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

[DisEnchantment]

Speculate at will

 

[Saylick]

Ryzen 3000+

I imagine it would look something like this...

 

[igor_kavinski]

Quickly created this (apologize for the lame colors).

HINT: Keep your eyes on the dark blue bar.

 

[Hail The Brain Slug]

Quickly created this (apologize for the lame colors).

HINT: Keep your eyes on the dark blue bar.

View attachment 103311

Now we need a 5950X in on this action

 

[igor_kavinski]

Now we need a 5950X in on this action

No offense to 5950X owners but that's so ancient now

 

[inf64]

Hey @igor_kavinski , have you managed to get some R24 (ST and MT) numbers from the 9950X ES user? Sorry if I missed it in this thread.

 

[igor_kavinski]

Hey @igor_kavinski , have you managed to get some R24 (ST and MT) numbers from the 9950X ES user? Sorry if I missed it in this thread.

Not yet. Soon!

 

[StinkyPinky]

Just warming up my 14900K guys, she'll be ready soon. Benchmarks incoming after I get some insurance quotes.

 

[Det0x]

Now we need a 5950X in on this action

Found a old highly-tuned 5950x screenshot in my benchmark folder 🤣
(unlimited PPT)

 

[Josh128]

Found a old highly-tuned 5950x screenshot in my benchmark folder 🤣
(unlimited PPT)
View attachment 103315

That might be the highest AM4 R23 score Ive ever seen, lol.

 

[Det0x]

Hold on, i found my old power/performance screens for the 5950X
(highly min/maxed in cold norwegian winter-air 😇)

16 Zen3 cores @ 49 watt = 20441 points in Cinebench r23 (~lowest powerlevel i could run with my old setup)


16 Zen3 cores @ 64 watt = 23701 points in Cinebench r23


16 Zen3 cores @ 88 watt = 27037 points in Cinebench r23


And if that was not enough, i even found optimized runs with my 7950X 🤣

PPT 35w, 60A TDC and 120A EDC limit = 14667 points


PPT 50w, 65A TDC and 125A EDC limit = 22832 points


PPT 65w, 70A TDC and 130A EDC limit = 28387 points


PPT 100w, 80A TDC and 150A EDC limit = 34574 points


PPT 130w, 100A TDC and 160A EDC limit = 37344 points


PPT 160w, 115A TDC and 175A EDC limit = 39068 points


PPT 210w, 140A TDC and 200A EDC limit = 40085 points (OBS: CPU only pulled 204w maximum)


All the runs above are highly min/maxed for Cinebench r23 power/performance, so i'm not sure if its wise/fair to add them (probably but not fully stable for daily usage)

 

[Det0x]

That might be the highest AM4 R23 score Ive ever seen, lol.

Think this was my topscore with the 5950x 😋

 

[Geddagod]

Green is 9950x
Blue is 7950x
Orange is 7950X3D

Compared to Zen 4, (using the 7950x), Zen 5 sees a ~5% perf/watt at ~3 watts, ~10% at ~4 watts, at 15% at ~6 watts per core.
Turin is prob going to end up using ~3 watts per core.

 

[poke01]

Amazing performance, ARL has a tough challenge

 

[Hans Gruber]

View attachment 103332
Green is 9950x
Blue is 7950x
Orange is 7950X3D

Compared to Zen 4, (using the 7950x), Zen 5 sees a ~5% perf/watt at ~3 watts, ~10% at ~4 watts, at 15% at ~6 watts per core.
Turin is prob going to end up using ~3 watts per core.

Any word yet on the Zen 5 silicon for consumers? I know it's N4 something. Did anybody find out if it's N4 or N4P. I doubt that it would be N4X.

 

[Geddagod]

Any word yet on the Zen 5 silicon for consumers? I know it's N4 something. Did anybody find out if it's N4 or N4P. I doubt that it would be N4X.

N4P

 

[Saylick]

N4P

Yep.

Not bad in the grand scheme of things. Same CCD die area (~70 mm2), >20% increase in density while maintaining clocks, +16% IPC uplift, and even a small boost to perf/W. Oh, and they managed to add AVX512 and allegedly running AVX512 doesn't reduce clocks. If you can't call that a success, Idk what does.

https://twitter.com/x/status/1813281978591944827

 

[Hans Gruber]

N4P

When Zen 5 is officially released. If people want to test their CPU's at set power settings. Below is straight from TSMC with regards to performance improvements of N4P vs N5.

Compared to N5, N4P will deliver an 11% performance boost, a 22% improvement in power efficiency, and a 6% improvement in transistor density.

 

[H433x0n]

View attachment 103332
Green is 9950x
Blue is 7950x
Orange is 7950X3D

Compared to Zen 4, (using the 7950x), Zen 5 sees a ~5% perf/watt at ~3 watts, ~10% at ~4 watts, at 15% at ~6 watts per core.
Turin is prob going to end up using ~3 watts per core.

This is modeled off of the really high scoring ES with the extreme cooling setup?

Would be curious to see this when a mainstream review outlet gets a hold of the 9950X.

 

[poke01]

Yep.

Not bad in the grand scheme of things. Same CCD die area (~70 mm2), >20% increase in density while maintaining clocks, +16% IPC uplift, and even a small boost to perf/W. Oh, and they managed to add AVX512 and allegedly running AVX512 doesn't reduce clocks. If you can't call that a success, Idk what does.

https://twitter.com/x/status/1813281978591944827

It’s very impressive.

 

[branch_suggestion]

Compared to N5, N4P will deliver an 11% performance boost, a 22% improvement in power efficiency, and a 6% improvement in transistor density.

Only relevant if using the same IP on both.
Z5 is different IP to Z4 and has different DTCO tweaks.

 

[Hans Gruber]

Only relevant if using the same IP on both.
Z5 is different IP to Z4 and has different DTCO tweaks.

I do not know what IP means other than internet protocol. Do you mean power? Like setting the absolute power to 65w, 95w and 105w and 120w and comparing the performance of Zen 4 to Zen 5 clock for clock. That is what I was thinking.

 

[branch_suggestion]

I do not know what IP means other than internet protocol. Do you mean power? Like setting the absolute power to 65w, 95w and 105w and 120w and comparing the performance of Zen 4 to Zen 5 clock for clock. That is what I was thinking.

Intellectual Property.
TSMC will measure raw xtor gains between processes but they usually present gains between processes through neutral IP, usually ARM cores.

 

[Ghostsonplanets]

I do not know what IP means other than internet protocol. Do you mean power? Like setting the absolute power to 65w, 95w and 105w and 120w and comparing the performance of Zen 4 to Zen 5 clock for clock. That is what I was thinking.

IP = Intellectual Property in this case

He's basically saying that the gains of N4P when compared to N5 would only apply when compared to a same design (e.g: Zen 4) that is available at both nodes.

Zen 5 is a new design/IP and thus will have different knobs and DTCO. Thus you can't take TSMC at face value and say Zen 5 gains when compared to Zen 4 are due to node alone.

 

[Hail The Brain Slug]

@igor_kavinski @Geddagod
Made possible by a generous benchmarking effort donation from @mostwanted002, I've got stock 5950X figures to add to the performance and performance per watt graphs. For your convenience I've collected the pertinent data here, but also attached screenshots.

Side note, the 5950X hit other limits (EDC/TDC) when PPT was increased to 160W, so 160W and up are done with unlimited EDC/TDC to leave PPT as the only constraining factor. Otherwise, every setting is stock.

Second side note, it didn't take well to 40W PPT and ended up running 44W PPT, which is why there are 4 extra watts core power compared to the rest of the power levels. The score is already so low due to poor scaling I'm giving it a pass.

40W - 3691 - 17.9W core
60W - 8328 - 33.3W core
80W - 13700 - 53.6W core
100W - 19260 - 73.6W core
120W - 23878 - 93.3W core
160W - 25822 - 133.3W core
200W - 27790 - 173.3W core
230W - 28155 - 198.5W core
253W - 28253 - 198.4W core
U/lW - 28390 - 198.4W core

 

[mostwanted002]

somewhat late to the party I guess.
Zen5 looks like a good upgrade compared to my current specs with each passing day.