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

[DisEnchantment]

Speculate at will

 

[Hitman928]

It has been proven by measuring directly from the pin of the SoC that the error between the “real power” and the estimated one is less than 5%.

How did they measure, “from the pin”? There would be many power pins to monitor and as far as I know, all Apple CPUs are socket-less.

 

[tsamolotoff]

. I’d love dual CCD VCache for certain productivity workloads. It would help compensate for the 16 cores being membw constrained by dual channel memory.

According to my personal tests, x3d at relative minus 200 mhz is roughly equal to non-extra-cache cores in 100% membw sparse matrix multi algo, so it's not that much to be honest.

More importantly than us talking about a very fast X3D part: it nullifies the weird scheduling business needed for Zen 4 parts on X3D parts. If both CCDs can hit the same frequency, then the V-Cache CCD (aside from some extreme niche cases) will always be the faster CCD. Which means AMD just needs to rely on regular old CPPC to handle allocating workloads to cores, no need for the Game Bar nonsense we had last time.

You don't really need the game bar, just set the driver (or bios) to 'prefer cache', it basically emulates one-ccd design in this case. What you actually need (and what driver/gamebar etc not always (almost not ever, really, at least on W10) provide) is to keep certain games / apps on cache-only ccd (cs2 for example), as spreading threads between ccds essentially limits the performance to non-x3d level (or even worse).

If both CCDs can hit the same frequency, then the V-Cache CCD (aside from some extreme niche cases) will always be the faster CCD

X3D chiplet is already a higher quality one (especially if you have post-release 7950x3d), it's just ACPI table configured by default in such a way that inferior-quality CCD1 takes precedence. As I said above, just swap it to 'prefer cache'.

 

[vanplayer]

Some guys on X searched 9950x price with google and got an unavailable link from Canada website

https://twitter.com/x/status/1801865240214200597

https://twitter.com/x/status/1801964819324579984

 

[Makaveli]

CC sells the 7950X for $719 I was expecting about a $100 increase so this seems close to that.

 

[Timmah!]

So you guys think, if Zen5 could achieve 5,7GHz at 1,2V or less, which is, if i understand currently the max voltage for v-cache, then no clock regression might be on cards this time around? As in, the lower clocks were result of this voltage limit, not cores having trouble to move heat away cause of the v-cache layer being in the way?
I would be down for dual v-cache 9950x3D, if the clocks were the same.

 

[PJVol]

All this talk about frequency residency is kinda alarming, reminding me of the Zen 2 era, when some parts could not even reach the stated Fboost max.

 

[SteinFG]

Lower 9000X MSRP makes sense, as X3D is now the premium version that will cost ~100 dollars more. And since there won't be much time between X3D and regular chips, regular ones won't have time to drop prices before X3D release.

 

[inquiss]

All this talk about frequency residency is kinda alarming, reminding me of the Zen 2 era, when some parts could not even reach the stated Fboost max.

Feels to me the opposite, they're talking about it boosting all core for longer, this is more reassuring to me..

 

[PJVol]

Feels to me the opposite, they're talking about it boosting all core for longer, this is more reassuring to me..

Have you ever had any Zen 2 CPU ?
I doubt that.

 

[biostud]

Do you think we will see better performance uplifts in 64+ core CPUs than in desktops? (MT obviously)

 

[SteinFG]

There will be GMI-wide for 64-core EPYCs, I assume (beacuse of the increased number of GMI links), Don't know how helpful it is though

 

[inquiss]

Have you ever had any Zen 2 CPU ?
I doubt that.

They comparing the boost behaviour being higher all core than zen 4. So you're saying zen 2 had some issues getting to top boost speed on your CPU. Is that relevant here? Not trying to be salty, just really not seeing the relevance so need some help.

I think zen 3 might have been the generation where they changed how they advertised boost clocks, leaving some MHz in the tank. I can't remember for sure, but that could be what you're talking again. Might make the point you're making moot, but again without context I don't know for sure.

 

[static shock]

I got in a search a leak of the top mobile cpu(HX AI) at 5.1Ghz matching the single core performance of 5.7Ghz 7950x in CB2024.

https://www.cpu-monkey.com/en/cpu-amd_ryzen_ai_9_hx_370

 

[Mopetar]

TIL. Geekerwan found that?

Interesting, but doesn’t address the other points about total power.

I'm not sure what the debate even is. Stop and think about it logically. The power you feed into a CPU is used to switch transistors. This action converts the power into waste heat. A negligible amount of power may flow through the CPU without doing any work or you'll have some capacitance built up for a transistor that isn't switched again in the future before powering the CPU off, but essentially all power put into will eventually be converted into waste heat.

A CPU cannot generate more heat than it has had power put into it, or we'd have to be drawing that power from a source other than the wall. Similarly, it can't generate less waste heat than the power that's been run through it or it would imply it's either passing that power through to somewhere else that consumes it or that it's storing that power internally in some way.

A CPU is basically a space heater that does some additional useful work with the electricity it consumes before turning it into heat. TDP will not be meaningfully less than the sustained max power consumed by the CPU.

 

[StefanR5R]

I'm not sure what the debate even is.

It was about measuring power use in given workloads. (Think of measuring Joule per task. Or Watt per task speed. At core level or socket level.) I.e. measurements unrelated to thermal design, related to power efficiency.

 

[Ajay]

When are the reviews for Zen5 coming out. Waited, what felt like forever for the reveal - now impatient to see what the real world performance of this new CPU will be.

 

[PJVol]

just really not seeing the relevance so need some help.

1. There were talks from amd reps about why boost clocks are not reached and all that residency BS after people and tech media started to complain that Zen2 is unable to do that.
2. Now there are again talks of frequency residency.

It's as simple as this.

I think zen 3 might have been the generation where they changed how they advertised boost clocks, leaving some MHz in the tank

So what prevents them from changing it again if they wanted to?

 

[uzzi38]

When are the reviews for Zen5 coming out. Waited, what felt like forever for the reveal - now impatient to see what the real world performance of this new CPU will be.

End of July. I think both GNR and STX NDAs lift then, if I'm remembering correctly.

 

[inquiss]

1. There were talks from amd reps about why boost clocks are not reached and all that residency BS after people and tech media started to complain that Zen2 is unable to do that.
2. Now there are again talks of frequency residency.

It's as simple as this.


So what prevents them from changing it again if they wanted to?

Yeah ok, thanks for clarifying, I still don't see how it's relevant. Some lingering skepticism based on some real world lived experience but contradictory to what's been said directly.

For Zen 2 the core only reached the boost clocks for small few portion of time, so for zen 3 onwards they changed how they communicate boost clocks. And in some conditions zen 3 and 4 achieve higher single thread boost beyond what is stated on the specs.

On your second point, on what prevents them changing again... Firstly, learning from the previous mistake. Secondly they've just communicated that all core is better. You're proposing both are incorrect, or at least that you have no trust because of a prior mistake. Seems a reach.

 

[Hail The Brain Slug]

Yeah ok, thanks for clarifying, I still don't see how it's relevant. Some lingering skepticism based on some real world lived experience but contradictory to what's been said directly.

For Zen 2 the core only reached the boost clocks for small few portion of time, so for zen 3 onwards they changed how they communicate boost clocks. And in some conditions zen 3 and 4 achieve higher single thread boost beyond what is stated on the specs.

On your second point, on what prevents them changing again... Firstly, learning from the previous mistake. Secondly they've just communicated that all core is better. You're proposing both are incorrect, or at least that you have no trust because of a prior mistake. Seems a reach.

What he's doing is hearing/reading the word "residency" and ignoring all context.

 

[StefanR5R]

Both so-called "Base Clock" (quote: "Represents the average effective base frequency of all cores.") and so-called "Max. Boost Clock" (quote: "Max boost for AMD processors is the maximum frequency achievable by a single core on the processor running a bursty single-threaded workload. Max boost will vary based on several factors, including, but not limited to: thermal paste; system cooling; motherboard design and BIOS; the latest AMD chipset driver; and the latest OS updates." AMD give an "up to" figure for this one.) are edge case clocks and as such quite unimportant for real performance. No need to get all worked up about these two really unimportant specifications. Especially to get worked up while ignoring AMD's fineprint which I quoted.

 

[eek2121]

All this talk about frequency residency is kinda alarming, reminding me of the Zen 2 era, when some parts could not even reach the stated Fboost max.

That was just early chiplet binning.

X3D chiplets have always clocked lower due to power and thermals.

What has been speculated is that the v/f curve may have been improved enough to eliminate that difference. A hypothetical AMD 9950X3D could, for example, have 3D V-Cache on both chiplets while still boosting to 5.75ghz. IIRC The 7950X3D only boosted to 5.25ghz. A boost to 5.75 GHz would an additional ~10% uplift on top of Zen5 IPC gains and X3D gains.

 

[PJVol]

For Zen 2 the core only reached the boost clocks for small few portion of time, so for zen 3 onwards they changed how they communicate boost clocks. And in some conditions zen 3 and 4 achieve higher single thread boost beyond what is stated on the specs.

I'm not sure what "communicate" means in that context, but with regards to boost behavior they have changed nothing

You're proposing both are incorrect, or at least that you have no trust because of a prior mistake.

Yes, that's it. Although I'd not take it as a mistake.

What has been speculated is that the v/f curve may have been improved enough to eliminate that difference. A hypothetical AMD 9950X3D could, for example, have 3D V-Cache on both chiplets while still boosting to 5.75ghz. IIRC The 7950X3D only boosted to 5.25ghz. A boost to 5.75 GHz would an additional ~10% uplift on top of Zen5 IPC gains and X3D gains.

Very well may be.

What he's doing is hearing/reading the word "residency" and ignoring all context.

Apparently you've never been told that speaking in the third person is bad manners

 

[B-Riz]

Lower 9000X MSRP makes sense, as X3D is now the premium version that wil costs ~100 dollars more. And since there won't be much time between X3D and regular chips, regular ones won't have time to drop prices before X3D release.

I believe AMD wants to flood the market and get people onto AM5 in the next few months.

The "budget" AM5 option is now the 7000 series, the new thing is 9000 and the hot stuff will the the 9000 X3D chips.

This is just all about sticking it to Intel, as, really they can; these chips are not expensive to produce and you get a lot of PC even with a B650 board.

 

[inquiss]

I'm not sure what "communicate" means in that context, but with regards to boost behavior they have changed nothing

Yes, that's it. Although I'd not take it as a mistake.


Very well may be.


Apparently you've never been told that speaking in the third person is bad manners

Communicate in this context means that the SKUs used to have no fat whatsoever in the boost, now there is some fat. They tell us a different (lower) max single core boost for the same silicon before and after the change. So a processor that can do 5 GHz single core in some limited timeframes would have been labelled as 5 GHz but now that same silicon is labelled as 4.85 GHz or something.

Of course, the above has nothing to do with zen 5, not should it.. They literally told us that SKUs with the same advertised single core boost will have a higher effective clock because it will stay there for longer. That's relevant.