Skip to main content

AMD Teases 5nm Ryzen 7000 ‘Raphael’ Zen 4 CPUs, Unveils Ryzen 7 5800X3D with 96MB of L3 Cache

AMD
(Image credit: AMD)

AMD’s Zen architecture propelled it to the top of the desktop PC market, but Intel’s Alder Lake CPUs have now taken the lead on our list of Best CPUs for gaming. Not to be outgunned, today AMD teased renders of the coming 5nm Zen 4 chips that will come with Ryzen 7000 branding and a radical new design that will drop into the new AM5 LGA socket. In addition, AMD demoed a Ryzen 7000 ‘Raphael’ chip running Halo Infinite with all cores surpassing 5.0 GHz during the benchmark. AMD says the Ryzen 7000 processors will arrive in the second half of the year. We have more in-depth details below.

In the interim, AMD will fire back at Alder Lake this spring with its Ryzen 7 5800X3D processor. This new chip features AMD’s new 3D V-Cache tech that packs a whopping 96MB of L3 cache onto a single processor via 3D-stacked SRAM, which AMD says will boost gaming by an average of 15% across a selected range of game titles at 1080p. AMD says the 5800X3D will be the fastest gaming chip in the world, and it provided its own benchmark results against Intel’s flagship Core i9-12900K to prove it.

AMD also announced its Ryzen 6000 ‘Rembrandt’ series mobile chips with the Zen 3+ CPU architecture, RDNA 2 graphics, and the 6nm process. You can read about those chips here.

The X3D Era Begins - Ryzen 7 5800X3D with 3D V-Cache

The Ryzen 7 5800X3D is the first AMD chip that will come to the consumer market with its 3D V-Cache technology, though AMD already uses this tech on its Milan-X processors for the data center. As a quick refresher, 3D V-Cache leverages a novel new technique that uses hybrid bonding to fuse an additional 64MB of 7nm SRAM cache stacked vertically atop the Ryzen compute chiplet, thus tripling the amount of L3 cache per Ryzen die. You can read the deep dive details here and here

AMD

(Image credit: AMD)

As AMD CEO Lisa Su demoed at Computex last year, 3D V-Cache can enable up to 192MB of L3 cache on a single consumer chip. However, the 192MB of L3 only applies to Ryzen models with two dies, and AMD’s new Ryzen 7 5800X3D only has a single die. That means it comes with 64MB of cache stacked atop the 32MB of cache already present on the chip, thus yielding 96MB of L3 cache. It remains unclear if AMD will bring another ‘X3D’ model based on the higher-end Ryzen 9 5900X or 5950X to market in the future. 

PriceCores | ThreadsL3 CacheP-Core Base/BoostE-Core Base/BoostTDP / PBP / MTPMemory Support
Ryzen 7 5800X3D?8P | 16 threads96MB3.4 / 4.5 GHz-105WDDR4-3200
Ryzen 7 5800X~$3908P | 16 threads32MB3.8 / 4.7 GHz- 105WDDR4-3200
Core i9-12900K / KF$589 (K) - $564 (KF)8P + 8E | 16 Cores / 24 threads30MB3.2 / 5.2 GHz2.4 / 3.9 GHz125W / 241WDDR4-3200 / DDR5-4800
Ryzen 9 5900X$54912P | 24 threads32MB3.7 / 4.8 GHz-105WDDR4-3200

The Ryzen 7 5800X3D comes with the same eight Zen 3 cores and 16 threads as the standard Ryzen 7 5800X, but has a lower 3.4-GHz base and 4.5 GHz boost frequency. That means AMD has reduced the base clock by 400 MHz and trimmed 200 MHz off the boost frequency. The chip will drop into existing 400- and 500-series motherboards, offering a hassle-free upgrade path for AMD users as they wait for Zen 4. AMD hasn’t announced pricing yet, but as we’ll explain below, it could be tricky.

Image 1 of 2

AMD

(Image credit: AMD)
Image 2 of 2

AMD

(Image credit: AMD)

The 5800X3D’s lower clock speeds are an inevitable tradeoff — AMD stacks the additional SRAM directly in the center of the compute die to isolate it from the heat-generating cores on the sides of the chiplet. However, due to the need for a uniform mating surface with the integrated heat spreader, AMD has to use silicon shims on top of the cores to provide an even surface for the top of the chiplet. Silicon is an excellent thermal conductor but it will inevitably trap at least some heat, thus resulting in less thermal headroom. It also consumes more power. That’s likely the source of the slightly lower clock speeds, but as you can see below, the net result is a big win for gaming performance.

AMD Ryzen 7 5800X3D Gaming Benchmarks

AMD’s stacked SRAM delivers 2TB/s of throughput between the two dies and minimal latency impact. Along with the additional cache capacity that’s kept closer to the cores, it leads to the types of performance gains that we would expect from the jump to a new chip architecture. Above we can see AMD’s internal test results, and as with all vendor-provided benchmarks, take them with the requisite grain of salt.

Image 1 of 2

AMD

(Image credit: AMD)
Image 2 of 2

AMD

(Image credit: AMD)

AMD measured an average of 15% more performance against the company’s fastest gaming chip, the Ryzen 9 5900X, and claims the Ryzen 7 5800X3D is now the fastest gaming chip in the world. Most of these games are sensitive to memory latency, so keeping data close to the execution cores yields big wins. However, the impact of the additional cache remains to be seen in games that aren’t as sensitive to latency.

Flipping over to AMD’s gaming benchmarks against the Core i9-12900K brings slimmer leads for the 5800X3D, and three of the titles end up in a tie. However, we see a 20% advantage in Final Fantasy XIV, and a 10% advantage in Shadow of the Tomb Raider and Far Cry 6. Overall, that works out to an average gain of ~7% across all six titles. Again, this is a relatively limited selection of game titles that tend to be sensitive to memory latency, meaning they play well to the 5800X3D’s strengths.

We’ve seen how the EPYC Milan-X chips profit in data center workloads from the increased cache, but it’s unclear how the 5800X3D will respond in typical desktop PC applications. We’re sure AMD would provide plenty of examples if there were significant uplift, so it’s possible the extra cache doesn’t provide much uplift for PCs in non-gaming tasks.

The Ryzen 7 5800X3D represents AMD’s attempt at taking back the gaming crown from Intel, but pricing will be tricky. If the Ryzen 7 5800X3D outperforms the 12900K in a broader range of games, it will command a much higher price tag than the $390 Ryzen 7 5800X. However, given that it has fewer cores than the Core i9-12900K and resides on the AM4 platform with older connectivity tech, it won’t be able to command a similar $590 price tag. It will be interesting to see pricing, especially given the fact that this type of packaging tech inevitably adds cost, and if AMD plans to bring more X3D models to market. We’ll certainly learn more before the launch in spring of this year.

5nm Zen 4 Chips for the Ryzen 7000 Family

(Image credit: AMD)

AMD also announced that the 5nm Zen 4 ‘Raphael’ chips would come to desktop PCs in the second half of 2022. These chips will fall into the Ryzen 7000 family and come with a completely redesigned integrated heatspreader (IHS) with cutouts along the periphery to accommodate the unique capacitor/SMD layout spread around the PCB. This design is somewhat similar to Intel’s older Skylake-X Core i9-7980XE HEDT models, but has more extensive cutouts. 

(Image credit: AMD)

The Raphael processors will drop into a new AM5 socket that supports both the PCIe 5.0 and DDR5 interfaces, matching Alder Lake on the connectivity front. This new socket marks a big departure for AMD — the company is moving from its long-lived Pin Grid Array (PGA) AM4 sockets to a Land Grid Array (LGA) AM5 layout. Despite the entirely different LGA1718 socket interface (1718 pins), the AM5 socket will still support AM4 coolers.

Rumors have long proliferated that you’ll need a conversion kit due to the AM5 socket’s higher Z-Height, but even if it requires a conversion kit, it’s encouraging that the AM4 cooler ecosystem will be alive and well for quite some time.

Image 1 of 2

AMD

(Image credit: AMD)
Image 2 of 2

AMD

(Image credit: AMD)

TSMC says its N5 process, the node AMD will presumably use for Raphael, offers a 30% power reduction or 15% performance improvement along with a 1.8X increase in density compared to the 7nm process that AMD currently uses in its Ryzen 5000 chips. AMD also has 5nm Zen 4 designs destined for the data center, too, but plans to use an HPC variant of N5 that has twice the density and performance of the standard N7 node, along with 1.25X the performance.

During the keynote, AMD demoed a Zen 4 processor running Halo Infinite in real-time, and all the cores boosted beyond 5.0 GHz, signaling the N5 process has some room for higher clocks.

The Ryzen 7000 branding marks another leapfrogging of Ryzen’s generational monikers — AMD is jumping from Zen 3-powered Ryzen 5000 chips for the desktop PC to Zen 4 Ryzen 7000 models, with no apparent 6000 series for desktop PCs. That comes because the new Ryzen 6000 processors also announced at the show come with the Zen3+ architecture with the 6nm process. But it does cloud the branding scheme again after the company recently corrected a similar issue and re-unified its stack after the Ryzen 4000 series that never came to mainstream client PCs.

Overall, AMD has a busy year ahead with the rollout of its Ryzen 6000 Rembrandt chips, Ryzen 7 5800X3D, and Zen 4 desktop PCs. As evidenced by the Zen 4 gaming will all cores hitting over 5.0 GHz, AMD is obviously far down the development road with its next line of processors. Now, all that’s left to do is execute on its roadmaps, and AMD has a pretty good track record of doing just that.

Paul Alcorn is the Deputy Managing Editor for Tom's Hardware US. He writes news and reviews on CPUs, storage and enterprise hardware.

  • VforV
    I find it so weird that only one Ryzen 3D CPU is launching and it's the "middle" tier 5800X3D... but if it really beats the 12900k in gaming I will really LMAO :ROFLMAO:

    Zen4 at 5Ghz though should be crazy good!
    Reply
  • shady28
    I guess they didn't fully solve the heat issues with 3D stack technology. I find it hard to believe the cache will completely overcome 5-10% lower clocks.

    Also, a single SKU, seems like this may be more of a technology demonstrator and marketing move than a part meant for mass consumption. I'd bet these mostly fall into the 'unobtanium' category with both high demand and limited production volume. If you want one, best sign up on pre-order as soon as it hits.
    Reply
  • PCWarrior
    So after all these headlines over 6 months ago about 3D Vcache AMD will launch just ONE cpu and that in 2-3 months from now. And that is not even the 12-core they demoed back then but a mid-tier octacore with lower clocks than its predecessor! Talking about too little too late. As for the games where they showed it winning against the 12900K, these happen (surprise surprise) to be amongst the games where the stock configuration of the 12900K (with the e-cores enabled) is lower than with the e-cores disabled. Disabling the e-cores (or assigning gaming to p-cores only which can be done in software) brings parity in these games. I suspected that something was off with this whole Zen3D Vcache launch with the lack of leaks. But I could never predict such a fiasco.
    Reply
  • shady28
    PCWarrior said:
    So after all these headlines over 6 months ago about 3D Vcache AMD will launch just ONE cpu and that in 2-3 months from now. And that is not even the 12-core they demoed back then but a mid-tier octacore with lower clocks than its predecessor! Talking about too little too late. As for the games where they showed it winning against the 12900K, these happen (surprise surprise) to be amongst the games where the stock configuration of the 12900K (with the e-cores enabled) is lower than with the e-cores disabled. Disabling the e-cores (or assigning gaming to p-cores only which can be done in software) brings parity in these games. I suspected that something was off with this whole Zen3D Vcache launch with the lack of leaks. But I could never predict such a fiasco.

    Lowering the clock like that, at best I think it will be a mixed bag. Some things will be faster where they're hindered by the cache, other things slower. This seems like more of a tech demo of 3D stacking to me. That said, 3D stacking is still very cool, though AMD is not the first at it by a long shot. It would be the first on a high power desktop CPU though (Intel Lakefield was 3D stacked, but very low power, like 10W or some such).
    Reply
  • rluker5
    Interesting that AMD finds the 12900k = the 5900X (5800X3D=1.1times5900X, 5800X3D=1.1times12900k) at 1080p in SOTTR. Suppose we will have to wait for reviews and not go by what AMD says again.
    Reply
  • rluker5
    VforV said:
    I find it so weird that only one Ryzen 3D CPU is launching and it's the "middle" tier 5800X3D... but if it really beats the 12900k in gaming I will really LMAO :ROFLMAO:

    Zen4 at 5Ghz though should be crazy good!
    Halo Infinite doesn't stress modern cpus. Using it is demonstrating low load clocks. High load clocks are yet to be seen.
    Reply
  • PCWarrior
    shady28 said:
    Lowering the clock like that, at best I think it will be a mixed bag. Some things will be faster where they're hindered by the cache, other things slower. This seems like more of a tech demo of 3D stacking to me. That said, 3D stacking is still very cool, though AMD is not the first at it by a long shot. It would be the first on a high power desktop CPU though (Intel Lakefield was 3D stacked, but very low power, like 10W or some such).
    This is a clear example of overpromising and underdelivering. Clearly it didn’t work out and the product should have been cancelled. But as they made all that noise about it they had to release something and we get this ... thing. Let’s not forget here how many people postponed their upgrades waiting over this mythical 3DVcache cpu lineup…
    Reply
  • sizzling
    Just playing devils advocate here. As AMD are having issues getting enough production of their existing CPU’s hence ignoring the lower tier/lower margin market. Could it be the only way to get any volume of production to market was to focus on a single cpu and they are just focusing on the one with the highest expected volume/profit margin.
    Reply
  • Makaveli
    PCWarrior said:
    So after all these headlines over 6 months ago about 3D Vcache AMD will launch just ONE cpu and that in 2-3 months from now. And that is not even the 12-core they demoed back then but a mid-tier octacore with lower clocks than its predecessor! Talking about too little too late. As for the games where they showed it winning against the 12900K, these happen (surprise surprise) to be amongst the games where the stock configuration of the 12900K (with the e-cores enabled) is lower than with the e-cores disabled. Disabling the e-cores (or assigning gaming to p-cores only which can be done in software) brings parity in these games. I suspected that something was off with this whole Zen3D Vcache launch with the lack of leaks. But I could never predict such a fiasco.

    There is a reason they only going with 8 core model for now.

    Why only Ryzen 7, not Ryzen 5/Ryzen 9? Because this is a new project for AMD, they want to find out how a processor like this will be welcomed into the market. A lot of users (myself included) expected AMD to go all-in with a big 16-core version, however anything Ryzen 9 requires two chiplets, and adding the extra V-Cache does require an extra cost in silicon and packaging. During a semiconductor shortage, I was told that this is the best way to get it into the hands of many people while also not in the super high-cost bracket. It also means one single unified 96 MB of L3 cache, without having to deal with two chiplets worth which might not be optimized immediately. Future versions of V-Cache on next-generation products may be expanded to other Ryzen members of the family.

    https://www.anandtech.com/show/17152/amd-cpus-in-2022-ces
    Reply
  • shady28
    sizzling said:
    Just playing devils advocate here. As AMD are having issues getting enough production of their existing CPU’s hence ignoring the lower tier/lower margin market. Could it be the only way to get any volume of production to market was to focus on a single cpu and they are just focusing on the one with the highest expected volume/profit margin.

    Well with one sku, it's probably about time to market, like they didn't have enough time to get a full line-up so they just did one chip.

    I seriously doubt they will have any volume to speak of to sell. They did this same thing with Zen 3, only 1M Zen 3 were made in 2020. That may sound like a lot, but it isn't. The PC market is like 300M per year. I totally expect this to be a repeat of that paper launch fiasco that the enthusiast press largely gave AMD a pass on.

    https://www.hardwaretimes.com/1-million-ryzen-5000-cpus-were-sold-in-q4-2020-thats-just-10-12-of-amds-7nm-capacity-at-tsmc-3-4-for-big-navi/
    Reply