9900X3D - Will AMD solve the split CCD issue

[mama]

I just saw another discussion start about how many threads one needs for gaming on the 9000 series. An issue which has been much discussed in respect of previous and current CPU generations. But undoubtedly the 7900X3D suffered against its 3D brethren for having only 6 3D cores. Will this generation be any different?

 

[oxrufiioxo]

When steve tested a theoretical 7600X3D it performed similar to the 7900X3D so being dual CCD probably mattered less than it just having 2 less cores. There still was a net gain over a single CCD 6 core vcache Ryzen part though.




Most of this is just going to come down to the games tested some behave really well others do not and if you test older games the results may swing one way or the other....

You see a similar thing with 14th generation in some games if you limit it to 6 cores....



lets be real though all these CPU perform really well and most end users would likely be unable to tell them apart with actual settings normal people use.

 

[SL2]

Fever 3D cores (6) isn't the way to go above the 8 core 7800X3D and its successor. It doesn't make any sense from a gaming standpoint and AMD knows it (they didn't send 7900X3D's to reviewers lol).

Did some brainstorming before, but yeah, you can always argue that it's too close to the 16 core model.

If 6 + 6 is a disappointment, and 8 + 4 is such a waste of sand, well then maybe AMD should have gone for 8 + 6 and called it 7920 X3D.

 

[Assimilator]

First, you need to explain why an irrelevant handful of FPS lower makes AMD's CPU design an "issue".

 

[Onasi]

First, you need to explain why an irrelevant handful of FPS lower makes AMD's CPU design an "issue".

Because every single user runs a 4090 and all they do is play games all day, compare CPUs and meticulously note when they see the vast, stark difference between 160 and 170 FPS in their PlayStation-studio made slow walking cinematic title du jour. It’s simply unacceptable and AMD can’t keep getting away with this. /s

 

[Assimilator]

Because every single user runs a 4090 and all they do is play games all day, compare CPUs and meticulously note when they see the vast, stark difference between 160 and 170 FPS in their PlayStation-studio made slow walking cinematic title de jour. It’s simply unacceptable and AMD can’t keep getting away with this. /s

God forbid they just play the game and enjoy it, instead of anger-wanking about things that have zero impact on their experience with said game. It's the same kind of brainrot as those who claim they will literally die if they don't play a game on a high-refresh-rate display.

 

[Vya Domus]

This is hardly an issue, just people spreading misinformation.

 

[natr0n]

Game engine dependent on how many cores will be used.

Researching irrelevancy really.

 

[DemonicRyzen666]

like what make a 3D V-cache bridge over to the other CCD?
that would just cause cache thrashing.

 

[SL2]

like what make a 3D V-cache bridge over to the other CCD?
that would just cause cache thrashing.

Yeah, I guess the only solution is making CCD's with more cores, but then again, mote than 8 is not really needed yet, so..

That guy from AMD that looks like Stephen Root's relative said something intriguing, but otoh that's his job.

 

[kapone32]

I know people are going to come at me but let us look at it from another tangent. Is the 5800X3D faster in feel than the 5900X in daily use?t Be honest. Look at the reviews, because of Vcache the clock is not as high. What the 7900X3D is is an answer to people like me that wanted the Gaming goodness of Vcache with the everyday snappiness of Dual CCD. It is not the issue that the narrative made. The craziest thing I have seen on TPU was that even staff were willing to spout the BS about the latency and how weak the CPU was and how the 7800X3D is a better choice when they did not have a sample and were not willing to buy one. So I actually bought a 7800X3D for myself and do you know what I found? It is slower in daily use than a 7900X3D so it was sold the next day. I guess cores do matter after all. The narrative has even made it even more attractive. With today's chips paying $5 more for 4 more cores and 8 more threads is worth it. If all you do is open Steam and play Games all day the 7800X3D is fine but no one who has a 7900X3D regrets buying it.

 

[JWNoctis]

I know people are going to come at me but let us look at it from another tangent. Is the 5800X3D faster in feel than the 5900X in daily use?t Be honest. Look at the reviews, because of Vcache the clock is not as high. What the 7900X3D is is an answer to people like me that wanted the Gaming goodness of Vcache with the everyday snappiness of Dual CCD. It is not the issue that the narrative made. The craziest thing I have seen on TPU was that even staff were willing to spout the BS about the latency and how weak the CPU was and how the 7800X3D is a better choice when they did not have a sample and were not willing to buy one. So I actually bought a 7800X3D for myself and do you know what I found? It is slower in daily use than a 7900X3D so it was sold the next day. I guess cores do matter after all. The narrative has even made it even more attractive. With today's chips paying $5 more for 4 more cores and 8 more threads is worth it. If all you do is open Steam and play Games all day the 7800X3D is fine but no one who has a 7900X3D regrets buying it.

Less cores to contend with the same chunk of cache and chiplet link, more cores in total, and higher boost clock (especially on non-V-cache chiplet) to boot. If only looking bad on benchmarks had the same effect on local prices as wherever you got yours.

I'd just say that both are nice kits.

 

[tabascosauz]

I just saw another discussion start about how many threads one needs for gaming on the 9000 series. An issue which has been much discussed in respect of previous and current CPU generations. But undoubtedly the 7900X3D suffered against its 3D brethren for having only 6 3D cores. Will this generation be any different?

I think so far the only indication anything will change has been unconfirmed rumors of Zen 6 implementing a new 2.5D solution. Doesn't seem very likely to show up in an incremental update like 9000 series.

It might be interesting to see what happens to Vcache this generation though. Last gen, Vcache node didn't change but the design was forced to, in order to fit - now it's another node shrink. Sounds plausible that they either had to stack like-on-like, find a new node for Vcache, or somehow reshuffle the N7 design for a 2nd time

I don't remember seeing anything about the 7900X3D being noticeably behind just because it's 6 cores.

 

[GerKNG]

X3D should only be V-Cache dies.

 

[SL2]

X3D should only be V-Cache dies.

Hey, I saw your old post on Gamers Nexus yesterday!

Double 3D doesn't solve every problem anyway AFAIK, the CPU is still divided into two CCD's, and reaching for the other CCD's cache will be a latency problem, even if it's larger.

 

[tabascosauz]

Hey, I saw your old post on Gamers Nexus yesterday!

Double 3D doesn't solve every problem anyway AFAIK, the CPU is still divided into two CCD's, and reaching for the other CCD's cache will be a latency problem, even if it's larger.

Agreed, it would just be more cost for even less gain. At least currently you can slap a CCD1 binned for freq and let AGESA/scheduler get you a bit more clock outside of games. 2 Vcache CCDs is just......still choosing between 2 Vcache CCDs lol

It could benefit those other niche workloads that love cache though

like what make a 3D V-cache bridge over to the other CCD?
that would just cause cache thrashing.

I'm pretty sure a faster interconnect in Zen 6 might make that redundant anyway. You don't see the MCDs on Navi31 needing an individual connection to each other, it's a fast enough solution.

 

[Onasi]

Double 3D doesn't solve every problem anyway AFAIK, the CPU is still divided into two CCD's, and reaching for the other CCD's cache will be a latency problem, even if it's larger.

Supposedly, AMD is working on minimizing the between-CCD latency for Zen 6 to a point where it might not be a concern anymore. That’s the rumor anyway, so we’ll see. I do agree with @GerKNG that the way they implemented the V-cache on 7900/7950X3D was clumsy and, for most people, probably made those CPUs more hassle than it’s worth. Personal opinion applies here, obviously.

 

[JustBenching]

I know people are going to come at me but let us look at it from another tangent. Is the 5800X3D faster in feel than the 5900X in daily use?t Be honest. Look at the reviews, because of Vcache the clock is not as high. What the 7900X3D is is an answer to people like me that wanted the Gaming goodness of Vcache with the everyday snappiness of Dual CCD. It is not the issue that the narrative made. The craziest thing I have seen on TPU was that even staff were willing to spout the BS about the latency and how weak the CPU was and how the 7800X3D is a better choice when they did not have a sample and were not willing to buy one. So I actually bought a 7800X3D for myself and do you know what I found? It is slower in daily use than a 7900X3D so it was sold the next day. I guess cores do matter after all. The narrative has even made it even more attractive. With today's chips paying $5 more for 4 more cores and 8 more threads is worth it. If all you do is open Steam and play Games all day the 7800X3D is fine but no one who has a 7900X3D regrets buying it.

Completely objectively, a 7900x 3d with 8 3d + 4 normal cores would just be flat out better than the current 6+6 Frankenstein. That is the issue. Yes obviously the 7900x 3d is faster than the 800 x3d in daily usage, but with a better configuration it could have been better for games too.

 

[Onasi]

@fevgatos
That’s true, but it’s economics that are in play for AMD, I think. 6+6 allows the use of two “scuffed” CCDs. 8+4 would require one to be “perfect”. And, seeing how there are no 4 core Raphael CPUs, I don’t think that AMD HAS anything to direct into making a 4-core CCD. So that would have to be a 6-core with two more artificially disabled. So the costs just go up.

 

[A Computer Guy]

Hey, I saw your old post on Gamers Nexus yesterday!

Double 3D doesn't solve every problem anyway AFAIK, the CPU is still divided into two CCD's, and reaching for the other CCD's cache will be a latency problem, even if it's larger.

Theoretically X3D with each CCD would minimally allow for performance consistency and prevents issues with non-homogenous core design when running applications. In other words process lasso and game mode optimizations to twiddle with the cores are not required. Perhaps not as important for your typical Excel user but when running VM's it might make a difference. I haven't seen any numbers on this yet which is why I said "theoretically" but it would be interesting to know if compute intensive VM's crossing between X3D and non-X3D cores are impacted by inconsistent performance. I believe AMD already tested dual X3D and determined it was a waste (or had no beneficial impact) for gaming.

 

[JustBenching]

@fevgatos
That’s true, but it’s economics that are in play for AMD, I think. 6+6 allows the use of two “scuffed” CCDs. 8+4 would require one to be “perfect”. And, seeing how there are no 4 core Raphael CPUs, I don’t think that AMD HAS anything to direct into making a 4-core CCD. So that would have to be a 6-core with two more artificially disabled. So the costs just go up.

Not in disagreement, im just explaining why people complain / don't like the current configuration of the 7900x3d.

 

[DemonicRyzen666]

I'm pretty sure a faster interconnect in Zen 6 might make that redundant anyway. You don't see the MCDs on Navi31 needing an individual connection to each other, it's a fast enough solution.

I thought that was because they were separated each by a 32bit memory controller that's linked to the/or couple of shader units or what they're calling "work groups" now?
I'm going to have to look a w1zzards review again.

:edit:
I see six shader engines & six 4x16 controllers.

 

[Bill_Bright]

Because every single user runs a 4090 and all they do is play games all day, compare CPUs and meticulously note when they see the vast, stark difference between 160 and 170 FPS in their PlayStation-studio made slow walking cinematic title du jour. It’s simply unacceptable and AMD can’t keep getting away with this. /s

God forbid they just play the game and enjoy it, instead of anger-wanking about things that have zero impact on their experience with said game. It's the same kind of brainrot as those who claim they will literally die if they don't play a game on a high-refresh-rate display.

This is why I don't like most benchmark programs. People don't know how to use them. Remember the Windows Experience Index (WEI) included in Windows? It was actually a great little basic (basic by intentional design) benchmarking program. It's still in there, BTW, but Microsoft removed easy access to it simply because users didn't know how to properly use it. And more importantly, they didn't understand the results. Instead of using it to set a baseline for their own computer, then compare before and after changes to that baseline value, they were using it to compare their system with other systems, then cry (and of course, blame Microsoft) when they didn't get the same or better score as someone else.

Using Onasi's example of 160 vs 170FPS, the 160 "sufferer" would be convinced there was something wrong with their systems when in fact there was nothing wrong. And they were so convinced, they even talked their minds into believing they could easily see those differences.

Yet it is highly unlikely it was only seen by the benchmark scores. Even in a double-blind, side-by-side, A-B tests with two identical systems, it is likely they would not actually see any differences.

And to that, EVERY computer becomes a unique computer within the first couple minutes after the very first boot - as networking, security, personalizations, etc. get setup by the user.

God forbid they just play the game and enjoy it

^^^THIS^^^ They can't because they are spoiled brats! I remember putting my Space Invaders "tape" into my cassette reader, going off to make and eat breakfast then coming back see the game just finish loading into my C-64 - then thoroughly enjoy playing for hours. Even Pong a few years before that, in black and white was entertaining.

Game developers today know most gamers cannot afford $1000 graphics cards and similarly priced CPUs, 256GB of RAM, etc. So they code their games to provide great "game play" with much lessor systems.

But of course, you can't have a good gaming experience if your best "frenemy" gets 3 more FPS than you.

 

[kapone32]

Completely objectively, a 7900x 3d with 8 3d + 4 normal cores would just be flat out better than the current 6+6 Frankenstein. That is the issue. Yes obviously the 7900x 3d is faster than the 800 x3d in daily usage, but with a better configuration it could have been better for games too.

That is another notion that makes no sense. If every single Game used Vcache then sure but why would you want to lose 2 cores that run at 5.6 GHz just to make it the same amount of cores on 1 CCD as the console that run at 5 Ghz. 6 cores in each is fine as it is because of that. If that was true the 7950X3D would blow the 7900X3D away in Games but it doesn't.

 

[JustBenching]

That is another notion that makes no sense. If every single Game used Vcache then sure but why would you want to lose 2 cores that run at 5.6 GHz just to make it the same amount of cores on 1 CCD as the console that run at 5 Ghz. 6 cores in each is fine as it is because of that.

Because the performance difference from downclocking 2 cores from 5.6 to 5ghz is near 0. Okay, maybe it's 3%. That's nothing. And im talking about fully MT workloads obviously.

If that was true the 7950X3D would blow the 7900X3D away in Games but it doesn't.

Of course it does. There are even games that the 7950x 3d is 30+ fps ahead. And that's just averages, in lows the difference will be bigger