Saturday, January 8th 2022
Two AMD Ryzen 7000 Series Processors Based on Zen 4 Core Appear: 16-Core and 8-Core SKUs
AMD's Ryzen 7000 series of desktop processors based on the novel Zen 4 core architecture are scheduled to arrive in the second half of 2022. While we are not sure just how big the architectural differences will be going from Zen 3 (with or without 3D V-cache) to the new Zen 4 core, we have some leaked information that confirms the existence of two SKUs that reveal additional details about the processor configuration. In the MilkyWay@Home project, aiming to create a model of the Milky Way galaxy by utilizing countless PCs across the globe, we found two next-generation Ryzen 7000 SKUs. The MilkyWay@Home project isn't a benchmark. However, it is a valuable reference where the next generation processors appeared.
First in line is the 100-000000666-21_N CPU, a codename for an eight-core, sixteen-threaded design. This model should correspond to the AMD Ryzen 7 7800X CPU, a successor to the Ryzen 7 5800X model. Next in line is the 100-000000665-21_N CPU with 16 cores and 32 threads, a successor to the Ryzen 9 5950X named Ryzen 9 7950X. One important thing to note is that these new CPUs feature different level two (L2) cache configurations. With the previous generation 5000 series "Vermeer" processors, the L2 cache was locked at 512 KB per core. However, according to today's leak, the upgraded Zen 4 IP will bring 1024 KB of L2 cache per core, doubling the cache size at one of the fastest levels.
Sources:
BenchLeaks, via Tom's Hardware
First in line is the 100-000000666-21_N CPU, a codename for an eight-core, sixteen-threaded design. This model should correspond to the AMD Ryzen 7 7800X CPU, a successor to the Ryzen 7 5800X model. Next in line is the 100-000000665-21_N CPU with 16 cores and 32 threads, a successor to the Ryzen 9 5950X named Ryzen 9 7950X. One important thing to note is that these new CPUs feature different level two (L2) cache configurations. With the previous generation 5000 series "Vermeer" processors, the L2 cache was locked at 512 KB per core. However, according to today's leak, the upgraded Zen 4 IP will bring 1024 KB of L2 cache per core, doubling the cache size at one of the fastest levels.
32 Comments on Two AMD Ryzen 7000 Series Processors Based on Zen 4 Core Appear: 16-Core and 8-Core SKUs
Guess we will have to wait and find out. Either way looking forward if AMD has some serious brass in its pockets with the 7000 series CPUs.
Doubling the cache mem would boost things up a notch.
The fact its only being released on the gaming focused 5800X, suggests its something that AMD is only doing to overtake(?) the 12600K, 12700K & 12900K in gaming benchmarks, until they're ready to launch the new 7000 series in Q3 or Q4.
AMD will not like Intel being able to say they have the fastest gaming CPU in their marketing for most of 2022...It just gives them a free pass to gain market share, if the 3D cache 5800X can match (or beat) Alder Lake for gaming, 3D cache will have served its purpose.
So it would make sense that AMD do not use it for Zen4, it's expensive, hot/inefficient, and difficult to get linear scaling across multiple CCDs.
More cache in general on the 5nm process would be the superior solution.
The advantage is there, but from what I've heard it's mostly high end gaming, and if gaming performance is CPU limited.
Only a small amount of systems will end up like that, and all the mid range gaming machines, boring home and office machines won't benefit from it.
Can I ask where are you getting this stuff from?Really? 3Dcache has done its part? It is not even out yet. Since when 5800x is a gaming focused CPU?
Intel, Alder Lake has matched AMD in performance and core count. The Big.Little has done it's part. Move to 4core set up now. Thank you
Do you realize how silly you sound now?
I don’t think power or price is the main issue. I believe AMD is allocating 90%+ of production to the Epyc lines.
So I guess what I’m saying is there are no technological or pricing hurdles to offer a two chiplet 3d cache solution. The problem is most likely production allocation which no manufacturer likes to admit too. There just isn’t enough 3d cache chips for the desktop given the much higher priority and more lucrative enterprise market that Milan-X serves. Plus these cache chips are made using TSMC N7 node which is super constrained at the moment.
If this is indeed the case, then future Zen 4 chips can have 3d cache on one or more chiplets should manufacturing constraints ease.
When it comes to server workloads, licensing costs may of course matter, but this probably have more to do with some edge cases where they expect some significant performance gains. While in general cases such an increase in L3 cache may yield ~2-15% extra performance, this is irrelevant as servers are usually built with a very specific purpose in mind. It's quite possible that some edge cases may receive 2-3x performance improvements from this. You'll be surprised of how many companies relies on some business-critical custom software, developed by overpaid consultants, resulting in some horribly bloated Java code. L3 is a spillover cache from L2, and contain recently evicted L2 caches, and chances are that it will be mostly instruction cache lines that get a hit here. So increased L3 will mostly help cache misses in code, which is associated with very bloated code.
I actually like the approach of modular CPU features, and I hope this technology gets to a point where the cores themselves can be extended with another layer on top.Heh ;)
For all of those who think chopping off 1-2 ms at the end of the latency chain matters, especially for a game with a real tick rate of 30 Hz (120 Hz interpolated locally). :P
The Enterprise version benefit great from huge and large caches. It's not just to cover the latency but to speed things up significant. It's one approach to the more complicated laws here.
PS: The timing of the early announcement last year may have been prompted by Intel's moves though.
That $1000 chip you're talking about (or rather $800 as we're not quite back at old school pricing yet) will have lower priced counterparts at $300 - 600, and those are the ones I'm talking about. Rather than selling a 3D cache chiplet for those prices, AMD would instead put them in a $10,000 EPYC CPU, because money. Production allocation issues drives up cost.Do you have a source for that?
AMD is only just now considering AM5 sockets.