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

[DisEnchantment]

Speculate at will

 

[static shock]

None of this is true.

In reality it was. AMD got much hurt due to the 3nm TSMC yields. Yelding rate grew too slow for the Time-To-market AMD needs for win versus Intel. And versus Apple too.

 

[Josh128]

How about a competition? The one who guesses the closest scores for CBR23 for 120w PPT, 230w PPT and unlimited PPT, gets a free CPU benchmark of their choice (terms and conditions apply). Voting stops after 60w result is posted.

Terms and conditions: Nothing that requires uploading to an online database or needs an online profile to run.

And the competition begins with the first screenshot from the super awesome ES user:

View attachment 103040

**EDIT: Changed my 230W to from 44.5 to 45.5, unlimited to 46.5K (7/14/24 21:05 US Central Time. Locked in now, let the chips fall where they may!

120W PPT = 39.5K

230W PPT= 45.5K

Unlimited PPT = 46.5K

This is exciting!!

 

[Timmah!]

my tip is 230W PPT = 51,5K

because reasons

 

[soresu]

and some aren't even humans!

Oh no I have been found out, you'll never take me alive!

 

[Det0x]

120W: 39,500
230W: 45,000
Unlimited: 46,500

Are you getting ready for a R23 bench maraton, so i we have a comparison for the ES ? 🤣
🙏

 

[biostud]

Can the vsoc problems have been improved allowing for higher vsoc?

 

[Hail The Brain Slug]

Are you geting ready for a R23 bench maraton, so i we have a comparison for the ES ? 🤣
🙏

You know it

 

[gaav87]

You know it

1 screenshot per day this is painful

 

[DrMrLordX]

In most cases wouldn’t the real life differences be most significant in the non 3D models? And besides benchmarking and fun and giggles, be less than 5% for the 3D models in real life scenarios?

Depends on which real-life scenario. If the workload hits main memory a lot, it'll make a difference. If not, then main memory speed won't have as large an influence. The 3D cache die is supposed to reduce the number of applications that have to go to main memory often, but it isn't large enough to eliminate those possibilities.

 

[Josh128]

Heres my 5900X @60W in R23 for comparison. Just 60W PPT, everything else stock.

5864 @60W. So 97.73 points/W.

The 9950X ES number of 12424 at 40W is 310.6 points/W!

So we are looking at a staggering +318% increase in points per watt efficiency here! Admittedly, my 5900X performance really tanks under about 70W PPT, but apparently the 9950X does much better at that level. Would love to see how that changes if both chips are compared at their most efficient points!

**UPDATE: Did a run at stock settings, scored 19730 @141W. This yields 140 points per watt, an efficiency increase of 43% vs what my 5900X does at 60W. Still less than half of the Zen 5 at 40W, which is impressive, because the 5900X falls off so much at 40W PPT I didnt even bother to run it. I can only imagine the points/watt in the sweet spot of 16 core Zen 5, likely somewhere between 160W and 190W I would guess.

 

[csbin]

9900X: Default 33K, PBO 34.5K
So 9950x: 230W ppt ~42K, Unlimited ~45K

AMD Ryzen 9 9900X 12-Core "Zen 5" CPU Performance In Cinebench R23 Leaks, 20% Uplift Over 7900X With PBO

We have received the first performance numbers of AMD's 12-core Ryzen 9 9900X "Zen 5" Desktop CPU within the Cinebench R23 benchmark.

wccftech.com

 

[Hail The Brain Slug]

How about a competition? The one who guesses the closest scores for CBR23 for 120w PPT, 230w PPT and unlimited PPT, gets a free CPU benchmark of their choice (terms and conditions apply). Voting stops after 60w result is posted.

Terms and conditions: Nothing that requires uploading to an online database or needs an online profile to run.

And the competition begins with the first screenshot from the super awesome ES user:

View attachment 103040

I went to start running my suite at the list of PPT tiers and noticed the 40W bench was run at stock/JEDEC, presumably to provide more power to the cores. I was going to run through them all tonight but now that he keeps changing the memory setup I won't waste my time until I know the config for each PPT level. Just ran 40W at stock/JEDEC here but stock on my X670E Aorus Xtreme sets to 2000FCLK, not 1800FCLK as pictured in his screenshot.

Heres my 5900X @60W in R23 for comparison. Just 60W PPT, everything else stock.

5864 @60W. So 97.73 points/W.

The 9950X ES number of 12424 at 40W is 310.6 points/W!

So we are looking at a staggering +318% increase in points per watt efficiency here! Admittedly, my 5900X performance really tanks under about 70W PPT, but apparently the 9950X does much better at that level. Would love to see how that changes if both chips are compared at their most efficient points!

**UPDATE: Did a run at stock settings, scored 19730 @141W. This yields 140 points per watt, an efficiency increase of 43% vs what my 5900X does at 60W. Still less than half of the Zen 5 at 40W, which is impressive, because the 5900X falls off so much at 40W PPT I didnt even bother to run it. I can only imagine the points/watt in the sweet spot of 16 core Zen 5, likely somewhere between 160W and 190W I would guess.

View attachment 103066

Considering my 7950X outscored his by quite a lot at the same PPT and core power, it seems less than impressive.

Lowering my FCLK to 1800 simply increased the power available to the cores and bumped the score a few percent.

Benchmarks:
Fully AUTO (2000FCLK)


Manual 1800FCLK

 

[H433x0n]

How about a competition? The one who guesses the closest scores for CBR23 for 120w PPT, 230w PPT and unlimited PPT, gets a free CPU benchmark of their choice (terms and conditions apply). Voting stops after 60w result is posted.

Terms and conditions: Nothing that requires uploading to an online database or needs an online profile to run.

And the competition begins with the first screenshot from the super awesome ES user:

View attachment 103040

42K 230W, 44K unlimited.

Is there anyway this guy can just run a standard configuration? No CO and with memory at DDR5-6000 or DDR5-6400 to get an idea of how this will perform for majority of users?

 

[igor_kavinski]

No CO and with memory at DDR5-6000 or DDR5-6400 to get an idea of how this will perform for majority of users?

The one for CBR23 is stock DDR5-4800 coz that benchmark isn't influenced much by RAM speed anyway.

 

[igor_kavinski]

42K 230W, 44K unlimited.

Gotta estimate 120W score too, to be eligible.

 

[Abwx]

The one for CBR23 is stock DDR5-4800 coz that benchmark isn't influenced much by RAM speed anyway.

I would think that stock RAM speed should be at least 5200 like the 7950X.

Anyway in the waiting of those CB scores, that should be 46k@230W PPT, there s yet another GB submission for a 9950X ES as well as for Strix Point GPU.

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Eng Sample: 100-000001277-60_Y processor.

browser.geekbench.com

ASUSTeK COMPUTER INC. ProArt P16 H7606WV_H7606WV - Geekbench

Benchmark results for an ASUSTeK COMPUTER INC. ProArt P16 H7606WV_H7606WV with an AMD Ryzen AI 9 HX 370 processor.

browser.geekbench.com

ASUSTeK COMPUTER INC. ProArt P16 H7606WV_H7606WV - Geekbench

Benchmark results for an ASUSTeK COMPUTER INC. ProArt P16 H7606WV_H7606WV with an AMD Ryzen AI 9 HX 370 processor.

browser.geekbench.com

All links and automated GB scores scans can be found here :

x.com

x.com

 

[Philste]

This is such a weird launch again from AMD. 15th of July today and we know nothing about ZEN5. No prices, no release day. Why they are doing it like this?

 

[KompuKare]

This is such a weird launch again from AMD. 15th of July today and we know nothing about ZEN5. No prices, no release day. Why they are doing it like this?

Last minute panic about max wattage and prices?
They probably know what prices they could sell at while still being very profitable but want to "reward" early adopters, if they can?

 

[tsamolotoff]

... my 5800X3D's I/O die just became 32% more efficient (in five minutes)

I don't think this option is available on AM5 boards, definitely not on Gigabyte ones

Also, found the comparison of delidded vs not delidded by mister dom himself for ya,

[Official] Zen 4 X3D Owner's Club (7800x3D /...

This thread was created to serve as a meeting place to share thoughts, speculate, and theory-craft on the upcoming AMD Zen 4 Raphael-X 7XXXx3D chips: The Ryzen 7 7800X3D, Ryzen 9 7900X3D, and Ryzen 9 7950X3d. After launch, this thread will serve as a place for owners to share experience...

www.overclock.net

I'd say delta would become even less significant if both ccds are gently polished after the hot iron procedure to remove last bits of S(h)TIM.

 

[SteinFG]

The price will probably be announced the same time as reviews drop. GPUs ben doing this for some time

 

[coercitiv]

but want to "reward" early adopters, if they can?

Oh yes, this is the name we must give it. AMD EAR - Early Adopter "Rewards"

 

[biostud]

This is such a weird launch again from AMD. 15th of July today and we know nothing about ZEN5. No prices, no release day. Why they are doing it like this?

Besides a small audience of enthusiasts, do the market really care?

 

[PJVol]

I don't think this option is available on AM5 boards, definitely not on Gigabyte ones

It never was on AM4 Gigabyte boards also. Guess which vendor was caught cheating first?

Also, found the comparison of delidded vs not delidded by mister dom himself for ya

I'm not sure I'm supposed to know who "mister dom himself" is and that his "hottest/coldest" temperature delta conclusions are correct, since they're based on peak core temperatures rather than average.
But even in this case, the data on the CCD0 core temps don't differ much on the subject discussed.

 

[tsamolotoff]

I'm not sure I'm supposed to know who "mister dom himself" is and that his "hottest/coldest" temperature delta conclusions are correct, since they're based on peak core temperatures rather than average.

Det0x , of course.

 

[CouncilorIrissa]

This is such a weird launch again from AMD. 15th of July today and we know nothing about ZEN5. No prices, no release day. Why they are doing it like this?

Speak of the devil.

AMD Ryzen 9000 series officially launching July 31st, Ryzen 7 9700X is 12% faster than 5800X3D - VideoCardz.com

AMD finally confirms Ryzen 9000 launch date, but not pricing New desktop CPUs are launching on July 31st. It’s official now, confirmed by several retailers stocking the new Ryzen 9000 CPUs or preparing for preorders and AMD themselves: the official launch date for Zen5-based desktop CPUs is...

videocardz.com