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

[DisEnchantment]

Speculate at will

 

[soresu]

Because of silicon cost designers need to get the max PPA because other wise the chips will cost us more than most of us are really willing to spend.

Going with chiplets does help with that somewhat.

It is cheaper to design a smol chip that can be used more than once in MCM configurations.

 

[fastandfurious6]

wheres mah zen6 6ghz 2nm 32core ultraX3D 1gb L3 cache mobile

9955HX3D in Q1 '25 hopeful?

TSMC just cant print enough money

 

[Jan Olšan]

If they did what you suggest they wouldn't cut back on the transistor budget for the core, they'd cut back on the transistor budget for the CCD. Which would mean fewer cores on the die or less L3.

This sounds extremely unrealistic to me, I don't think that would be considered for even a second. Going to six cores would never happen. And even cutting the L3 cache would be massive self-own.

The used 8core fabric setup and 32MB for 2 previous designs, it absolutely would be a big complication to change it. And it would drop performance in games (cache) and overall (core count cut). Last but not least it would look extremely silly. And not just look, it would be an actual hard regression.

I mean, shooting their own leg by lowering the max core count they can offer in servers and desktop - what could go wrong? They would actually be better of refreshing Zen 4 probably, bad as that would be. Realistically, at that point the wise choice is IMHO accept the die size inflation from using worse process node. Might even be what the did.

 

[eek2121]

All this talk about Zen 4 being “mid”, “average”, etc. makes me giggle.

Zen 5 is easily the most efficient x86 core out there, beating the previous Zen 3-based 5950X (which was the previous perf/watt champ), and people are complaining.

The TDP of the 9700X was dropped by 45-50% (120W for X3D vs 65W for 9700X) and it is still ahead of every non-X3D chip out there, and is close to X3D chips in gaming performance, despite the TDP handicap.

EDIT: To put differently, in gaming, the 9700X does about as well as the 7800X3D at about half the power. (88W vs 162W, fudging the numbers a bit because i don’t want to pull up calculator)

 

[H433x0n]

All this talk about Zen 4 being “mid”, “average”, etc. makes me giggle.

Zen 5 is easily the most efficient x86 core out there, beating the previous Zen 3-based 5950X (which was the previous perf/watt champ), and people are complaining.

The TDP of the 9700X was dropped by 45-50% (120W for X3D vs 65W for 9700X) and it is still ahead of every non-X3D chip out there, and is close to X3D chips in gaming performance, despite the TDP handicap.

slight nitpick, the 7950X is significantly more efficient than the 5950X. When both are limited to 105W the 7950X is >=35% more efficient.

 

[eek2121]

slight nitpick, the 7950X is significantly more efficient than the 5950X. When both are limited to 105W the 7950X is >=35% more efficient.

I am referring to out of the box. I do agree with you. My 7950X @65W/88W beats my 5950X @ 105W/142W.

 

[Tuna-Fish]

EDIT: To put differently, in gaming, the 9700X does about as well as the 7800X3D at about half the power. (88W vs 162W, fudging the numbers a bit because i don’t want to pull up calculator)

7800X3D doesn't actually use nearly that much power, despite the PPT. The peak power measured at the socket for Prime is <83W, and the typical heavy gaming load is something like 50W.

 

[Hail The Brain Slug]

7800X3D doesn't actually use nearly that much power, despite the PPT. The peak power measured at the socket for Prime is <83W, and the typical heavy gaming load is something like 50W.

Even the 7950X3D doesn't use nearly as much power as the 7950X despite having ~95% of the MT performance and the same power limits. I can't find a workload that causes mine to draw more than ~140W PPT but my 7950X will gulp 200-230W easily.

 

[StefanR5R]

Even the 7950X3D doesn't use nearly as much power as the 7950X despite having ~95% of the MT performance and the same power limits. I can't find a workload that causes mine to draw more than ~140W PPT but my 7950X will gulp 200-230W easily.

Stock power limits differ between the two. Did you compare with lifted PPT limit of the 7950X3D? And even if you did, I am guessing that AMD put certain constraints WRT on-chip clocks or currents or hot-spot thermals on the 3D V-cache chips which as a side effect prevent them to spiral off into uselessly inefficient spheres like X SKUs can.

(edit: better addressed by post #13,811)

 

[Doug S]

This sounds extremely unrealistic to me, I don't think that would be considered for even a second. Going to six cores would never happen. And even cutting the L3 cache would be massive self-own.

The used 8core fabric setup and 32MB for 2 previous designs, it absolutely would be a big complication to change it. And it would drop performance in games (cache) and overall (core count cut). Last but not least it would look extremely silly. And not just look, it would be an actual hard regression.

I mean, shooting their own leg by lowering the max core count they can offer in servers and desktop - what could go wrong? They would actually be better of refreshing Zen 4 probably, bad as that would be. Realistically, at that point the wise choice is IMHO accept the die size inflation from using worse process node. Might even be what the did.

If they were designing for N3 at first and had to backport (continuing the assumption made in the post I originally replied to, I'm not endorsing that assumption) they might have originally planned on 10 cores per die, and had to back off to 8.

What is unrealistic is redesigning a core to be smaller because of a change to the process roadmap. Such a major redesign would set your schedule back so far that it would obviate the reason for doing it.

 

[Mopetar]

The X3D chips always use a lot less power just because the v-cache limits the voltage. With Zen 4 where AMD upped the TDP by a lot, this translates to a massive reduction in power use and if the application can use the added cache, a performance bump on top of that.

Power use corresponds to the square of the voltage applied, so a 10% reduction in voltage is actually close to a 20% reduction in power. The lower frequency that goes along with that also reduces the power used in a linear fashion.

 

[DAPUNISHER]

The latest juicy tidbit?

https://www.techradar.com/computing...rrive-soon-to-supercharge-your-next-gaming-pc

 

[CouncilorIrissa]

The latest juicy tidbit?

https://www.techradar.com/computing...rrive-soon-to-supercharge-your-next-gaming-pc

Impressive. I wish I were able to make eight paragraph long articles that say next to nothing out of a single tweet and get paid for it.

 

[Makaveli]

The X3D chips always use a lot less power just because the v-cache limits the voltage. With Zen 4 where AMD upped the TDP by a lot, this translates to a massive reduction in power use and if the application can use the added cache, a performance bump on top of that.

Power use corresponds to the square of the voltage applied, so a 10% reduction in voltage is actually close to a 20% reduction in power. The lower frequency that goes along with that also reduces the power used in a linear fashion.

Voltage limits and lot less trips out to main memory if data is in the cache.

Can be seen in 5800X vs 5800X3D and 7700X vs 7800X3D etc

 

[DAPUNISHER]

Impressive. I wish I were able to make eight paragraph long articles that say next to nothing out of a single tweet and get paid for it.

Some don't use twitter and say it asks them to sign in to see the posts, so I did not direct link.

Any comments besides sarcasm and jelly?

 

[CouncilorIrissa]

Some don't use twitter and say it asks them to sign in to see the posts, so I did not direct link.

Any comments besides sarcasm and jelly?

The tweet itself is a nothingburger imo, nor do I share the author's excitement.

In fact, the news is disappointing because if the X3D chips are this close it probably means that vanilla models are not that competitive.

 

[DAPUNISHER]

The tweet itself is a nothingburger imo, nor do I share the author's excitement.

In fact, the news is disappointing because if the X3D chips are this close it probably means that vanilla models are not that competitive.

Which is exactly what the author referenced.

And it comes on top of our theories from yesterday about why AMD might want to push out these 3D V-Cache spins early – due to worries about how gaming performance of vanilla Ryzen 9000 might stack up to Intel’s incoming next-gen CPUs, Arrow Lake.

 

[CouncilorIrissa]

Which is exactly what the author referenced.

Well he knows just as much as us plebs do so this statement does not amount to much, eh.

 

[biostud]

If they are priced more than +25% of current zen4 prices, they will be a hard sell.

 

[DAPUNISHER]

Well he knows just as much as us plebs do so this statement does not amount to much, eh.

You are using the shoot the messenger fallacy. I explained why I chose the source. Now please stop attacking it and stay on topic.

 

[eek2121]

7800X3D doesn't actually use nearly that much power, despite the PPT. The peak power measured at the socket for Prime is <83W, and the typical heavy gaming load is something like 50W.

Well, I don't know of any Zen 4 chip that uses a ton of power gaming. However, you can absolutely pull 162 watts out of X3D chip, you just need the right test.

Zen 5 appears to have a much improved v/f curve, so I suspect it will clock higher, possibly, as @uzzi38 stated, as high as 5.7ghz. It will all depend on AMD and binning, but even 5.5ghz would be a huge deal. Previous gen topped out at 5.25ghz for the X3D chip, so any frequency gains are a welcome bonus and we could see up to an 8% improvement on top of Zen 5 PPC improvements and X3D boosts.

 

[Hail The Brain Slug]

Well, I don't know of any Zen 4 chip that uses a ton of power gaming. However, you can absolutely pull 162 watts out of X3D chip, you just need the right test.

Zen 5 appears to have a much improved v/f curve, so I suspect it will clock higher, possibly, as @uzzi38 stated, as high as 5.7ghz. It will all depend on AMD and binning, but even 5.5ghz would be a huge deal. Previous gen topped out at 5.25ghz for the X3D chip, so any frequency gains are a welcome bonus and we could see up to an 8% improvement on top of Zen 5 PPC improvements and X3D boosts.

If you know of a workload that will get my 7950X3D to actually hit its PPT cap, please tell me. I tried several different workloads and found nothing that took it meaningfully beyond around 140 watts. I've got it on water so it is not clocking down due to thermal limits.

 

[Nothingness]

If you know of a workload that will get my 7950X3D to actually hit its PPT cap, please tell me. I tried several different workloads and found nothing that took it meaningfully beyond around 140 watts. I've got it on water so it is not clocking down due to thermal limits.

Did you try Prime95 or yCruncher with AVX512 enabled and all cores running with HT?

 

[Hail The Brain Slug]

Did you try Prime95 or yCruncher with AVX512 enabled and all cores running with HT?

I couldn't get any configuration of prime95 to draw 162W PPT. I also tried ycruncher with AVX 512 enabled, blender benchmark, cinebenches. I am running through these workloads again to make sure I'm not just misremembering.

Edit: Confirmed, the maximum PPT I can get under any of these workloads is about 145W. Most are 125-135W. No thermal, power, or current limiting hit at any point.

 

[tsamolotoff]

Did you try Prime95 or yCruncher with AVX512 enabled and all cores running with HT?

Thanks to hard FIT caps, those type of workloads don't generate much heat as maximum allowed voltage is just too low (below 0.9v ish). The highest power draw I saw was in cinebench 23 or 20, ~180 or 190W or so.