AMD’s “together we advance_PCs”l ivestream presentation just wrapped up moments ago, where AMD CEO Dr. Lisa Su set the stage for the release of the next generation of AMD Ryzen desktop CPUs. Building off of AMD’s Ryzen 7000 announcement back at Computex 2022, the eagerly anticipated presentation laid out AMD’s launch plans for their first family of Zen 4 architecture-based CPUs, which will see AMD kick things off with a quartet of enthusiast-focused chips. Topping out with the 16 core Ryzen 9 7950X, AMD’s Ryzen 7000 chips will be launching in just over 4 weeks’ time, on September 27th, with AMD expecting to handily retake the performance crown across virtually all categories of the PC CPU space, from gaming to content creation.
Driving AMD’s gains in this newest generation of desktop CPUs is a combination of architectural improvements underpinning the Zen 4 architecture, as well as moving production of the CPU core chiplets to TSMC’s leading-edge 5nm process. The combination of which will allow AMD to deliver what they are saying is now a 13% increase in IPC over their Zen 3 architecture – up from an 11% claim as of Computex – as well as a sizable increase in CPU clockspeeds. The top-end Ryzen 9 7950X will have a maximum turbo clockspeed of 5.7GHz, 800MHz (16%) higher than the equivalent Ryzen 9 5950X. As a result, AMD expects to deliver a 29% generational increase in single-threaded performance, and even more in multi-threaded workloads.
Launching in conjunction with the new Ryzen 7000 series chips will be AMD’s AM5 platform, which will be cornerstone of AMD’s consumer desktop platform through at least 2025. AM5 introduces DDR5 and PCIe 5.0 support for AMD’s chips, as well as a new LGA socket. 4 chipsets are planned thus far – X670 Extreme, X670, B660 Extreme, and B660 – with X670 series boards available for the September launch, and B660 series boards set to follow in October.
Ryzen 7000 Family: 4 Zen 4 SKUs
For regular AnandTech readers and AMD followers, AMD’s product stack for next month’s launch should look very familiar. AMD has done very well for themselves by starting things off with an enthusiast-focused desktop launch that’s built around a group of four high-end processors, and for the Ryzen 7000 generation, AMD isn’t deviating from this.
As established by AMD back at Computex, Ryzen 7000 chips top out at 16 cores. So across AMD’s product stack, the core counts are the same in this generation versus the last. AMD’s top SKU will offer 16 cores, followed by 12, 8, and finally 6 CPU cores. And as before, AMD is building their chips using up to two Zen 4 CPU core chiplets (CCDs), each comprising 8 Zen 4 CPU cores.
|AMD Ryzen 7000 Series Processors
Zen 4 Microarchitecture
|Ryzen 9 7950X||16C / 32T||4.5GHz||5.7GHz||16 MB||64 MB||170 W||$699|
|Ryzen 9 7900X||12C / 24T||4.7GHz||5.6GHz||12 MB||64 MB||170 W||$549|
|Ryzen 7 7700X||8C / 16T||4.5GHz||5.4GHz||8 MB||32 MB||105 W||$399|
|Ryzen 5 7600X||6C / 12T||4.7GHz||5.3GHz||6 MB||32 MB||105 W||$299|
As an aside, AMD is not disclosing any further details about the integrated RDNA2 architecture GPU at this time. So while all of the Ryzen 7000 chips come with an iGPU, we still do not have specifications to speak of. Regardess, AMD has made it clear at multiple points that the iGPU for these desktop chips is a relatively small configuration for basic desktop work, and is not designed to be a high-performance GPU like on AMD’s APUs.
Ryzen 9 7950X: The Fastest CPU In the World
Leading the pack is the Ryzen 9 7950X. The very best of AMD’s Zen 4 architecture will use two fully enabled CCDs to deliver a total of 16 CPU cores. Thanks in large part to TSMC’s 5nm process, it comes with some very high clockspeeds as well; the base clockspeed alone is 4.5GHz, and the turbo clockspeed will reach up to 5.7GHz for a single thread. The latter is actually 200MHz higher than AMD’s Computex presentation, where the company hit just 5.5GHz.
AMD is advertising this chip as coming with 80MB of cache. That breaks down to 64MB of L3 cache – 32MB on each CCD – as well as 1MB of L2 cache paired with each Zen 4 CPU core. This is the same amount of L3 cache as found on the Ryzen 5000/Zen 3 family, but the amount of L2 cache per core has doubled over the previous generation.
The 7950X will also be AMD’s most power hungry mainstream desktop Zen chip yet. The TDP for the chip is 170W, and as we know from previous AMD disclosures, the Power Package Tracking (PPT) limit for these chips will be 230W. This is one of the major reasons that AMD is seeing such significant MT performance gains – exceeding their average ST performance gains – as previous Ryzen 5000 chips were often TDP limited when all of their CPU cores were under heavy load, leaving them well short of hitting the higher clockspeeds they’d otherwise be able to support.
Since AMD is the first CPU vendor out of the gate with this generation of parts, the 7950X doesn’t have any direct competition – though if you’re considering an Intel Core i9 12900K, AMD would be more than happy to upsell you on a 7950X instead. According to AMD’s figures the fully fleshed out Zen 4 chip should easily be the most powerful desktop CPU in the world, something we’ll of course need to confirm at review time, but is not hard to believe given AMD’s architecture updates and improvements brought by TSMC’s 5nm process. Still, a bit farther down the line it will eventually do battle with Intel’s Raptor Lake CPUs once those launch.
AMD is, a little surprisingly, pricing their new flagship chip at $699. This is actually $100 below the launch price of the 5950X back in late 2020, wich hit the streets at $799. The higher price of that chip was not especially well received, but none the less for quite a while AMD sold everything they could make. Still, the chip shortage has abated and AMD has more access to both wafers and substrates, and the company will not be going without competition for the entire life of the chip.
Ryzen 9 7900X
A step below the 7950X we have the Ryzen 9 7900X. This is another two CCD part, but with only 6 cores enabled on each CCD for a total of 12 CPU cores and 24 threads. This part also retains the full 64MB of L3 cache that comes with the 7950X, which is identical to how the L3 cache scales down (or rather, doesn’t) on the Ryzen 5000 series.
With its TDP set at the same 170W value as the 7950X, the 7900X actually gains a bit in base clockspeeds. This part will offer 4.7GHz at its base, 200MHz higher than the 7950X. However the maximum turbo clockspeed is 5.6GHz – a bit under the 7950X – so if you want the absolute best single-threaded performance, the 7950X is positioned and configured to be the fastest option for both ST and MT workloads.
Ryzen 9 7900X will be priced at $549, which is the same price as AMD’s previous 12 core part, the 5900X. So for a true apples-to-apples comparison on generational performance gains at a constant price, this will be one point to look at.
Ryzen 7 7700X
Going farther down AMD’s Ryzen 7000 product stack, we have the Ryzen 7 7700X. This is the 8 core Zen 4 counterpart to AMD’s 5800X in the previous generation. Like its predecessor, AMD is using just a single Zen 4 CCD here, but it is a fully-enabled CCD. Base clockspeeds will be 4.5GHz, while the maximum turbo clockspeed will top out at 5.4GHz.
Since it’s just one CCD, this also means that the L3 cache is halved to 32MB. Combined with the 8MB of L2 cache spread amongst the Zen 4 CPU cores, and you’ll see AMD advertising this chip as offering a total of 40MB of cache.
Meanwhile, unlike the 12 and 16 core chips, the 7700X will have a more traditional TDP of 105 watts. This makes the chip easier to power (and to cool), though it will be interesting to see what this means for turbo clocks under heavy MT workloads. With fewer CPU cores overall there’s less silicon to light up, but that means AMD also isn’t getting the additional electrical headroom to push clockspeeds higher.
The 7700X will hit the street with a $399 price tag, which like the 7950X, is a slight price drop over the previous generation (where 5800X was priced at $449).
Ryzen 5 7600X
Bringing up the bottom of the Ryzen 7000 launch stack is the sole Ryzen 5 class chip, the Ryzen 5 7600X. Like its predecessor this is a 6 core part, allowing for a total of 12 threads. Driving it is a single Zen 4 CCD with 2 of the CPU cores disabled.
The base clockspeed of the chip will stand at 4.7GHz (seeing a similar boost as the 7900X), while the turbo clockspeed will be the lowest of all of the initial Ryzen 7000 chips at 5.3GHz. And yet even as the slowest of the Zen 4 processors, AMD is boldly claiming that the 7600X should, on average, be 5% faster in gaming than Intel’s Core i9 12900K (never mind the cheaper chips).
This chip is seeing a TDP bump, however. Whereas AMD’s top 6 core chip in the previous generation was 65W, now even the bottom of the stack is starting at 105W. So this should not be considered a petite chip – AMD expects it to both draw and deliver quite a bit of power.
Zen 4 Architecture Tease: +13% Desktop IPC & AVX-512
Underpinning these new chips is of course AMD’s Zen 4 architecture. At just half an hour long, AMD’s presentation was not nearly lengthy enough to dive deeply into the Zen 4 architecture (AMD has to save something for the launch). But CTO Mark Papermaster did take the stage to offer a tease of sorts of additional technical details on the Zen 4 architecture.
First and foremost, AMD has revised their IPC estimate for Zen 4. As they have continued to tune the chips and firmware, in “typical desktop applications” they are now seeing a 13% average (geomean) increase in IPC versus Zen 3.
Locking things down at an iso-frequency of 4GHz, according to AMD they are seeing anywhere between a 1% gain and a 39% gain, depending on the workload. The net result is that, as is often the case, the exact performance gains vary with the task. Still, a lot of AMD’s benchmarks do come in fairly close to that 13% mark.
For this generation Papermaster also brought back one of our favorite diagrams, the IPC breakdown, showing how the various architecture changes to Zen 4 contributed to the overall IPC gain.
According to AMD, the biggest contributors in this generation are their new front-end , followed by load/store improvements, and then an improved branch predictor, execution engine, and the larger 1MB of L2 cache per CPU core. Specific technical details on what AMD has changed here will have to remain for another time, but Papermaster did let it slip that for the execution engine, the operand cache size has been increased by 50%; so it’s clear that AMD has been playing with their various caches by quite a bit at the CPU core level for Zen 4.
Papermaster also confirmed for the first time that Zen 4 – including the Ryzen 7000 series – will support AVX-512 instructions. AVX-512 is a bit of a mess of standards, so besides the foundation (AVX-512F) instructions, it’s still not entirely clear which subsets of AVX-512 AMD will support. But Papermaster did explicitly mention Vector Neural Network Instructions (VNNI) as among the additional subsets supported.
Critically, however, AMD is diverging from Intel in one important aspect: whereas Intel built a true, 512-bit wide SIMD machine for executing AVX-512 instructions, AMD did not. Instead, AMD will be executing these instructions over two cycles. This means AMD’s implementation still benefits from all of the additional instructions, register file space, and other technical improvements that came as part of AVX-512, but they won’t gain the innate doubling in SIMD throughput.
In discussing the rationale for AMD’s decision, Papermaster cited the extreme power requirements for a true 512-bit SIMD block as the biggest impetus for keeping AMD’s SIMD design at 256-bits. As we’ve already seen in Intel chips with AVX-512 support, the massive throughput of a 512-bit SIMD combined with its high density results in a hard spike in power consumption when using it, requiring Intel’s chips to downclock on AVX-512 workloads (sometimes severely) in order to keep power and thermals in check. Using a narrower 256-bit SIMD means that AMD won’t need to light up nearly as many transistors at once, which will in turn make it easier to keep clockspeeds and power consumption more consistent. At the same time, I don’t think AMD minds that the die space requirements for a 256-bit SIMD are significantly less than a 512-bit SIMD; a full 512-bit SIMD is a lot of transistors to build, and a lot of transistors to fire up during heavy workloads.
Finally, not to be understated is the impact TSMC’s 5nm process to AMD’s CPU cores and their resulting chiplets. AMD’s CCD chiplets have used TSMC 7nm for both the Ryzen 3000 (Zen 2) and Ryzen 5000 (Zen 3) generations – so AMD’s desktop CPUs have been on 7nm for over 3 years now. As a result, the shift to 5nm has been a long time coming, and the first time in quite a while that AMD has been able to enjoy a full node shrink for their desktop CPUs.
The benefits of 5nm are evident everywhere from clockspeeds to power consumption to die size. The improvements to clockspeeds in particular represent a major shift for AMD, whom a couple of generations ago faced a clockspeed deficit versus Intel, and as a result limited AMD’s overall single-threaded performance. Now AMD is looking at their top chips going well over 5GHz, delivering a meaningful clockspeed boost on top of their 13% IPC gains – for a total of a 29% gain in single-threaded performance.
Meanwhile, despite AMD’s architectural additions to Zen 4 – including the double-sized 1MB L2 cache – a Zen 4 CPU core plus its associated cache is smaller than a Zen 3 core and its 512KB of L2 cache. According to Papermaster, the resulting Zen 4 core is 18% smaller than its predecessor, measuring 3.84mm2 in size. And, being ever so cheeky, he also threw in a comparison to the size of Intel’s Golden Cove core, which is found on their Alder Lake processors. According to AMD’s figures, a Zen 4 core is outright half the size of a Golden Cove core, which gives AMD a significant advantage in die size (and thus at some level, manufacturing costs).
We’ve known for some time that TSMC’s 5nm process has a significant lead in density versus the Intel 7 process, and that is undoubtedly playing a huge part in AMD’s resulting size advantage. None the less, as AMD is trying hard to curate the idea of Zen 4 being a leadership-class design, area efficiency is one more aspect that AMD is competing with – and can hold over Intel’s head.
Finally, in a passing comment, Papermaster also disclosed the size of the metal stack AMD is using. Altogether, AMD is using a 15 layer telescoping stack.
Zen 4 vs. Zen 3: A Performance Preview
So between their architectural changes and their access to TSMC’s 5nm process, what does all of this mean for AMD? In short, the company is quite boldly proclaiming that they have the fastest desktop CPUs on the planet, and that the Ryzen 7000 chips will be delivering significant improvements over both Intel’s current chips as well as AMD’s previous-generation Ryzen 5000 parts.
|AMD Desktop CPU Generations|
|CPU Architecture||Zen 4||Zen 3||Zen 2|
|CPU Cores||Up To 16C / 32T||Up To 16C / 32T||Up To 16C / 32T|
|CPU PCIe Lanes||24x PCIe 5.0||24x PCIe 4.0||24x PCIe 4.0|
|Manufacturing Process||CCD: TSMC N5
IOD: TSMC N6
|CCD: TSMC N7
IOD: GloFo 12nm
|CCD: TSMC N7
IOD: GloFo 12nm
And while all first-party benchmarks should be taken with a grain of salt, here is what AMD is claiming for power and performance.
We’ll first start with iso (like-for-like) comparisons between Ryzen generations, with iso-power.
Interestingly, AMD offered performance figures for three different TDPs: 65W, 105W, and 170W. The greatest performance gains are actually at the lowest TDPs, where the 7950X saw a 74% increase in Cinebench R23 MT performance. These advantages actually decreased as TDPs went up, dropping to 37% at 105W, and finally 35% at 170W.
This is meant to underscore how AMD’s focus on energy efficiency, combined with TMSC’s 5nm process, has delivered significant gains across the board, but especially at lower TDPs. When AMD isn’t loading CPU clockspeeds into the stratosphere – which is always well into the diminishing returns of the voltage/frequency curve – Zen 4 is significantly more power efficient than its predecessor. And even when AMD dose go for broke on clockspeeds, Zen 4 still comes out well ahead.
On the whole, AMD is claiming that Zen 4 will deliver up to 49% more performance than Zen 3 at iso-power. Note that this is an “up to” figure and not an average. But it goes to show what Zen 4 should be capable of when it gets to put its best foot forward.
Meanwhile, taking things from the other direction on Cinebench, AMD also provided iso-performance figures. According to the company, Zen 4 (7950X) consumes up to 62% less energy for the same level of performance as a Zen 3 chip (5950X). And while AMD is going to be plowing most of those gains back into improving performance (clockspeeds) for these Ryzen 7000 desktop parts, it’s a further example of what AMD is touting to be significant efficiency gains.
As for how things stack up against arch-rival Intel, AMD is offering similarly bullish figures, albeit with an extra degree of cherry picking. In Chaos V-Ray, for example, AMD is claiming that the 7950X offers 47% better performance per watt than the 12900K.
And in gaming workloads, the company says that even the (relatively) diminutive 6-core 7600X will outperform the 12900K by an average of 5%. Which to be sure, that doesn’t have AMD winning everything, but a 6 core configuration is about as close as you can get to Zen 4 fighting with one hand behind its proverbial back.
Overall, AMD is claiming that the Ryzen 7000 series will deliver 11% better gaming performance than the 12900K, as well as 44% better performance in highly multithreaded content creation applications. Which, if this proves true in independent evaluation, certainly stands to put the Ryzen 7000 well into the lead for this forthcoming generation of PC processors.
Ryzen 7000: Coming September 27th
Bringing things to a close for now, AMD’s first step into the Zen 4 era will kick off on September 27th. That is when all four of the initial Ryzen 7000 CPUs will go up for sale, as will the first wave of the associated AM5 platform motherboards.
According to AMD, while initial launch logistics are always an interesting challenge (especially now as shipping is still a little wacky), the overall supply of Ryzen 7000 chips is expected to be good. While we’re rather accustomed to seeing launch-day allocations of new high-end hardware sell out, the overall message from AMD is that Ryzen 7000 chips should be plentiful. So unlike the Ryzen 5000 launch, which was very close to the peak of the chip and substrate shortage, Ryzen 7000 chips should be more regularly available. And AMD has invested in the wafer and substrate capacity to help ensure that happens.
Meanwhile on the motherboard front, the launch of AM5 boards will be spread over two months. For the initial September 27th launch, AMD’s board partners will have their high-end X670/X670E boards available. The more mainstream B650 boards – including the newly announced B650E chipset – will then be a month later. So potential buyers will have to weigh taking an X670 board at launch, versus waiting to grab cheaper B650 boards.
According to AMD, motherboard prices will start at $125. This is presumably for entry-level B650 boards without any PCIe 5.0 functionality. PCIe 5.0 will undoubtedly bring up the price on B650E and other boards, but at this point we don’t have a good answer as to just how much PCIe 5.0 will add to the production costs and retail prices of AM5 motherboards.
In conjunction with AM5 boards, the 27th will also see the release of the first factory overclocked memory kits specifically tuned for the Ryzen 7000 series. For this generation AMD will be offering the EXtended Profiles for Overclocking (EXPO) standard, which is AMD’s counterpart to Intel’s XMP DIMMs when it comes to pre-programed profiles for factory overclocked memory. AMD expects at least 15 EXPO-enabled memory kits to be available at launch, at speeds up to DDR5-6400.
And with that, the Zen 4 die is cast. AMD is aiming for nothing less than a leadership position in this generation, and come September 27th, we’ll get to see first-hand how well AMD’s engineers can deliver on the company’s ambition. If nothing else, after a pandemic, a chip supply shortage, and more, it’s going to be nice to get back to a more regular CPU launch in the PC space.