9900X3D - Will AMD solve the split CCD issue

[Assimilator]

my cpu is better than your cpu.... Dual CCD cpu suck thread..... Or at least that's what it has turn into.

This. Basically a stupid thread about a stupid non-issue has drawn in stupid people who are trying to support their stupid opinions with stupid benchmarks.

The rest of us are too busy actually playing games.

 

[AusWolf]

Does the 9900X3D have the same proportionally lower clock speed?

That's pretty much guaranteed as long as the extra cache is an extra heat insulation layer on top of the CPU die.

 

[SL2]

I guess most of you have seen it by now.

 

[AusWolf]

I guess most of you have seen it by now.
View attachment 350809

That adds another question on top of the ones you posted above: Will the increased L1 and L2 bandwidth make the 3D V-cache redundant?

 

[SL2]

That's pretty much guaranteed as long as the extra cache is an extra heat insulation layer on top of the CPU die.

Not necessarily, it improved from Ryzen 5000 (12% difference) to 7000 (7%), although it's debatable what it means. The set clock speeds doesn't show us what the CPU limits are anyway, or even if they're pushed the same amount.

Basically a stupid thread about a stupid non-issue has drawn in stupid people who are trying to support their stupid opinions with stupid benchmarks.

Yeah, it's hard to accept what makes a nerd tick from time to time.

 

[Sunny and 75]

I play at x FPS.

This^

This conversation so far makes me feel pretty good about pairing my future 9950x / 9900X3D with ECC RAM and RX 6700 XT. I won't have to worry about any of these high performance issues.

This^

actually playing games.

And this^

 

[Apocalypsee]

I guess most of you have seen it by now.
View attachment 350809

I like that AMD is increasing bandwidth of the caches (though slight latency increase) instead of focusing pure clockspeed, although still disappointed that they don't increase its size since they are on die shrink, and not increasing core count per CCD.

 

[JustBenching]

I like that AMD is increasing bandwidth of the caches (though slight latency increase) instead of focusing pure clockspeed, although still disappointed that they don't increase its size since they are on die shrink, and not increasing core count per CCD.

Core count per ccd increase is what all we intel diehards are waiting for. Ill be buying a 12core single ccd 3 d chip day one, amd forbid.

 

[Cowboystrekk]

I doubt you could get 10% improvement in your FPS by RAM tuning, but even if you do, it's still nothing by my standards. If you can't get at least 30-50% better (which is only possible with a GPU upgrade in most cases), then it's not worth bothering with, imo.


Nothing murders anything. Both the 7800X3D and 14900K are excellent CPUs for different reasons, and I don't think anyone buying either of them is unhappy with their choice.


If someone thinks that the tradeoffs of having 2 CCDs are greater than the benefits, then they don't need more than 8 cores to begin with. Simple as.


I couldn't notice any difference between the 7700X and 7800X3D at any RAM speed even when I still had my 7800 XT. I guess my standards are too low as well.

In certain cases you can get a lot more than 10% from ramtuning Depends on the game and cpu. In general the 3D cpus gain less from ramtuning than non 3Ds due to L3 cache outweighing it.

This is xmp/expo 6000 compared wirh tuned 6200 for ryzen and 7000 fpr intel. If you run default at 4800 difference will be much bigger.

 

[AusWolf]

In certain cases you can get a lot more than 10% from ramtuning Depends on the game and cpu. In general the 3D cpus gain less from ramtuning than non 3Ds due to L3 cache outweighing it.
View attachment 350943
View attachment 350944
View attachment 350945
View attachment 350946
This is xmp/expo 6000 compared wirh tuned 6200 for ryzen and 7000 fpr intel. If you run default at 4800 difference will be much bigger.

Let me guess... a 4090 was used during the test?

I like that AMD is increasing bandwidth of the caches (though slight latency increase) instead of focusing pure clockspeed, although still disappointed that they don't increase its size since they are on die shrink, and not increasing core count per CCD.

They are increasing the L1 data cache from 32 to 48 kb. What that does to performance, we'll see. If I had to guess based on Intel having made the same move from Comet to Rocket Lake, I'd say, not much.

 

[Cowboystrekk]

Let me guess... a 4090 was used during the test?


They are increasing the L1 data cache from 32 to 48 kb. What that does to performance, we'll see. If I had to guess based on Intel having made the same move from Comet to Rocket Lake, I'd say, not much.

Yes, 4090, but if you use dlss and run high refresh the difference can be noticable. Going from 60 to 70fps on 1% lows is noticable for many

 

[A Computer Guy]

Yes, 4090, but if you use dlss and run high refresh the difference can be noticable. Going from 60 to 70fps on 1% lows is noticable for many

There was a poll on that at TPU. (https://www.techpowerup.com/forums/...he-difference-past-60fps-gaming-other.322224/)
It's still open it seems if anyone hasn't voted yet.

 

[Cowboystrekk]

I seriously hope AMD either launches a 9900X3D with larhe cache on both ccds or ditch the 9900X3D altogether. The 7900X3D was ashitshow of a cpu failing to reach 7800X3Ds performance except for in productivity. A 7600X3D would have made more sense considering price.

There was a poll on that at TPU. (https://www.techpowerup.com/forums/...he-difference-past-60fps-gaming-other.322224/)
It's still open it seems if anyone hasn't voted yet.

Yeah, I'm not saying it's a lot, but tuning ram is a free upgrade that if done within safe margins nets you anywhere from 5-15% with little effort. About the same as upgrading from a 13600K to a 13900K in gaming, except it's free

 

[A Computer Guy]

I seriously hope AMD either launches a 9900X3D with larhe cache on both ccds or ditch the 9900X3D altogether. The 7900X3D was ashitshow of a cpu failing to reach 7800X3Ds performance except for in productivity. A 7600X3D would have made more sense considering price.


Yeah, I'm not saying it's a lot, but tuning ram is a free upgrade that if done within safe margins nets you anywhere from 5-15% with little effort. About the same as upgrading from a 13600K to a 13900K in gaming, except it's free

I don't understand the expectation. 7800X3D has 8 X3D cores for gaming, the 7900X3D has 6 X3D cores for gaming, of course the 7800X3D should do at least a bit better because 8 > 6 right?.

 

[JustBenching]

I don't understand the expectation. 7800X3D has 8 X3D cores for gaming, the 7900X3D has 6 X3D cores for gaming, of course the 7800X3D should do at least a bit better because 8 > 6 right?.

Well isn't that counterproductive? You are paying more for the 3d variant for the gaming performance, you sacrifice performance in other workloads and ocing to get that extra cache to help you in games and then you realize you don't really get that extra gaming oomph because it's just a 6 core chip for gaming.

Id wager the normal 7700x with some slightly tuned ram and / or ocing would end up faster than the 7900x 3d, which is rather absurd.

 

[wNotyarD]

I don't understand the expectation. 7800X3D has 8 X3D cores for gaming, the 7900X3D has 6 X3D cores for gaming, of course the 7800X3D should do at least a bit better because 8 > 6 right?.

The expectation comes from "bigger numbers equal better models, right?" philosophy.
I personally don't think the x900X3D (assuming a hypothetical 9900X3D is still 6+6) are bad, but have odd characteristics that flare up when more than 6 cores are required/taxed. If AMD makes the 9900 models to be 8+4, though, they'll become a lot better. I just wouldn't bet on it happening.

 

[SL2]

I seriously hope AMD either launches a 9900X3D with larhe cache on both ccds or ditch the 9900X3D altogether.

You don't have to buy it you know.

I still think AMD should go for 8+4 cores, or 8+6. People think 4 cores on a CCD is a waste, but so is selling the 7900X3D for lower than 7800X3D.

 

[A Computer Guy]

Well isn't that counterproductive? You are paying more for the 3d variant for the gaming performance, you sacrifice performance in other workloads and ocing to get that extra cache to help you in games and then you realize you don't really get that extra gaming oomph because it's just a 6 core chip for gaming.

Id wager the normal 7700x with some slightly tuned ram and / or ocing would end up faster than the 7900x 3d, which is rather absurd.

Well it's a new kind of offering that wasn't available before. You still get the total number of cores for good productivity - BUT - you do get X3D gaming cores advantage compared to non-X3D parts. If you play games that take advantage of the X3D cache it's a win-win scenario for productivity and gaming if the price is right. In theory anyway I see the best use case for 7900X3D is for dual use systems productivity/gaming at a much lower cost than 7950X3D.

 

[rv8000]

I doubt you could get 10% improvement in your FPS by RAM tuning, but even if you do, it's still nothing by my standards. If you can't get at least 30-50% better (which is only possible with a GPU upgrade in most cases), then it's not worth bothering with, imo.


Nothing murders anything. Both the 7800X3D and 14900K are excellent CPUs for different reasons, and I don't think anyone buying either of them is unhappy with their choice.


If someone thinks that the tradeoffs of having 2 CCDs are greater than the benefits, then they don't need more than 8 cores to begin with. Simple as.


I couldn't notice any difference between the 7700X and 7800X3D at any RAM speed even when I still had my 7800 XT. I guess my standards are too low as well.

Back when I was first testing the 7900X3D some imprecise manual tuning (only primaries and trefi) vs expo/xmp gave a 3% boost. Tack on a bump to IF and cp2077 was seeing a ~6% boost at 1080 low; all of this with very early agesa on AM5 before memory/IF improvements.

I don’t think its unreasonable to get a 10% boost with highly and properly tuned memory (6400) and sync’d IF.

 

[JustBenching]

Well it's a new kind of offering that wasn't available before. You still get the total number of cores for good productivity - BUT - you do get X3D gaming cores advantage compared to non-X3D parts. If you play games that take advantage of the X3D cache it's a win-win scenario for productivity and gaming if the price is right. In theory anyway I see the best use case for 7900X3D is for dual use systems productivity/gaming at a much lower cost than 7950X3D.

The normal non 3d cpus already offer plenty of gaming performance, it's not like they are slouches. So I'd imagine someone opting for the 3d means they want more than just good gaming performance, but Uber good, but the 7900x 3d kinda doesn't have. If the price is low, like it currently is, different story, but for the original asking prices, naaah.

 

[wNotyarD]

The normal non 3d cpus already offer plenty of gaming performance, it's not like they are slouches. So I'd imagine someone opting for the 3d means they want more than just good gaming performance, but Uber good, but the 7900x 3d kinda doesn't have. If the price is low, like it currently is, different story, but for the original asking prices, naaah.

Yesn't. It's the Schrödinger's processor: it has and hasn't uber good performance.
When it works it is amazing (on par with the 7800X3D and 7950X3D), and when it doesn't it is still at least as good as the non-X3D (and certainly faster than a 7600/7600X).

You can argue that for any value spent you don't want variance, it should be amazing all the time and I'd agree with you. But you can't say it outright doesn't have uber good performance because it does (albeit not always).

 

[rv8000]

Yesn't. It's the Schrödinger's processor: it has and hasn't uber good performance.
When it works it is amazing (on par with the 7800X3D and 7950X3D), and when it doesn't it is still at least as good as the non-X3D (and certainly faster than a 7600/7600X).

You can argue that for any value spent you don't want variance, it should be amazing all the time and I'd agree with you. But you can't say it outright doesn't have uber good performance because it does (albeit not always).

The 7900X3D is a good cpu, it’s just the 7800X3D and 7950X3D cover either end of their respective niche better. The primary reason to have bought one since the SKU was available was that you needed more than 8 cores, had multiple use cases (gaming, rendering, etc…), and didn't have the budget for the 7950X3D.

Dependent on thread assignment the 7900X3D and 7950X3D can both be faster than the 7800X3D due to having higher core frequencies on the vcache CCD. If you have a mixed workload, it really is a good CPU, even better with the price dropping through the floor. It’s mostly a matter of understanding your needs.

 

[A Computer Guy]

Yesn't. It's the Schrödinger's processor: it has and hasn't uber good performance.
When it works it is amazing (on par with the 7800X3D and 7950X3D), and when it doesn't it is still at least as good as the non-X3D (and certainly faster than a 7600/7600X).

You can argue that for any value spent you don't want variance, it should be amazing all the time and I'd agree with you. But you can't say it outright doesn't have uber good performance because it does (albeit not always).

So now if we try and take this and try to veer it back on topic what does this mean for the potential of 9900X3D/9950X3D and their non-dual CCD counterparts?

For starters if we take AMD's original research on 5950x dual X3D there was no advantage in dual X3D to my recollection from the GN video so I don't think dual X3D is going to happen anytime soon for consumer platforms unless there is a game changing technical or marketable use case for it such as "AI-X3D". Maybe for AM5 EPYC where there could be advantages on the server side especially when competing with Xeon.

If we look at data like HU collected (

) we can probably expect to see similar patterns of performance with the new parts in the lineup. If there isn't a decent enough uplift to be marketable the new X3D parts might find themselves in a particular situation where the price/performance ratio of the new parts aren't worth it and older lower priced 7000 X3D parts are just too good to ignore (even for a new build) in the same way many running 5800X3D users didn't feel the need to upgrade to 7800X3D. When I see results like this (

) 7800X3D doesn't look as great as it sounds especially if it's cost prohibitive to change platform.

 

[rv8000]

So now if we try and take this and try to veer it back on topic what does this mean for the potential of 9900X3D/9950X3D and their non-dual CCD counterparts?

For starters if we take AMD's original research on 5950x dual X3D there was no advantage in dual X3D to my recollection from the GN video so I don't think dual X3D is going to happen anytime soon for consumer platforms unless there is a game changing technical or marketable use case for it such as "AI-X3D". Maybe for AM5 EPYC where there could be advantages on the server side especially when competing with Xeon.

If we look at data like HU collected (

) we can probably expect to see similar patterns of performance with the new parts in the lineup. If there isn't a decent enough uplift to be marketable the new X3D parts might find themselves in a particular situation where the price/performance ratio of the new parts aren't worth it and older lower priced 7000 X3D parts are just too good to ignore (even for a new build) in the same way many running 5800X3D users didn't feel the need to upgrade to 7800X3D. When I see results like this (

) 7800X3D doesn't look as great as it sounds especially if it's cost prohibitive to change platform.

With limited application for things outside of gaming, there is no advantage. You’d be losing productivity performance on dual CCD X3D variants because your giving up 600-700mhz on the second CCD, while gaming remains limited to 8 cores or less.

 

[Zach_01]

I’d like to see a 10-12core single CCD/CCX 3D variant at some point if 3DV cache is a thing still in future.
But unless AMD needs this core count CCX on its server SKUs I don’t see it happening.
Because, we like it or not the Ryzen design is a server type one and happens to work well on PC platform and work better for games with 3DVcache addition. Enables AMD to have universal design, keep cost low and profit margins high.

Server needs drive mostly the development of these chips.

Personally I do not need productivity performance. Other than normal everyday tasks, I game but I do want to have higher than 8 cores for some VMs occasionally.
A x950X3D is the most appealing SKU to me but the cost is high. At least initially.

I’m a bit puzzled for what to get for my next upgrade. Not that I plan to switch soon. I guess I will decide it when the need for upgrade gets strong.