Poll: Ryzen 7950X3D, 7900X3D, 7800X3D

[Iron Man]

Some more info fresh in https://videocardz.com/newz/amd-ryz...with-similar-single-core-performance-to-7950x

Unfortunately it doesn’t really show much other than the fact that reviewers are starting to get them in their possession.

 

[VinylRichTea]

Some more info fresh in https://videocardz.com/newz/amd-ryz...with-similar-single-core-performance-to-7950x

Unfortunately it doesn’t really show much other than the fact that reviewers are starting to get them in their possession.

I don't know, it seems slower in multi-core than a 7950x and similar single-core performance, it's not looking good for first impressions otherwise we'd be seeing some impressive 3dmark scores popping up immediately instead of Ashes and Geekbench. I'm betting the only one worth having is the 7800x3d due to having a full 3d cache and the 7900x3d and 7950x3d being patchwork 3d cache.

 

[Iron Man]

I don't know, it seems slower in multi-core than a 7950x and similar single-core performance, it's not looking good for first impressions otherwise we'd be seeing some impressive 3dmark scores popping up immediately instead of Ashes and Geekbench. I'm betting the only one worth having is the 7800x3d due to having a full 3d cache and the 7900x3d and 7950x3d being patchwork 3d cache.

I think the article said they don’t know what the memory configuration was which could have an effect as well so still too early to tell. Hopefully we will get more leaks soon with more detailed info.

 

[Meddling-Monk]

I think given what we know of the x3D chips, and especially the 7950X variant (only one CCD having the extra cache, unknown how games are supposed to know or able to call on that CCD to to process instead of other one given its usually handled by the OS, the lower power limit, etc) lower performance outside of games that (somehow) manage to use the cache part of the chip was always an expectation really.

The 7800X3D however, as said by others, I think is the one to watch overall. The 7900X3D and 7950X3D sounds like a cash grab that's not really needed, or not really useful "right now".

 

[Grim5]

How much faster is the 5800X3D vs the 5800X in this?

That number from Toms is crap. In reality this score is the same as many other 7900x with the same gpu posted as per the Reddit thread. I believe toms number looks stupid because they are comparing between different directx versions

We now have some numbers for the 7950x3d; I haven't double checked them but wccftech claims it's 10% slower in single and multithread than non X3d model

AMD Ryzen 9 7950X3D 3D V-Cache CPU Benchmarks Leak: 10% Slower In Multi-Thread & Similar Single-Core As 7950X

The first unofficial AMD Ryzen 9 7950X3D 3D V-Cache CPU benchmarks have leaked out within the Geekbench 5 database.

wccftech.com

 

[uscool]

slower in multi thread isnt really a suprise knowing the ccd with 8 cores cache on 7950x3d will have lower clock speed

 

[Deleted member 258511]

slower in multi thread isnt really a suprise knowing the ccd with 8 cores cache on 7950x3d will have lower clock speed

Single is a little disappointing though if this is the case.

 

[cliffy]

Single is a little disappointing though if this is the case.

If the single core score is slower then surely thats probably down to needing a windows update so the scheduler knows which CCD to use

 

[Deleted member 258511]

If the single core score is slower then surely thats probably down to needing a windows update so the scheduler knows which CCD to use

Yes most probably. The thing is, there are too many variables and not enough information to go on with these. At least we are starting to get something though. Not long now.

 

[delta0]

I don't know, it seems slower in multi-core than a 7950x and similar single-core performance, it's not looking good for first impressions otherwise we'd be seeing some impressive 3dmark scores popping up immediately instead of Ashes and Geekbench. I'm betting the only one worth having is the 7800x3d due to having a full 3d cache and the 7900x3d and 7950x3d being patchwork 3d cache.

I’m not sure that’s a negative. There are 2 CCDs and neither of them are as quick as a 7950X in non-3D cache applications.

What do you mean by patchwork 3D cache and full 3D cache? They all have the same 3D cache.

 

[Meddling-Monk]

I’m not sure that’s a negative. There are 2 CCDs and neither of them are as quick as a 7950X in non-3D cache applications.

What do you mean by patchwork 3D cache and full 3D cache? They all have the same 3D cache.

I think they mean that the 7950X3D has the extra 3D Cache, but it's bolted onto only one of the CCD. So both CCD do not have the same amount of L3 Cache. One has more, one less. So this is more "patchwork", and not like on the 7800X3D where it is all on the one chip. This also means there's no need to get ways to have the correct cores (that have the extra cache) process for the game (on the 7900X3D and 7950X3D), as it's universal on the 7800X3D.

 

[delta0]

I think they mean that the 7950X3D has the extra 3D Cache, but it's bolted onto only one of the CCD. So both CCD do not have the same amount of L3 Cache. One has more, one less. So this is more "patchwork", and not like on the 7800X3D where it is all on the one chip. This also means there's no need to get ways to have the correct cores (that have the extra cache) process for the game (on the 7900X3D and 7950X3D), as it's universal on the 7800X3D.

Cache sizes
7950X3D - 144 MB (64 MB CCD, 64 MB V-Cache + 16 MB L2)
7900X3D - 140 MB (64 MB CCD, 64 MB V-Cache + 12 MB L2)
7800X3D - 104 MB (32 MB CCD, 64 MB V-Cache + 8 MB L2)

These are the caches for each CPU. The 7800X3D has zero advantages here. The 7950X3D can run 3D cache programs like a 7800X3D can but also has another CCD that can provide 8 cores to support non 3D cache processes. It’s far more versatile being able to support 3D cache and 16 cores at the same time.

 

[Meddling-Monk]

These are the caches for each CPU. The 7800X3D has zero advantages here. The 7950X3D can run 3D cache programs like a 7800X3D can but also has another CCD that can provide 8 cores to support non 3D cache processes. It’s far more versatile being able to support 3D cache and 16 cores at the same time.

I don't think what you posted is in dispute. What is is dispute is how exactly does the cores get assigned the "right" CCD for best performance? I don't think that has been explained (well) yet. As it doesn't look like it's using the Intel approach to P/E cores. And hence, because (currently) it looks like there's no way to determine which cores is called up to do what, you can just as easily have it in reverse where you have the normal (non VCache) CCD cores processing and end up with the VCache CCD core supporting it instead. Hence, it's more patchwork rather than on the 7800X3D where no matter which core is used, it's always one with VCache.

 

[Iron Man]

I don't think what you posted is in dispute. What is is dispute is how exactly does the cores get assigned the "right" CCD for best performance? I don't think that has been explained (well) yet. As it doesn't look like it's using the Intel approach to P/E cores. And hence, because (currently) it looks like there's no way to determine which cores is called up to do what, you can just as easily have it in reverse where you have the normal (non VCache) CCD cores processing and end up with the VCache CCD core supporting it instead. Hence, it's more patchwork rather than on the 7800X3D where no matter which core is used, it's always one with VCache.

This was discussed by AMD and Microsoft at CES 23 (article about it here https://www.techradar.com/news/micr...-7000-x3d-cpus-run-games-faster-in-windows-11).

As Microsoft have already done this previously for Intel and made windows 11 work for their mix of performance and efficiency cores I’m inclined to think that Microsoft can get this to work for the 7xxxX3D CPUs too.

 

[delta0]

I don't think what you posted is in dispute. What is is dispute is how exactly does the cores get assigned the "right" CCD for best performance? I don't think that has been explained (well) yet. As it doesn't look like it's using the Intel approach to P/E cores. And hence, because (currently) it looks like there's no way to determine which cores is called up to do what, you can just as easily have it in reverse where you have the normal (non VCache) CCD cores processing and end up with the VCache CCD core supporting it instead. Hence, it's more patchwork rather than on the 7800X3D where no matter which core is used, it's always one with VCache.

From what has been mentioned already AMD will ringfence processes that can be sent to the right CCD. They are doing this with Microsoft. I think the 7950X3D gives you the best of both worlds and more as it can run both CCDs at the same time.

 

[Meddling-Monk]

Thank you both, that is most informative. That was missing information for many of us I suspect, or at least, certainly for me.

 

[delta0]

As with all these things the benchmarks and real world performance will be the most interesting. I don’t think we will see the 7800X3D benchmarks soon but in theory someone could shutdown the non-3D cache CCD on a 7950X3D so that nothing can run on it and the performance should be similar/same.

 

[Quixote]

For me, the 7950X3D & 7900X3D have great potential for mixed - use systems.
If it works out how we hope, you can give up like 5% in productivity performance for top-of-the-tree gaming & cache-restricted performance.
Of course, there are a lot of 'ifs' there, so I'll be waiting for proper benchmarks & reviews.
One thing I'm hoping is that the 3D cache on Ryzen 7 will reduce the importance of good RAM, like it did on the Ryzen 5.
One thing productivity uses want is a lot of RAM (IMO 64GB is the 'standard' now).
It would be great if that can be cheap RAM with slack-ish timings, as we need more of it.
Standard Ryzen 7 is known to be more senstive to Ram speed & timings than Intel (for gaming stuff), so I hope the 3D cache versions can mitigate that.

 

[Fire_fly]

As with all these things the benchmarks and real world performance will be the most interesting. I don’t think we will see the 7800X3D benchmarks soon but in theory someone could shutdown the non-3D cache CCD on a 7950X3D so that nothing can run on it and the performance should be similar/same.

Back with Zen 1 a few websites started making their own simulated versions of unreleased Ryzen 5 CPUs by disabling core and tweaking the clock speeds. They got fairly close the to performance of the final CPUs doing this. With a bit of luck we should get an idea of 7800X3D performance relatively quickly.

For me, the 7950X3D & 7900X3D have great potential for mixed - use systems.
If it works out how we hope, you can give up like 5% in productivity performance for top-of-the-tree gaming & cache-restricted performance.
Of course, there are a lot of 'ifs' there, so I'll be waiting for proper benchmarks & reviews.
One thing I'm hoping is that the 3D cache on Ryzen 7 will reduce the importance of good RAM, like it did on the Ryzen 5.
One thing productivity uses want is a lot of RAM (IMO 64GB is the 'standard' now).
It would be great if that can be cheap RAM with slack-ish timings, as we need more of it.
Standard Ryzen 7 is known to be more senstive to Ram speed & timings than Intel (for gaming stuff), so I hope the 3D cache versions can mitigate that.

My experience of mixed use systems, ones where you can do two things at once i.e. video rendering and gaming, is that sometimes you can get a combination of programs that do weird things which don't get shown in benchmarks. I had an experience of this last year when I built a system with 12700 for a friend.

When doing some stability test on it though I would take the opportunity to compare it to my 3900X with a lengthy handbrake transcode. Everything was great with the 12700 when I set it going and it gave excellent performance, better than my 3900X. However, all that changed when I opened chrome to browse OCUK. Handbrake was immediately moved to the E cores, causing the performance to crater. Chrome was using a tiny fraction of the P cores which were all now mostly at idle. As soon as I closed Chrome Handbrake shifted back to the P cores. It made a system with a similar CPU completely unusable for what I need to do with my system. One the other hand the person that I built it for could not be happier with it as they don't run the same combination of programs that I do.

My main concern with the configuration of the dies in these CPUs is that we will get interactions between programmes both wanting the cores on the die with the extra cache and the scheduler not being smart enough to know if a the program actually needs the extra cache. It will probably lead to some very wired performance issues that won't show though in just pure single program benchmarks.

 

[Grim5]

slower in multi thread isnt really a suprise knowing the ccd with 8 cores cache on 7950x3d will have lower clock speed

Yep there is absolutely nothing interesting or surprising from any of these leaks so far, best to wait for real tests