is the RTX 5070. And on the Radeon side, you really want an RX 9070 GRE or RX 9070 for the best experience.

Test Setup and Hardware for GPU Benchmarks

We've used several different PCs for our testing over the years. The latest update switches to an AMD Zen 5 processor: the unparalleled Ryzen 7 9800X3D and its 3D V-Cache-enhanced performance.

Here are the specifications for our latest GPU test PC.

Tom's Hardware 2026 GPU Benchmarks Testbed

CPU: AMD Ryzen 7 9800X3D
CPU Cooler: Thermalright Phantom Spirit 120 SE
Memory: G.Skill Trident Z5 32GB (2x16GB) DDR5-6000 CL30
Motherboard: Asus TUF Gaming X670E-Plus Wifi
SSD: Inland Performance Plus 4TB
PSU: MSI MPG Ai1600TS 1600W

We test across the three most common gaming resolutions, 1080p, 1440p, and 4K, using a mix of high and ultra settings, depending on the title. Where possible, we use 'reference' cards for all of these tests, like Nvidia's Founders Edition models and AMD's reference designs. Most midrange and lower GPUs don't get reference models, however, and in some cases we only have factory-overclocked cards for testing. We do our best to select cards that are close to the reference specs in such cases.

For each graphics card, we follow the same testing procedure. We run one pass of each benchmark to "warm up" the GPU after launching the game, then perform our actual test runs across each resolution.

We carefully review our test data and check for anomalies. For example, we always expect the RTX 5080 to be faster than the RTX 5070 Ti. If it's not, and we're not in a CPU limited situation, we'll recheck both cards to ensure that our standings our accurate. We also check and retest in cases of subtler issues, as when a transient hitch or frame-time spike causes a large dip in 1% low FPS.

Due to the length of time required for testing each GPU, updated drivers and game patches inevitably come out that can impact performance. We periodically retest a few sample cards to verify our results are still valid, and if not, we go through and retest the affected game(s) and GPU(s). We may also add games to our test suite over time, if one comes out that is popular and conducive to testing. See what makes a good game benchmark for our selection criteria.

GPU Benchmarks: Individual Game Charts

The above tables provide a summary of performance, but for those that want to see the individual game charts, for both the standard and ray tracing test suites, we've got those as well. These charts only cover current-gen GPUs for readability.

These charts are up to date as of June 2026.

GPU Benchmarks — 1080p

GPU Benchmarks — 1440p

GPU Benchmarks — 4K

GPU Benchmarks — Power, Clocks, and Temperatures

Most of our discussion has focused on performance, but for those interested in power and other aspects of the GPUs, here are the appropriate charts. We'll run these from highest to lowest settings, as 4K ultra tends to be the most strenuous workload on most of these GPUs.

If you're looking for the legacy GPU hierarchy, head over to page two! We moved it to a separate page to help improve load times in our CMS as well as for the main website. And if you're looking to comment on the GPU benchmarks hierarchy, head over to our forums and join the discussion!

Choosing a Graphics Card

Choosing a Graphics Card

Choosing a Graphics Card

Which graphics card do you need? To help you decide, we created this GPU benchmarks hierarchy, comprising 48 GPUs from three generations of hardware for Nvidia and AMD and also including Intel's Battlemage cards. Not surprisingly, the fastest cards are using the latest GPU architectures, though they're not always a major upgrade over the prior generation.

Of course, it's not just about playing games. Many applications use the GPU for other work. But a good graphics card for gaming will typically do equally well in complex GPU computational workloads. Buy one of the top cards and you can run games at high resolutions and frame rates with the effects turned all the way up, and you'll be able to do content creation work as needed. Drop down to the middle and lower portions of the list and you'll need to start dialing down the settings to get acceptable performance in regular gameplay and GPU benchmarks.

If your main goal is gaming, you can't forget about the CPU. Getting the best possible gaming GPU won't help you much if your CPU is underpowered or many years out of date. So be sure to check out the Best CPUs for gaming page, as well as our CPU Benchmarks Hierarchy to make sure you have the right CPU for the level of gaming performance you're looking to achieve.

And don't forget about your monitor. Be sure to get a display whose refresh rate range matches the average frame rates of the graphics card that you want in the games that you play. Check out our list of the best gaming monitors for a starting point.

GPU Benchmarks from 2022–2024

Our 'legacy' GPU benchmarks used different hardware. Here are the details for the 2022–2024 configuration that used an Alder Lake 12900K CPU.

Tom's Hardware 2022–2024 GPU Testbed

Intel Core i9-12900K
MSI Pro Z690-A WiFi DDR4
Corsair 2x16GB DDR4-3600 CL16
Crucial P5 Plus 2TB
Cooler Master MWE 1250 V2 Gold
Cooler Master PL360 Flux
Cooler Master HAF500
Windows 11 Pro 64-bit

We have upgraded our test system and changed our test suite for 2025, effectively resetting our benchmarking and rankings. However, the old data — collected using a Core i9-12900K PC — remains valid. We aren't testing new GPUs with the old testbed, as that would double the time required for what is already a lengthy process, but we felt it would be helpful to some to keep the final data available.

We also had an interim system that never quite got wrapped up, which had a Core i9-13900K Raptor Lake CPU. Our original CPU was one of the impacted units that failed over time, which explains in retrospect why so much of the data felt a bit questionable. Thankfully, the new Ryzen 7 9800X3D system seems to be running just fine. Here are the 12900K results.

For these GPU benchmarks, we tested nearly every GPU released between 2016 and 2024, plus a few extras. All graphics cards were tested at 1080p medium and 1080p ultra, and we sorted the table by the 1080p ultra results. Where it made sense, we also tested at 1440p ultra and 4K ultra. All the scores are scaled relative to the top-ranking 1080p ultra card of the time, which in our suite is the RTX 4090 — especially at 4K and 1440p.

The above summary charts show the relative performance of the cards we've tested across the past several generations of hardware at 1080p ultra — swipe through the above gallery if you want to see the 1080p medium, 1440p, and 4K ultra images. There are a few missing options (e.g., the GT 1030, RX 550, and several Titan cards), but otherwise it's basically complete. Note that we also have data in the table below for some of the other older GPUs.

The eight games used for our standard GPU benchmarks hierarchy are Borderlands 3 (DX12), Far Cry 6 (DX12), Flight Simulator (DX11 Nvidia, DX12 AMD/Intel), Forza Horizon 5 (DX12), Horizon Zero Dawn (DX12), Red Dead Redemption 2 (Vulkan), Total War Warhammer 3 (DX11), and Watch Dogs Legion (DX12). The fps score is the geometric mean (equal weighting) of the eight games. Note that the specifications column links directly to our original review for the various GPUs.

*: GPU couldn't run all tests, so the overall score is slightly skewed at 1080p ultra.

While the RTX 4090 does technically take first place at 1080p ultra, it's the 1440p and especially 4K numbers that impress. It's less than 2% faster than the RTX 4080 Super at 1080p ultra, but that increases to 9% at 1440p and then 25% at 4K. Also note that the fps numbers in our table incorporate both the average and minimum fps into a single score — with the average given more weight than the 1% low fps.

Again, keep in mind that we're not including any ray tracing or DLSS results in the above table, as we use the same test suite with the same settings on all current and previous generation graphics cards. Since only RTX cards support DLSS (and RTX 40-series if you want DLSS 3), that would drastically limit which cards we could directly compare. You can see DLSS 2/3 and FSR 2 upscaling results in our RTX 4070 review if you want to check out how the various upscaling modes can help.

The RTX 4090 comes at a steep price, though on paper it's not that much worse than the previous generation RTX 3090. In fact, we'd say it's a lot better in some respects, as the 3090 was only a minor improvement in performance compared to the 3080 at the time of launch, but with more than double the VRAM. Nvidia pulled out all the stops with the 4090, increasing the core counts, clock speeds, and power limits to push it beyond all contenders. There are two problems with the 4090, however: It's not available at MSRP any longer, due to demand from the AI sector — it often costs $2,000 or more — and there are still concerns with pulling 450W of power over the 16-pin connector.

Stepping down from the RTX 4090, the RTX 4080 Super and RX 7900 XTX trade blows at higher resolutions, while CPU bottlenecks come into play at 1080p. We'll be switching our testbed in the near future, with the current results from our 13900K testing in the charts at the bottom of the page.

Outside of the latest releases from AMD and Nvidia, the RX 6000- and RTX 30-series chips still perform reasonably well and if you're using such a card, there may not be any need to upgrade at present. Intel's Arc GPUs also fall into this category and are something of a wild card.

We've been testing and retesting GPUs periodically, and the Arc chips running the latest drivers now complete all of our benchmarks without any major anomalies. (Minecraft was previously a problem, though Intel has finally sorted that out.) They're not great on efficiency, but overall performance and pricing for the A750 is quite good.

Turning to the previous generation GPUs, the RTX 20-series and GTX 16-series chips end up scattered throughout the results, along with the RX 5000-series. The general rule of thumb is that you get one or two "model upgrades" with the newer architectures, so for example the RTX 2080 Super comes in just below the RTX 3060 Ti, while the RX 5700 XT basically matches the newer and less expensive RX 6600 XT.

Go back far enough and you can see how modern games at ultra settings severely punish cards that don't have more than 4GB VRAM. We've been saying for a few years now that 4GB was just scraping by, and these days we'd avoid buying anything with less than 8GB of VRAM — 12GB or more is the minimum we'd want with a mainstream GPU, and 16GB or more for high-end and above. Old cards like the GTX 1060 3GB and GTX 1050 actually failed to run some of our tests, which skews their results a bit, even though they do better at 1080p medium.

Now let's switch over to the ray tracing hierarchy.

Ray Tracing GPU Benchmarks 2022–2024

Enabling ray tracing, particularly with demanding games like many of those we're using in our DXR test suite, can cause framerates to drop off a cliff. We're testing with "medium" and "ultra" ray tracing settings. Medium generally means using the medium graphics preset but turning on ray tracing effects (set to "medium" if that's an option; otherwise, "on"), while ultra turns on all of the RT options at more or less maximum quality.

Because ray tracing is so much more demanding, we're sorting these results by the 1080p medium scores. That's also because the RX 6500 XT, RX 6400, and Arc A380 basically can't handle ray tracing even at these settings, and testing at anything more than 1080p medium would be fruitless.

The five ray tracing games used are Bright Memory Infinite, Control Ultimate Edition, Cyberpunk 2077, Metro Exodus Enhanced, and Minecraft — all of these use the DirectX 12 / DX12 Ultimate API. The fps score is the geometric mean (equal weighting) of the five games, and the percentage is scaled relative to the fastest GPU in the list, which again is the GeForce RTX 4090.

If you want to see what the future may hold with ray tracing, check out our Alan Wake 2 benchmarks where the full path tracing barely manages playable performance even with upscaling on non-Nvidia GPUs. However, and this is a critical point, the number of games where RT truly makes a striking difference in visuals is extremely limited. For most games, we still feel running pure rasterization rendering makes more sense.

If you felt the RTX 4090 performance was impressive at 4K in our standard test suite, just take a look at the results with ray tracing. Nvidia put even more ray tracing enhancements into the Ada Lovelace architecture, and those start to show up here. There are still further potential performance improvements for ray tracing with SER, OMM, and DMM — not to mention DLSS 3, though that ends up being a bit of a mixed bag, since the generated frames don't include new user input and add latency.

If you want a real kick in the pants, we also ran many of the faster ray tracing GPUs through Cyberpunk 2077's RT Overdrive mode, which implements full "path tracing" (full ray tracing, without any rasterization) — as well as Alan Wake 2, which uses path tracing at higher settings, and Black Myth: Wukong that supports full ray tracing. Those games provide a glimpse of how future games could behave, and why upscaling and AI techniques like frame generation are here to stay.

Even at 1080p medium, a relatively tame setting for DXR (DirectX Raytracing), the RTX 4090 roars past all contenders and leads the previous generation RTX 3090 Ti by 41%. At 1080p ultra, the lead grows to 53%, and it's nearly 64% at 1440p. Nvidia made claims before the RTX 4090 launch that it was "2x to 4x faster than the RTX 3090 Ti" — factoring in DLSS 3's Frame Generation technology — but even without DLSS 3, the 4090 is 72% faster than the 3090 Ti at 4K.

AMD continued to relegate DXR and ray tracing to secondary status, focusing more on improving rasterization performance — and on reducing manufacturing costs through the use of chiplets on the new RDNA 3 GPUs. As such, the ray tracing performance from AMD isn't particularly impressive. The top RX 7900 XTX basically matches Nvidia's previous generation RTX 3080 12GB, which puts it barely ahead of the RTX 4070 — and that's not even in all DXR games. There are some minor improvements for RT performance in RDNA 3, though, as the 7800 XT for example ends up basically tied with the RX 6800 XT in rasterization performance but is 10% faster in DXR performance.

Intel's Arc A7-series parts show a decent blend of performance in general, with the A750 coming in ahead of the RTX 3060 overall. With the latest drivers (and with vsync forced off in the options.txt file), Minecraft performance also looks much more in line with the other Arc DXR results.

You can also see what DLSS Quality mode did for performance in DXR games on the RTX 4090 in our review, but the short summary is that it boosted performance by 78% at 4K ultra. DLSS 3 frame generation improved framerates another 30% to 100% in our testing, though we recommend exercising (extreme) caution when looking at FPS with the feature enabled. It can boost framerates in benchmarks, but when actually playing games it often doesn't feel much faster than without the feature.

Overall, with DLSS 2, the 4090 in our ray tracing test suite is nearly four times as fast as AMD's RX 7900 XTX. Ouch. AMD's FSR 2 and FSR 3 can help as well, and AMD continues to work on increasing the rate of adoption, but it still trails DLSS both in the number of games supported and in the overall image quality. Only two of the games in our DXR suite have FSR2 support. By comparison, all of the DXR games we're testing support DLSS2 — and one also supports DLSS3.

Without FSR2, AMD's fastest GPUs can only clear 60 fps at 1080p ultra, while remaining decently playable at 1440p with 40–50 fps on average. But native 4K DXR remains out of reach for just about every GPU, with only the 3090 Ti and above breaking the 30 fps mark on the composite score — and a couple of games still come up short on the 3090 Ti.

AMD also has FSR 3 frame generation. Like DLSS3, it adds latency, and AMD requires the integration of Anti-Lag+ support in games that use FSR 3. But Anti-Lag+ only works with AMD GPUs, which means non-AMD cards will likely incur a larger latency penalty. We've tested it in Avatar: Frontiers of Pandora and found it worked pretty well, but that was not the case in Forspoken and Immortals of Aveum. It has since gained a lot more traction, though quality and latency remain quite variable — it can look and run well in one game, and then fall flat in another.

The midrange GPUs like the RTX 3070 and RX 6700 XT basically manage 1080p ultra and not much more, while the bottom tier of DXR-capable GPUs barely manage 1080p medium — and the RX 6500 XT can't even do that, with single digit framerates in most of our test suite, and one game that wouldn't even work at our chosen "medium" settings. (Control requires at least 6GB VRAM to let you enable ray tracing.)

Intel's Arc A380 ends up just ahead of the RX 6500 XT in ray tracing performance, which is interesting considering it only has 8 RTUs going up against AMD's 16 Ray Accelerators. Intel posted a deep dive into its ray tracing hardware, and Arc seems reasonably impressive, except for the fact that the number of RTUs severely limits performance. The top-end A770 still only has 32 RTUs, which proves sufficient for it to pull ahead (barely) of the RTX 3060 in DXR testing, but it can't go much further than that. Arc A750 and above also ends up ahead of AMD's RX 6750 XT in DXR performance, showing just how poor AMD's RDNA 2 hardware is when it comes to ray tracing.

It's also interesting to look at the generational performance of Nvidia's RTX cards. The slowest 20-series GPU, the RTX 2060, still outperforms the newer RTX 3050 by a bit, but the fastest RTX 2080 Ti comes in a bit behind the RTX 3070. Where the 2080 Ti basically doubled the performance of the 2060, the 3090 delivers about triple the performance of the 3050.

Test System and How We Test for GPU Benchmarks

For each graphics card, we follow the same testing procedure. We run one pass of each benchmark to "warm up" the GPU after launching the game, then run at least two passes at each setting/resolution combination. If the two runs are basically identical (within 0.5% or less difference), we use the faster of the two runs. If there's more than a small difference, we run the test at least twice more to determine what "normal" performance is supposed to be.

We also look at all the data and check for anomalies, so for example RTX 3070 Ti, RTX 3070, and RTX 3060 Ti all generally going to perform within a narrow range — 3070 Ti is about 5% faster than 3070, which is about 5% faster than 3060 Ti. If we see games where there are clear outliers (i.e. performance is more than 10% higher for the cards just mentioned), we'll go back and retest whatever cards are showing the anomaly and figure out what the "correct" result would be.

Due to the length of time required for testing each GPU, updated drivers and game patches inevitably will come out that can impact performance. We periodically retest a few sample cards to verify our results are still valid, and if not, we go through and retest the affected game(s) and GPU(s). We may also add games to our test suite over the coming year, if one comes out that is popular and conducive to testing — see our what makes a good game benchmark for our selection criteria.

GPU Benchmarks: Individual Game Charts

The above tables provide a summary of performance, but for those that want to see the individual game charts, for both the standard and ray tracing test suites, we've got those as well. We're only including more recent GPUs in these charts, as otherwise things get very messy. These are also using a newer test suite and a 13900K CPU, which changes the performance slightly from the above table, simply because our newest tests are more relevant (but haven't been run on a lot of the older GPUs shown in the tables).

These charts were up to date as of November 11, 2024.

GPU Benchmarks — 1080p Medium

GPU Benchmarks — 1080p Ultra

GPU Benchmarks — 1440p Ultra

GPU Benchmarks — 4K Ultra

GPU Benchmarks — Power, Clocks, and Temperatures

Most of our discussion has focused on performance, but for those interested in power and other aspects of the GPUs, here are the appropriate charts.

Our even older 2020–2021 'legacy' GPU benchmarks used another set of hardware with different games. Here are the details for the Coffee Lake 9900K and Z390 PC.

Tom's Hardware 2020–2021 GPU Testbed

Intel Core i9-9900K
Corsair H150i Pro RGB
MSI MEG Z390 Ace
Corsair 2x16GB DDR4-3200
XPG SX8200 Pro 2TB
Windows 10 Pro (21H1)

The results below combine results from nine games with six resolution and setting combinations. All of the scores are combined (via a geometric mean calculation) into a single overall result, which tends to penalize the fastest and slowest GPUs — CPU bottlenecks come into play at 1080p medium, while VRAM limitations can kill performance at 4K ultra.

These results have not been updated since early 2022, when we added the RTX 3050 and RX 6500 XT to the list. We won't be adding future GPUs to this table, so there's no RTX 40-series, RX 7000-series, Arc, 3090 Ti, 6950 XT, 6750 XT, or 6650 XT, but it does help to provide a look at a slightly less demanding suite of games, where 6GB or more VRAM isn't generally required at 1080p ultra settings. You can use these older results to help inform your purchase decisions, if you don't typically run the latest games at maxed out settings.

Ancient Legacy GPU Hierarchy

Below is our legacy desktop GPU hierarchy dating back to the late 1990s. We have not tested most of these cards in many years, driver support has ended on most models, and the relative rankings are pretty coarse. Note that we also don't factor in memory bandwidth or features like AMD's Infinity Cache or Nvidia's larger L2 cache on Ada Lovelace. The list below is mostly intended to show relative performance between architectures from a similar time period.

We sorted the table by the theoretical GFLOPS, though on architectures that don't support unified shaders, we only have data for "Gops/s" (giga operations per second). That's GeForce 7 and Radeon X1000 and earlier — basically anything from before 2007. We've put an asterisk (*) next to the GPU names for those cards, and they comprise the latter part of the table. Comparing pre-2007 GPUs against each other should be relatively meaningful, but trying to compare those older GPUs against newer GPUs gets a bit convoluted.

These results are, at best, merely theoretical and we don't have any recent benchmarks for most of the GPUs. As one recent example, AMD's RX 7900 GRE ranks above the RTX 4070 Ti Super, even though Nvidia's card nearly matches the RX 7900 XT in rasterization performance and easily beats even the 7900 XTX in ray tracing performance. Take the following with a healthy dose of skepticism and a liberal sprinkling of salt, in other words, but it does contain a list of just about every major desktop GPU from the past 25 years.