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

[DisEnchantment]

Speculate at will

 

[adroc_thurston]

It does seem like there is a manufacturing tweak / optimization rather than just better binning.

It is just bins.

 

[Josh128]

It is just bins.

And a year and a half of normal ongoing N4 process tweaks likely increasing yield over time, so they now have capacity to do a high volume part like this.

One half of the puzzle seems solved, now I wonder if the rumored 2X3D or X3D2 epeen halo will come to pass.

 

[adroc_thurston]

And a year and a half of normal ongoing N4 process tweaks likely increasing yield over time

Dawg this isn't 2020, N5 family ramped half a decade ago.

so they now have capacity to do a high volume part like this.

They always had but I mean why not pocket the margin.

 

[Josh128]

Dawg this isn't 2020, N5 family ramped half a decade ago.

Process improvement doesnt stop at ramp.

 

[adroc_thurston]

Process improvement doesnt stop at ramp.

Nothing to 'develop' here, it's mature as cow manure.
You're not magically gonna get better yield or whatever when ramping a second product family on a 5yo node.

 

[Tuna-Fish]

It's possible it's a new stepping that let them tighten it a bit, not on the process side but on the design side.

Could also be just binning. We won't know until someone gets one and reads the stepping out of it.

 

[jdubs03]

Nothing to 'develop' here, it's mature as cow manure.
You're not magically gonna get better yield or whatever when ramping a second product family on a 5yo node.

So it’s excrement now. 😂

 

[Josh128]

Doesnt matter bois, muh tuned Arrow Lake or Raptor Lake matches or is faster than all Zen 5 X3D series, tuned or not. Only the "smart people" know this. Only the "best" reviews reveal this fact, all other reviews showing otherwise are FUD.

https://twitter.com/x/status/1995115643956301989

 

[Joe NYC]

Doesnt matter bois, muh tuned Arrow Lake or Raptor Lake matches or is faster than all Zen 5 X3D series, tuned or not. Only the "smart people" know this. Only the "best" reviews reveal this fact, all other reviews showing otherwise are FUD.

https://twitter.com/x/status/1995115643956301989

Applying 1490 KiloFarad to Intel chip buy only feeding raisins to AMD chips...

 

[Joe NYC]

I strongly doubt this. What are you basing this on?

I was basing it on my hunch is that the clock speed deficit between 9700x and 9800x3d was not based on CCD binning but on something else, related to V-Cache.

MLID says that there is a new Zen 5 stepping, which would confirm what was only a hunch on my end. But their analysis seems to have missed missing the main point of no clock speed deficit due to V-Cache.

 

[LightningZ71]

Hey. I didn't say it was better. I said if you tune it to within an inch of it's life, and failing y-cruncher, and give it all of the watts, you can eek out a few wins on games with a working set that fits in the 2-3 MB range and pull reasonably close in several others.

 

[inquiss]

Hey. I didn't say it was better. I said if you tune it to within an inch of it's life, and failing y-cruncher, and give it all of the watts, you can eek out a few wins on games with a working set that fits in the 2-3 MB range and pull reasonably close in several others.

So whats the point. It's means nothing.

 

[DrMrLordX]

So whats the point. It's means nothing.

It's cool for enthusiasts to be able to do that, but let's be honest, if they put the same amount of effort into an AMD system then most of those wins would go away.

 

[Joe NYC]

9850x3d boxed

https://twitter.com/x/status/1996589022249099630

 

[ToTTenTranz]

9850x3d boxed

https://twitter.com/x/status/1996589022249099630

The absence of the 9950X3D2 is leaving me disappointed and a bit anxious.

 

[Gideon]

Geekbench score for the 9850X3D

COLORFUL CVN B850M GAMING FROZEN V14A - Geekbench

Benchmark results for a COLORFUL CVN B850M GAMING FROZEN V14A with an AMD Ryzen 7 9850X3D processor.

browser.geekbench.com

 

[CouncilorIrissa]

LENOVO 83Q8 - Geekbench

Benchmark results for a LENOVO 83Q8 with an AMD Ryzen AI 9 HX 470 w/ Radeon 890M processor.

browser.geekbench.com

Looks like the refresh will support LPDDR5X-8533.

 

[Joe NYC]

Geekbench score for the 9850X3D

COLORFUL CVN B850M GAMING FROZEN V14A - Geekbench

Benchmark results for a COLORFUL CVN B850M GAMING FROZEN V14A with an AMD Ryzen 7 9850X3D processor.

browser.geekbench.com

How is the multicore score for 9850x3d lower than most of the 9800x3d submissions?

 

[Tuna-Fish]

How is the multicore score for 9850x3d lower than most of the 9800x3d submissions?

Low memory clock?

 

[Joe NYC]

Low memory clock?

Yeah, looks like memory is DDR5 running at 4800. I wonder how much it affects the ST performance as well.

 

[Det0x]

How is the multicore score for 9850x3d lower than most of the 9800x3d submissions?

Something wrong / unoptimized with the setup if a 9850X3D loses to a 9800X3D

Looking at this from a purely overclocking / hwbot lens:

9850X3D is a superbinned 9800X3D, how much better the silicon quality on the CCD is, depend on what sample you are comparing it against.
The very best 9800X3D can probably match it / get close, but there are very few of this caliber out there in the wild.
Average sample vs average sample i would say + ~200-300mhz clockspeed once overclocked
(difference is larger when running stock 5.2ghz vs 5.6ghz like reviewers do)

So the 9850X3D is just a way to get very strong silicon, without binning hundreds of 9800X3D's
IO-die quality (IMC/FCLK) is still all random, so no guarantee there.

If your average user already own a 9800X3D it makes very little sense to buy a 9850X3D
These new cpus are for overclockers or ppl that are building new systems.

 

[Joe NYC]

Something wrong / unoptimized with the setup if a 9850X3D loses to a 9800X3D

Looking at this from a purely overclocking / hwbot lens:

9850X3D is a superbinned 9800X3D, how much better the silicon quality on the CCD is, depend on what sample you are comparing it against.
The very best 9800X3D can probably match it / get close, but there are very few of this caliber out there in the wild.
Average sample vs average sample i would say + ~200-300mhz clockspeed once overclocked
(difference is larger when running stock 5.2ghz vs 5.6ghz like reviewers do)

So the 9850X3D is just a way to get very strong silicon, without binning hundreds of 9800X3D's
IO-die quality (IMC/FCLK) is still all random, so no guarantee there.

If your average user already own a 9800X3D it makes very little sense to buy a 9850X3D
These new cpus are for overclockers or ppl that are building new systems.

When it comes out, most of the reviews will be running stock clock speeds but we will get apples to apples memory clock speed comparison (rather than memory running at 4800, that the GB run showed).

I am curious if (or by how much) the ST results change when running at least normal (DDR5-6000) rather than suboptimal settings. MT results should change more but I am more curious about ST.

 

[Det0x]

5600/5200 = ~7.7% as the upper theoretical ST limit
Real world ~4-5%

 

[Joe NYC]

5600/5200 = ~7.7% as the upper theoretical ST limit
Real world ~4-5%

It's already showing 4.9% with slow memory:

https://twitter.com/x/status/1997243685969506797

 

[adroc_thurston]

It's already showing 4.9% with slow memory:

https://twitter.com/x/status/1997243685969506797

it's passmark, you can safely ignore it.