That Epyc does not look like a good candidate for showcasing future gaming performance with V-Cache: it's 8 cores distributed on 8 chiplets.I have seen some gaming benchmarks with a Epyc 72F3 (8 cores, 32 MB per core, 256 MB total) vs. Intel Core i9-11900K; I don’t know how reliable they were though. The Epyc managed to win with a much lower clock, but not by a huge amount. The cpu can only do so much before you run into gpu bottlenecks.
Insightful look at the impact of cache vs core count on games:
Written article:
How CPU Cores & Cache Impact Gaming Performance
At some point you may have heard someone say that for gaming you need X amount of cores. Examples include "6 is more than enough cores," or...www.techspot.com
"When CPU limited in today's games, cache generally provides the largest performance gains and this is why we see less of a performance variation between the various Zen 3-based (Ryzen 5000 series) processors ranging from 6 to 16 cores. [...] AMD Zen 3 CPUs all feature 32MB of L3 cache per CCD, that's 32MB total cache in the case of the Ryzen 5 and Ryzen 7 parts, and Ryzen 9 getting a 64MB total broken into two separate dies. Intel CPUs though see a fundamental change in L3 cache capacity depending on core count. The 10th-gen 6-core i5 models get 12 MB of L3, 8-core i7’s get 16 MB, and the 10-core i9 20 MB."
View attachment 49017
View attachment 49018
PS. Interestingly, when AMD introduced the (much derided, and now most likely abandoned) marketing term "GameCache", they might have been preparing for the introduction of stacked L3 cache, now marketed as V-Cache.
I was involved in that discussion about 2 weeks ago or so. I kind of gave up on it. There are a bunch of technological arguments, but just because something is possible doesn’t mean that it will be a product. There doesn’t seem to be any room for the 4 high cache variant between top end Ryzen and Threadripper.
At this point, I suspect the 1 vs. 4 consumer product is irrelevant since I think the 4 layer part will be a later product. They will likely intro the single layer part first and the 4+ layer part will be quite a bit later, probably closer to Zen 4 launch. There may be a lot of overlap between Milan-X3D and Genoa. Genoa is a completely new platform; new socket, new memory, new interconnect, etc. Parts of the server market move very slowly. They spec a machine and then stick with it for years. It will take quite a bit of time for Genoa to actually achieve market penetration. There are still some early adopters in HPC and a few other markets, but a lot of AMD’s sales are likely to be Milan or Milan-X3D for quite a while, even after Genoa. A Milan-X3D even with just a single layer of stacked cache will dominate everything else for some applications. That could be up to 768 MB of L3.
I am still amazed that some very smart people think that AMD should create a whole new IO die for Zen 4, and release a CPU that would be competitive in MT (which is the Threadripper territory):
View attachment 49030
While people on this thread insist that AMD NOT release SKUs, that AMD NOT compete to win in ST, while releasing such CPU would be effortless in comparison.
I think there is a good case to be made on working on the Threadripper platform and Threadripper CPU pricing to address its high cost, making it a much smoother, more linear transition from regular Ryzen (up to 16 cores) to Ryzen Threadripper as an HEDT platform with > 16 cores.
Right now, the cost increment between 16 core Ryzen and 24 core Threadripper is too high, which in turn hurts the platform popularity, leaving it low volume.
Extra memory channels of the Threadripper platform would come in handy for highly multithreaded applications, which tend to be more memory bandwidth dependent vs. single thread performance apps, such as games, which tend to be more memory latency dependent, which can be addressed (lowered) with extra L3.
AMD could release some Milan X3D SKUs with fewer CCDs and V-Cache.
4 CCD Milan X3D with up to 32 cores should (in nearly all cases) gain performance vs. 8 CCD equivalent. So it could lower AMD cost and increase performance.
I would be shocked if AMD released new core configurations for Ryzen at all.
I suspect AMD secretly wants to merge Threadripper with EPYC.
People have stopped bothering to argue with you on this since you are making up straw-man arguments. In your mind, AMD is giving up on single thread performance if they don’t or can’t release a 288 MB L3 cache variant for the consumer market? No one said or implied that other than you. I personally think they will do fine with 96 MB. A larger cache variant Ryzen part would break market segmentation and doesn’t have any place in the stack. What would it be priced at vs. threadripper 5000 parts that could have high cache per core size without X3D?I am still amazed that some very smart people think that AMD should create a whole new IO die for Zen 4, and release a CPU that would be competitive in MT (which is the Threadripper territory):
View attachment 49030
While people on this thread insist that AMD NOT release SKUs, that AMD NOT compete to win in ST, while releasing such CPU would be effortless in comparison.
I think there is a good case to be made on working on the Threadripper platform and Threadripper CPU pricing to address its high cost, making it a much smoother, more linear transition from regular Ryzen (up to 16 cores) to Ryzen Threadripper as an HEDT platform with > 16 cores.
Right now, the cost increment between 16 core Ryzen and 24 core Threadripper is too high, which in turn hurts the platform popularity, leaving it low volume.
Extra memory channels of the Threadripper platform would come in handy for highly multithreaded applications, which tend to be more memory bandwidth dependent vs. single thread performance apps, such as games, which tend to be more memory latency dependent, which can be addressed (lowered) with extra L3.
AMD could release some Milan X3D SKUs with fewer CCDs and V-Cache.
4 CCD Milan X3D with up to 32 cores should (in nearly all cases) gain performance vs. 8 CCD equivalent. So it could lower AMD cost and increase performance.
People have stopped bothering to argue with you on this since you are making up straw-man arguments. In your mind, AMD is giving up on single thread performance if they don’t or can’t release a 288 MB L3 cache variant for the consumer market? No one said or implied that other than you. I personally think they will do fine with 96 MB.
A larger cache variant Ryzen part would break market segmentation and doesn’t have any place in the stack. What would it be priced at vs. threadripper 5000 parts that could have high cache per core size without X3D?
Anyway, it likely isn’t relevant since the more than 1 layer device probably isn’t going to be available in the same time frame. I know people trying to get Milan processors at the moment and they can’t get them. This is very bad in the server market when a company specs a machine, gets evaluation machines, and then can’t actually get the parts to build the production machines. Perhaps covid is to blame for the delays; I don’t know.
Threadripper 5000 isn’t going to be available until AMD can mostly satisfy Milan demand. Current rumors say threadripper 5000 availability will not be until November now. Milan-X3D (presumably 4+ layer devices) is rumored to not be available until sometime in 2022.
So, I guess from what you seem to believe, AMD is just completely SOL, since they are probably only going to have 96 MB L3 parts in the consumer market at the end of the year.
It has happened frequently that people in forums try to build up unreasonable expectations for upcoming AMD parts. I don’t know if this is just fanboyism or an attempt to make truly incredible AMD releases seem more like a disappointment. I am wondering which category you fall into.
AMD Zen 3 with v-cache will have 192mb and will be available on desktop and server only according to the hot chips presentation.
Yes, later date.Is there a date for this?
12:47PM EDT - Gaming was a main target for Zen 3
Funny how the "Dunning-Kruger crowd" (people who use the expression "Dunning-Kruger") didn't see it even in the rearview mirror, not to mention being able to make a projection about the future.
That's for 2 CCDs, IIRC.AMD Zen 3 with v-cache will have 192mb and will be available on desktop and server only according to the hot chips presentation.
Another straw man slain. . . congrats?
Bold of you to assume it will be present at all on single CCD parts. Remember, it is more expensive to make these chips. Margins are higher on the 5900X and 5950x.That's for 2 CCDs, IIRC.
The main thrust of my argument is that AMD will do what it takes to keep the Gaming Performance crown against Alder Lake.
You were of the opposite opinion.
I posted proof that AMD did it in the future:
12:47PM EDT - Gaming was a main target for Zen 3
Your reply is more sophistry.
Some rumors from Moore's Las is dead:
- Zen 3D will likely be called 5000 XT
I assume you mean AMD did it in the past, but either way, what you posted is not proof of your argument in any way.
It's pretty simple, I'll gladly admit I was wrong if AMD comes out with a 4+ high stack Zen 3d Ryzen product. Will you admit you were wrong if they don't?
I know what you mean. As a sophist would say: "past does not predict the future"
My prediction is conditioned on assumption (with low degree of confidence) that Zen 3D with just a single layer of V-Cache would not convincingly beat Alder Lake in gaming, therefore AMD will need to release > 1 to maintain leadership in gaming performance.
Alder Lake can turn out particularly bad in gaming, then all it would take is a single layer of L3.
As far as my assumption (with low degree of confidence) that Alder Lake will beat Zen 3 in gaming, that assumption got just a tiny bit stronger. Alder Lake is pretty good at something - it beats Zen 3 in Puget workstation benchmark:
Alder Lake Benchmark Results Surface, but Beware the Fine Print - ExtremeTech
Just a single layer still triples the L3 cache size, and the further layers very likely won't have the same effect on top.Zen 3D with just a single layer of V-Cache
I know what you mean. As a sophist would say: "past does not predict the future"
My prediction is conditioned on assumption (with low degree of confidence) that Zen 3D with just a single layer of V-Cache would not convincingly beat Alder Lake in gaming, therefore AMD will need to release > 1 to maintain leadership in gaming performance.
Alder Lake can turn out particularly bad in gaming, then all it would take is a single layer of L3.
As far as my assumption (with low degree of confidence) that Alder Lake will beat Zen 3 in gaming, that assumption got just a tiny bit stronger. Alder Lake is pretty good at something - it beats Zen 3 in Puget workstation benchmark:
Alder Lake Benchmark Results Surface, but Beware the Fine Print - ExtremeTech
Ok, but earlier you said the stacks should be relatively cheap and offer large % increases with every cache double. If this were the case, why not go 4 hi and just dominate from the beginning? They could easily wipe the floor with anything Intel put forward and do it for only a small increase in price. What constitutes a convincing win for Zen 3 here and at what point do they just say good enough or is it only good enough if they win convincingly?
What if AMD wins gaming with 1 stack but loses in other classic benchmark scenarios like rendering, encoding, and office work. Do you still think they should go 4+ then or is it just for gaming that 4+ is worth it?