Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[Geddagod]

I think we missed an.important part with V cache we can lower CPU Power so more power to GPU and if it's maxed lower total system power means less heat

OEMs aren't decreasing power to the 3D V-cache CPUs afaik

 

[511]

OEMs aren't decreasing power to the 3D V-cache CPUs afaik

Well OEMs are stupid in this regard no matter the vendor

 

[MS_AT]

https://videocardz.com/newz/amd-ryz...mb-of-l3-cache-and-200w-reportedly-on-the-way some people will be happy if this turns out to be true

 

[Josh128]

https://videocardz.com/newz/amd-ryz...mb-of-l3-cache-and-200w-reportedly-on-the-way some people will be happy if this turns out to be true

Only way this will really be newsworthy is if this SKU manages to unify both CCDs and L3s to form one 16 core coherent CCX. Apparently that is out of the question, so the use case for this consists of basically nothing outside of bragging rights.

That being the case, Im more curious about the new 8 core version. Would it be more likely that its a "9700X3D" that is clocked lower out of the box, say, 5.0 GHz, or a "9850X3D" that clocks higher? Out of the box 5.2GHz seems is already not that reliant on highly binned CCDs, so it would be interesting if they pull off something cool like a 5.5GHz out of the box boost. An "XT" style refresh if you will.

 

[ToTTenTranz]

I think there'd be more market for a Threadripper where one CCD got VCache than there will be for a Granite Ridge getting 2 CCDs with VCache each.

Only way this will really be newsworthy is if this SKU manages to unify both CCDs and L3s to form one 16 core coherent CCX. Apparently that is out of the question, so the use case for this consists of basically nothing outside of bragging rights.

And perhaps software video transcoding?

 

[Josh128]

I think there'd be more market for a Threadripper where one CCD got VCache than there will be for a Granite Ridge getting 2 CCDs with VCache each.






And perhaps software video transcoding?

View attachment 128215

That 4% advantage example is a super dubious claim for the benefits of X3D in transcoding. Thats margin of error stuff. I'd bet adding a second vcache die would not add another 4%, if anything at all. Certainly nothing worth the extra $100-$200 (15%-30%) that they will surely be asking.

 

[Win2012R2]

Well well well... should have done it a year ago, would have sold nicely

 

[yuri69]

This certainly seems like a better "filler SKU" than those "XT" models.

 

[gdansk]

Watch one be around 9800X3D speeds and the other be around 9950X3D speeds so you end up with a ~7% difference between CCDs anyway.
It's AMD. You know they want to.

 

[Josh128]

Watch one be around 9800X3D speeds and the other be around 9950X3D speeds so you end up with a ~7% difference between CCDs anyway.
It's AMD. You know they want to.

Almost certainly wont be both a 5.7 boost. They do that and this really will be the enthusiast halo approach, and the price will be eye watering.

 

[fastandfurious6]

Ryzen 9 9999X3DX2

 

[Hail The Brain Slug]

Ryzen 9 9999X3DX2

I was expecting 9955X3DX2

 

[Hail The Brain Slug]

The use case for this consists of basically nothing outside of bragging rights.

I've been doing zen 5 x3d vs zen 5 performance comparisons with Unreal - engine compiling, shader compiling, cooking assets, building HLODs, etc and the V$ seems to provide anywhere from 3-10% uplift. A dual V$ chip would provide some measurable improvement overall for this.

 

[igor_kavinski]

https://videocardz.com/newz/amd-ryz...mb-of-l3-cache-and-200w-reportedly-on-the-way some people will be happy if this turns out to be true

I will hate them with a vengeance if they do that now. I seriously will. I spent over $800 for my lame 9950X3D

For their sake, I hope they don't do that.

 

[Hail The Brain Slug]

I will hate them with a vengeance if they do that now. I seriously will. I spent over $800 for my lame 9950X3D

For their sake, I hope they don't do that.

I just ordered a second 9950X3D for a new unreal workstation. Amazon already charged me but I managed to still cancel it when I saw the news this morning. I'll rob the 9950X3D out of my gaming rig until this CPU materializes (or doesn't)

 

[Josh128]

This cant come soon enough. Quite a drought in discussable news lately.

 

[igor_kavinski]

This cant come soon enough. Quite a drought in discussable news lately.

Yeah, at least the benchmarks will be enlightening. But way to go crushing the spirits of your most ardent fans, AMD.

 

[eek2121]

I've been doing zen 5 x3d vs zen 5 performance comparisons with Unreal - engine compiling, shader compiling, cooking assets, building HLODs, etc and the V$ seems to provide anywhere from 3-10% uplift. A dual V$ chip would provide some measurable improvement overall for this.

Yep, it provides a bit of an uplift in certain developer workloads, gaming, and more. The dual CCD part likely won’t add much more performance, however the TDP is higher, so all-core clocks will also probably see a slight boost. It wouldn’t surprise me if it received it’s own numbering scheme (9955X3D or something similar)

MSRP will probably be around $999.

 

[Tuna-Fish]

9999X3D priced at $999?

 

[igor_kavinski]

My guess is $899.

 

[LightningZ71]

As I suggested the last three times this product came up, it should have been called a 4XXX series EPYC model. That's where it'll see the biggest gains, server and workstation class workloads.

 

[Hail The Brain Slug]

Yep, it provides a bit of an uplift in certain developer workloads, gaming, and more. The dual CCD part likely won’t add much more performance

All of the workloads I mentioned are embarrassingly parallel and should benefit to various extents from the second CCD having cache.

Of course, an extra 3-10% on top of a 9950X3D hardly justifies $1000, or $900, or even $800.

Heck, the 3-10% gain a 9950X3D has over a 9950X in my workloads doesn't justify the currently 30% price premium ($670 vs. $515)

However, I have found cases working with Unreal where the scheduling puts it on the frequency CCD when the workload actually would benefit measurably from the cache CCD, and in some cases I haven't found any way for me to fix that since it's not managed through thread affinity and process lasso would do nothing. Actually while writing this I remembered the UEFI should have a preference toggle for cache vs. frequency CCD, I should try that and see if it helps.

Edit: It did, but it didn't. Now it just round robins the threads between both the cache and frequency CCD. I suppose that probably gives the best uplift since now there's one worker per physical core.

 

[igor_kavinski]

Edit: It did, but it didn't. Now it just round robins the threads between both the cache and frequency CCD. God, this scheduling crap is dumb.

Do you have gamebar installed? Can you put the Unreal engine EXE on gamebar, making it think it's a game?

 

[Hail The Brain Slug]

Do you have gamebar installed? Can you put the Unreal engine EXE on gamebar, making it think it's a game?

I just realized this is the fastest way to do the work, 16 workers with one per physical core is faster than 16 workers on 8 cache cores. I should do some further testing to see how much having two cache CCD's would speed it up vs. having the scheduling improved this way.

 

[igor_kavinski]

As I suggested the last three times this product came up, it should have been called a 4XXX series EPYC model. That's where it'll see the biggest gains, server and workstation class workloads.

They will probably do both, consumer and Epyc 4005 models.