Question Zen 6 Speculation Thread

[CouncilorIrissa]

I can confirm the Zen 3 data, I have an old manual 3.5GHz ST run from my previous 5900X. 3600 CL17 Patriot Viper RAM, XMP settings. 329 ST points/GHz. Very interesting that Zen 4 had more R23 ST IPC increase from previous gen than Zen 5 does. Until now I thought it was the other way around.

I wish we could pin this article, because the thread is going in circles.
Cinebench is scalar SSE/scalar AVX spam, it cares little for vector execution, which is the weak point of Zen 5.

• All formerly 1-cycle latency SIMD instructions now have 2-cycle latency. Applies to all widths - including scalar.

 

[fastandfurious6]

MT freq, max. TDP assumptions as well in order to complete your entry

desktop defaults 200W ~5GHZ no tuning = ~70000

Laptop defaults ~130W 4.5GHZ = ~60000

Laptop ~80W 4GHZ = ~50000

Laptop OC = 70000+ on par with desktop

 

[OneEng2]

So let's see...

16 P core, 32 E cores, and 4 LPE cores for Nova Lake at the top-of-the-stack.

Let's assume 450 for the "IPC" of the P as calculated in my chart and 375 for the E's.
Furthermore let's assume 5GHz for the P cores, 4GHz for the E cores, and 2.5GHz for the LPE "islands."

That would produce a CB R23 MT score of 87,500. "Don't call it a comeback!"

It's not AT ALL a silly set of assumptions.

This is why I have been skeptical that Zen 6 will be besting NVL 52c in CB.

That would be a monstrous MT score that desktop Zen 6 wont be touching without some hefty OC. However, I think that guesstimate could be a bit high unless its pulling 300W+, which, knowing Intel, would not be surprising at all. My guess would still be around 80-83K or so. In any case, this is exactly why I believe 24C Zen 6 will top out around 230-250W. 200W would just be leaving too much perf. on the table for AMD given what Intell will likely be doing.

The skymont cores were pretty low power (~4W / E Core?). Even 32 of them only eats up 128W.

For CB, I think NVL will be unbeatable....... but those mont cores wont amount to a hill of beans in DC were I expect Zen 6 to dominate.

 

[gdansk]

For CB, I think NVL will be unbeatable

Indeed, they've accelerated Cinebench. The question is how much faster in these embarrassingly parallel programs with no memory bottlenecks.

And not to ignore the fun people are having here estimating that ratio but when does it justify the additional area/cost? Look at the 9800X3D vs 285K. The Intel chip is 75-80% faster in CB for not much more money, usually 10%. And yet the 9800X3D has 108 pages of completed builds on PCPartPicker and the 285K 4 pages. bLLC should make that less one-sided but it seems like CB nT - after a certain point - doesn't map to $$$.

 

[Hulk]

That would be a monstrous MT score that desktop Zen 6 wont be touching without some hefty OC. However, I think that guesstimate could be a bit high unless its pulling 300W+, which, knowing Intel, would not be surprising at all. My guess would still be around 80-83K or so. In any case, this is exactly why I believe 24C Zen 6 will top out around 230-250W. 200W would just be leaving too much perf. on the table for AMD given what Intell will likely be doing.

I think you are right. More likely stock clocks might be 4.5/3.5/2, which would result in a score 77,400.

You know why I think this is likely? The Vulcan in me knows this is not a logical reason, but for the past few generations Intel and AMD have been neck-and-neck in this benchmark and I don't expect things to change.

There are so many variables. Intel could have very low IPC increase in this benchmark or clocks could be low without IPC and AMD could have very nice frequency and IPC improvement. Who knows if Intel's top dog will even be 52 core? Change of plans, yield issues, or whatever? On the otherhand I have a feeling Zen 6 is going to come out swinging.

That's why I still believe this battle will be decided with ST performance.

 

[Doug S]

I track CB scores in a spreadsheet

Hey wait just a minute now, I thought people here were supposed to make their arguments on this site based on rumors, gut feelings and vibes, not actual hard data!

 

[inquiss]

Link please.

Please be so kind as to provide a link to non-public information that won't exist in the public domain.

Every. Time.

A link to a reputable site would make something more credible, if the site only deals in public information i.e. if AMD has already confirmed it. Since you're asking a question that AMD themselves has not confirmed, you must know that it's only going to be rumour mill sites picking up people on Twitter or even here. Some of that will be nonsense, some of it will be leaded non-public information. The link itself doesn't mean anything. One of these days someone will provide you a link when you ask this question that is based on a wccftech article that links back these forums just to prove this point.

 

[marees]

wccftech article that links back these forums just to prove this point.

Wccftech seem incapable of parsing rumours properly

Here they claim entire rdna 5 stack will be gddr7 when we "know" that AT3 & AT4 will be lpddr5x/lpddr6

https://twitter.com/x/status/1985688212875223055

 

[poke01]

AMD will be fixing its most annoying issue next gen, idle power. Zen6 might be worth it even from Zen5

 

[StefanR5R]

As for the Cinebench R23 IPC, why would you say it was probably not part of the workload mix the projection was based on?? Zen 5's official IPC calculation slide certainly included it.

View attachment 133232

The rumored projection would be (if it was indeed done by AMD in 2023, showing their x86 microarchitectures of the 3rd...6th generation) based on a benchmark suite which is a) stable over many years, b) relevant to the several business segments including datacenter, in other words, likely SPECrate 2017 Integer. That is, such a projection would not be based on an arbitrary benchmark mix which would change from occasion to occasion.

AMD's Computex 2024 keynote presentation (launch of Granite Ridge and Strix Point), in contrast, contains a completely arbitrary mix of benchmarks, selected a posteriori just for this keynote presentation, such that the geomean equals a desired value.

 

[inquiss]

Wccftech seem incapable of parsing rumours properly

Here they claim entire rdna 5 stack will be gddr7 when we "know" that AT3 & AT4 will be lpddr5x/lpddr6

https://twitter.com/x/status/1985688212875223055

Perfect, so we the forum dwellers know more than they do. Which proves the point that 'link', for non-public info does not a confirmation make

 

[Fjodor2001]

Please be so kind as to provide a link to non-public information that won't exist in the public domain.

Every. Time.

A link to a reputable site would make something more credible, if the site only deals in public information i.e. if AMD has already confirmed it. Since you're asking a question that AMD themselves has not confirmed, you must know that it's only going to be rumour mill sites picking up people on Twitter or even here. Some of that will be nonsense, some of it will be leaded non-public information. The link itself doesn't mean anything. One of these days someone will provide you a link when you ask this question that is based on a wccftech article that links back these forums just to prove this point.

If there is no source or even reasoning provided along with a claim, then how can it be trusted to be correct?

Some people on Internet forums often tend to claim that whatever fact is for certain, when it's actually just guesses / speculation.

 

[511]

I track CB scores in a spreadsheet and I look through quite a lot of data before I "sit on" a number for a particular core. I'm actually seeing +12.8% for MT and +9.3% ST when moving from Zen 4 to Zen 5.

For the record a number of years ago we had a "thing" over IPC. IPC hasn't really been an accurate terms in decades. When we use the term we are talking about "throughput" or the "rate of work" per cycle in a specific workload.

Here is my chart, which is waiting to be updated with Zen 6 data. I don't look for the absolute best, ie... just restarted computer, all non essential apps shut down, tightest timing that can be run for a bench score. I look at what most review sites are getting on these CPU's and then I average results to something I have some measure of faith in.

View attachment 133251

Here is my data:
View attachment 133252

and just for fun:

View attachment 133253

nice chart

 

[511]

I track CB scores in a spreadsheet and I look through quite a lot of data before I "sit on" a number for a particular core. I'm actually seeing +12.8% for MT and +9.3% ST when moving from Zen 4 to Zen 5.

For the record a number of years ago we had a "thing" over IPC. IPC hasn't really been an accurate terms in decades. When we use the term we are talking about "throughput" or the "rate of work" per cycle in a specific workload.

Here is my chart, which is waiting to be updated with Zen 6 data. I don't look for the absolute best, ie... just restarted computer, all non essential apps shut down, tightest timing that can be run for a bench score. I look at what most review sites are getting on these CPU's and then I average results to something I have some measure of faith in.

View attachment 133251

Here is my data:
View attachment 133252

and just for fun:

View attachment 133253

Skymont is just to funny 35% vs RPL Gracemont

 

[Joe NYC]

AMD will be fixing its most annoying issue next gen, idle power. Zen6 might be worth it even from Zen5

Yes. And that saved power than is also additive to the performance during high CPU utilization.

 

[lixlax]

AMD will be fixing its most annoying issue next gen, idle power. Zen6 might be worth it even from Zen5

Something has been done already in the last 1,5 years or so. Currently my system idles around 65-70W, before it was 80-85W (that has been about the standard for normal single CCD Ryzen systems with Expo/XMP enabled). I've changed CPU and GPU in the meantime, but these had almost no effect to idle power draw. I didn't catch the exact moment, but it must've been a BIOS, Windows or a chipset driver update.
I hope Zen 6 can lower it even further.

 

[Hulk]

That would be a monstrous MT score that desktop Zen 6 wont be touching without some hefty OC. However, I think that guesstimate could be a bit high unless its pulling 300W+, which, knowing Intel, would not be surprising at all. My guess would still be around 80-83K or so. In any case, this is exactly why I believe 24C Zen 6 will top out around 230-250W. 200W would just be leaving too much perf. on the table for AMD given what Intell will likely be doing.

I know I keep repeating this but I think as our projected MT compute numbers climb into the stratosphere the number of desktop users that will actually be using that compute grows smaller and smaller. Right now, with my 9950X and it's "measly" 16 cores and 32 threads there aren't a lot of cases where I'm fully loading all of those cores. Probably my biggest use case would be encoding video and running Pureraw (raw image processing) in the background while doing DAW work in the foreground. Enough compute has to be available to playback a mix without dropouts. But even with these apps, encoding is CPU based for me (I like the better efficiency/bit of CPU routines), PureRaw is primarily GPU, so there is generally some CPU cores left over for other tasks.

My point is that my "bottleneck" currently, if I were to complain, is ST performance, as most applications rely heavily on a couple cores, sometimes even one core. I notice even my HX370 based laptop scores ~20,000 CB R23 MT and allows me to multitask pretty effortlessly.

But, I DO realize I'm not everybody, there ARE definitely people out there who will be using applications that will load up all cores available (DC, Blender, etc..) or really hit the multitasking hard, but we're moving into an era where we have all the MT we need. We need more ST and/or software developers to start coding better for MT.

If Zen 6 can sustain a 10% frequency advantage over Zen 5 when lightly loaded, meaning 6.27GHz, then without any IPC improvement that is a win for me. Add in 5 or 10% for architecture and 8 additional cores and it's a winner. Of course if Intel can do the same then we have two great options as well have to look into power efficiency, thermals, pricing, motherboard features and support, etc...

 

[Hulk]

I wish we could pin this article, because the thread is going in circles.
Cinebench is scalar SSE/scalar AVX spam, it cares little for vector execution, which is the weak point of Zen 5.

That is a great article!

For better or worse part of what we do here between release cycles is go in circles. It keeps us occupied while we wait.

That being said it is definitely better for the discussion for us to all be well-versed on the facts of what we are discussing.

Thanks for posting.

 

[Joe NYC]

If there is no source or even reasoning provided along with a claim, then how can it be trusted to be correct?

Some people on Internet forums often tend to claim that whatever fact is for certain, when it's actually just guesses / speculation.

AMD said, during their earnings call, that customers are already running Venice in their datacenters.

For this to be true (and it is), the tape out would have to have taken place early 2025. So we are way past tape outs.

The only argument (and you can judge plausibility) is that AMD, for the first time since Zen was introduced, broke the commonality of the CCDs between Server and Client.

While the rumors are pointing to the opposite direction, that AMD is increasing commonality of the main CCD to include not only Server and Desktop, but also some Mobile models.

 

[luro]

If there is no source or even reasoning provided along with a claim, then how can it be trusted to be correct?

Some people on Internet forums often tend to claim that whatever fact is for certain, when it's actually just guesses / speculation.

Some people here have credibility because of their past leaks, but it's up to you to decide whether you believe them or not. Still, it's ridiculous to ask for a link or a source of a leak.

 

[Fjodor2001]

Some people here have credibility because of their past leaks, but it's up to you to decide whether you believe them or not. Still, it's ridiculous to ask for a link or a source of a leak.

Some people also do not have credibility of their past leaks.

And some people always state whatever they say as fact, regardless of whether it's insider info, or official public info, or reasoning, or just pure speculation, or whatever. Then it's hard to judge the credibility of each of their various claims.

Better to do like Joe NYC did above and provided reasoning along with the claim (or source if that exist, or whatever is suitable). Then people can judge the claim themselves, and question it if they do not think it make sense.

 

[luro]

Some people also do not have credibility of their past leaks.

Yes, and you are free to believe them or not. But asking a link for the source of a leak doesn't make sense.

 

[jpiniero]

Yes, and you are free to believe them or not. But asking a link for the source of a leak doesn't make sense.

That's OK, it's just voices in her head.

 

[Fjodor2001]

Yes, and you are free to believe them or not. But asking a link for the source of a leak doesn't make sense.

It was not mentioned whether the info/claim was from a leak or whatever, so then asking source/reasoning or similar makes perfect sense.

See also my updated post that you quoted.

That said, I guess asking for a link in this case could also be message saying:

Sorry I don't trust you, you've stated too much as fact before without anything to back it up, and it turned out to just be speculation.

Now the above is just me speculating though, and I don't have any link to source for that guess...

 

[Hulk]

Some people here have credibility because of their past leaks, but it's up to you to decide whether you believe them or not. Still, it's ridiculous to ask for a link or a source of a leak.

I think there is some nuance here. Sometimes when I've missed something, a leak, news, whatever I will ask where someone heard this or that. It's not that I'm doubting the person, I just want to have a look at the source as well.