Question Zen 6 Speculation Thread

[Hulk]

Difference is that the 48 AMD threads will be slower than the 48 Intel threads. Because for AMD it's 24C/48T, but for Intel it's 48C/48T.

(And yeah I know for Intel it's a mix of P+E cores while AMD uses only P cores, but the difference above will trump that.)

Since both look to be more than capable when it comes to MT, I believe will be decided based ST compute prowess/who can hit high clocks with low power/thermal load?

Let's face it 24C/48T or 48C/48T, for the desktop that's A LOT of MT compute. Outside of CB not a lot of benchmarks/apps (not counting running apps simultaneously) are going to hit all of those cores. But ALL apps/benchmarks will hit a handful of cores really hard.

 

[ETI4711]

Probably.

No one wants that.

Tom's Hardware (TH): When you view Zen 5c compact cores, do you think they only have a place in power-constrained environments \[mobile]? Could you see this coming over to desktop PCs, where power isn't a consideration?

Mike Clark (MC): \[...] If we keep building the compact cores in the way that we talked about—which I think we will; I don't know why I said it a little more theoretically—the hard part is really making sure we hit the right frequency point so that it's balanced with however many \[cores] you're going to put down. But let's say you're really good at that, then there's no reason not to put a compact core on a desktop.

Whether it's the same performance at a given core count to the customer and cheaper because there's less area used, or we can squeeze in even more cores on a desktop because of the compact cores. And we couldn’t leverage them \[performance cores] anyway because they were TDP-constrained when you got out to that many cores, so you may as well have used a compact core. I think as we get more experienced with Windows and see that the scheduling does work, well, I think you'll see us, in desktop, using the compact cores to both get more cores and be more cost-effective. Because it's wasted area \[for performance cores] because we can't run everything at that 5.7 GHz frequency.

An interview with AMD's Mike Clark, the Father of Zen — 'Zen Daddy' says 3nm Zen 5 is coming fast; also talks compact cores for desktop chips

AMD expands on its Zen 5 architecture.

www.tomshardware.com

 

[poke01]

Link please.

Look at M3 vs M4. M4 has higher clocks and better perf/w.

Since both look to be more than capable when it comes to MT, I believe will be decided based ST compute prowess/who can hit high clocks with low power/thermal load?

Let's face it 24C/48T or 48C/48T, for the desktop that's A LOT of MT compute. Outside of CB not a lot of benchmarks/apps (not counting running apps simultaneously) are going to hit all of those cores. But ALL apps/benchmarks will hit a handful of cores really hard.

I also think the flagship SKUs don’t matter anymore, it’s not about ST performance in DIY anymore but gaming perf. Who has the best gaming CPU will win that generation.

 

[511]

Link please.

Here N3 is N3B

 

[Fjodor2001]

Let's face it 24C/48T or 48C/48T, for the desktop that's A LOT of MT compute. Outside of CB not a lot of benchmarks/apps (not counting running apps simultaneously) are going to hit all of those cores. But ALL apps/benchmarks will hit a handful of cores really hard.

This depends on the type of workloads. There are already CPUs with a lot more cores than this.

Also, note that it may not only be a single App you’re using. You could e.g. be transcoding movies, compiling source code, running some anti-virus Sw, OS maintenance tasks, and watching movies / web pages or reading email, all at the same time. Most of those workloads can be executed in the background, and only the last few ones require human interaction.

 

[Thunder 57]

This depends on the type of workloads. There are already CPUs with a lot more cores than this.

Also, note that it may not only be a single App you’re using. You could e.g. be transcoding movies, compiling source code, running some anti-virus Sw, OS maintenance tasks, and watching movies / web pages or reading email, all at the same time. Most of those workloads can be executed in the background, and only the last few ones require human interaction.

Lol who is doing all of that at once? Probably nobody, ever. You are looking for ways to piss on AMD and make Intel look good in an AMD thread. You've been doing it for pages now.

 

[Fjodor2001]

Lol who is doing all of that at once? Probably nobody, ever. You are looking for ways to piss on AMD and make Intel look good in an AMD thread. You've been doing it for pages now.

It was just examples. Not everybody will be executing the same type of workloads at the same time. But many will be executing at least some workloads in parallel some of the time, which was the point.

Also, this is not an AMD vs Intel thing. Both will have 48T on DT in the next CPU generation, remember?

 

[poke01]

transcoding movies, compiling source code, running some anti-virus Sw, OS maintenance tasks, and watching movies / web pages or reading email, all at the same time.

What are all these people using now? Threadrippers?

You can do all of this on a 12-16 core machine. We need faster cores not more…

 

[adroc_thurston]

Also, note that it may not only be a single App you’re using. You could e.g. be transcoding movies, compiling source code, running some anti-virus Sw, OS maintenance tasks, and watching movies / web pages or reading email, all at the same time

You need maybe 8c for all that

 

[Joe NYC]

Look at M3 vs M4. M4 has higher clocks and better perf/w.


I also think the flagship SKUs don’t matter anymore, it’s not about ST performance in DIY anymore but gaming perf. Who has the best gaming CPU will win that generation.

I agree. I think it is Low Thread performance that matters most on client. 2 to 8 threads, maybe up to 16 threads.

The 2nd CCD of both Zen 6 and Nova Lake will be less relevant because the first CCD can cover pretty much all of the ground as far as client apps.

 

[MS_AT]

You need maybe 8c for all that

Nah, 1 core is plenty. If we had only 1 core CPUs we could have so much time back. I mean fire the compilation in the morning, go outside do some errands and then come back in the evening, and maybe it would have finished compiling. The effect is better if you limit the CPU to 1GHz.

We need faster cores not more…

That is harmful generalization. It all depends on what you are doing. I mean in mobile, it's indeed pointless as the device does not allow you to use all of this processing power due to OS support and device ergonomics (we need proper desktop mode for android, or dual boot to Linux/Windows).

Otherwise I would say we need more faster cores. It's then just on you to pick the CPU that meets your needs. Nobody is forcing you to buy 48c part if you can do everything you want with 8c machine. But some people would like to get more, and there exist a gap between when Threadripper starts being viable and what we have now. (For example, I find AM5 meeting all my needs for I/O but I could use more cores from time to time).

 

[adroc_thurston]

Nah, 1 core is plenty. If we had only 1 core CPUs we could have so much time back. I mean fire the compilation in the morning, go outside do some errands and then come back in the evening, and maybe it would have finished compiling. The effect is better if you limit the CPU to 1GHz.

now this is proper dramatism.

But some people would like to get more, and there exist a gap between when Threadripper starts being viable and what we have now

that's like 2% of the market.
whatever, those SKUs are for bragging rights.

 

[StefanR5R]

So you are saying that's the only CPU instruction that the CPUs are executing?

e.g.

"e.g." is shorthand for "for example".

In favourable circumstances for Intel it will be the case. Think Cinebench vs Y-cruncher. E cores have more but narrower execution units (4fma per cycle vs 2 on zen) so if software uses mostly scalar or narrow simd E cores might have a lead.

Thanks! I actually wanted to look up what the respective backends are equipped with (in the latest gen, assuming the upcoming gen will be similar) but then didn't go through with it. Now, the immediate next question would be about the energy consumption of the FMAs while they can be kept busy. Apparently this can only be answered satisfyingly after these CPUs actually become available.

And when we are past the questions of backend width and backend power requirements, we quickly get to the question of how well the frontends and the memory subsystems can keep these top end desktop CPUs utilized... So I stick with my idea that Fjodor2001 is a bit hastily getting to his conclusions.

PS, re Cinebench in particular: I have no doubt that those users who run Cinebench 24/7 are best served by the one CPU vendor who has been equipping his last several CPU generations with a dedicated Cinebench accelerator block. ;-) PPS, I admit to a personal bias to engineering/scientific use cases when it comes to CPU discussions.

Also, note that it may not only be a single App you’re using. You could e.g. be transcoding movies, compiling source code, running some anti-virus Sw, OS maintenance tasks, and watching movies / web pages or reading email, all at the same time. Most of those workloads can be executed in the background, and only the last few ones require human interaction.

Sounds like a lot of I/O waiting and cache thrashing.

I find AM5 meeting all my needs for I/O but I could use more cores from time to time

Several years back, when I personally started to need more cores for the first time (until then I only had use for faster cores, not more cores), I also needed proportionally more memory channels. (Plus better-than-Gigabit-Ethernet interconnect for spreading one application instance across the total available number of memory channels. It was for engineering work.)

 

[poke01]

is the 12+4+4+2 cores rumour for Medusa point true?

 

[jpiniero]

It was just examples. Not everybody will be executing the same type of workloads at the same time. But many will be executing at least some workloads in parallel some of the time, which was the point.

Also, this is not an AMD vs Intel thing. Both will have 48T on DT in the next CPU generation, remember?

Cinebench 4eva!!!!!!!!!!!!!!

 

[poke01]

If your workloads require 48 cores, is that for professional work or hobby?

Cause if it’s for a hobby I get why people are excited for dual channel 48 cores cause it’s cheap MT.

But any persons that needs that many cores for a living will need more memory bandwidth to go along with it.

 

[jpiniero]

If your workloads require 48 cores, is that for professional work or hobby?

Cause if it’s for a hobby I get why people are excited for dual channel 48 cores cause it’s cheap MT.

But any persons that needs that many cores for a living will need more memory bandwidth to go along with it.

Because they want to get a high score in Cinebench?

 

[branch_suggestion]

is the 12+4+4+2 cores rumour for Medusa point true?

MDS1-hi is 12+0+0+2.
CCX on the monodie is disabled for a myriad of reasons, simplest is power constraints make extra MT juicing irrelevant.
Most pertinent is not making validation engineers commit sudoku.

 

[OneEng2]

My impression is that Pat kicked bean counters to their nuts and that design is for performance without cost limits. It might not be financially valid product but it sure should not lack performance and AMD need to be on top of their game to retain top spot on performance race.

I am wondering this as well. Many seem to assume that ARL is a "dud" and brush aside NVL. I am wondering that with a tweak or two, NVL might not be strong competition for Zen 6.

Since both look to be more than capable when it comes to MT, I believe will be decided based ST compute prowess/who can hit high clocks with low power/thermal load?

Let's face it 24C/48T or 48C/48T, for the desktop that's A LOT of MT compute. Outside of CB not a lot of benchmarks/apps (not counting running apps simultaneously) are going to hit all of those cores. But ALL apps/benchmarks will hit a handful of cores really hard.

For the analytical among us (<0.001% of consumers) I totally agree .

I am still wondering about a resurgent Blue Man Group with an add campaign of "COREZILLA, the ultimate consumer processor"

Look at M3 vs M4. M4 has higher clocks and better perf/w.

Look at Zen 3 -> Zen 4. 7nm->5nm family (16% clock increase)

Moving from Zen 5 -> Zen 6 5nm -> 2nm family I suspect we might expect about the same so 5.7Ghz -> 6.6Ghz. The more I think about it, the more likely this appears. I think the reality will likely be more like 6.3Ghz due to yield considerations.

If your workloads require 48 cores, is that for professional work or hobby?

Cause if it’s for a hobby I get why people are excited for dual channel 48 cores cause it’s cheap MT.

But any persons that needs that many cores for a living will need more memory bandwidth to go along with it.

Other than CB, are there any other apps that a consumer would use regularly that could take advantage of such core counts without corresponding increases in bandwidth?

Because they want to get a high score in Cinebench?

Exactly.

I still think Intel will beat the drum of "more cores" for marketing reasons.

 

[eek2121]

Difference is that the 48 AMD threads will be slower than the 48 Intel threads. Because for AMD it's 24C/48T, but for Intel it's 48C/48T.

(And yeah I know for Intel it's a mix of P+E cores while AMD uses only P cores, but the difference above will trump that.)

You aren’t factoring power into this equation.

If Intel and AMD have similar power limits, Intel will have half the effective amount of power available to each core.

 

[marees]

MDS1-hi is 12+0+0+2.
CCX on the monodie is disabled for a myriad of reasons, simplest is power constraints make extra MT juicing irrelevant.
Most pertinent is not making validation engineers commit sudoku.

If I understood this correctly the 12 comes from one 12 p-core ccd & the 2lp comes from a monolithic medusa-point-1 but the 4p & 4c cores disabled right ?

 

[branch_suggestion]

If I understood this correctly the 12 comes from one 12 p-core ccd & the 2lp comes from a monolithic medusa-point-1 but the 4p & 4c cores disabled right ?

Correct, the LP island is very important.

 

[CouncilorIrissa]

Correct, the LP island is very important.

Only 2 LP cores sounds questionable imo, they might need to wake the CCD way too frequently.

 

[adroc_thurston]

Only 2 LP cores sounds questionable imo, they might need to wake the CCD way too frequently.

it's 4t.

 

[Hulk]

Okay prediction time.

Based on my rig running CB R23 (more fun to run/watch than R24) as the baseline for Zen 6 predictions....
This is how I've been running (and I hit it hard) for the past year with absolutely no issues and perfect stability.
For the record, I think it's nearly impossible for Raptor Lake to attain this score without a custom loop and be reliable.
That's why after 30+ years of Intel I moved to AMD.
Anyway...

9950X
4.87GHz CCD1, 4.55 CCD2 - Average clockspeed = 4.71GHz
200W Package Power
75 C on 280 AIO
1.09V
CB R23 MT score 41,000

What will the clockspeed and CB R23 score be for the 24/48 Zen 6 CPU be at 200W?

If AMD does not set Zen 6 at 200W "out of the box" then what power will they use and what will the frequency and score for CBR23 be?

I'm curious for the 6 to 7GHz people out there how many will be claiming over 6GHz in this actual real world example for Zen 6? Even if you shut down 8 cores and made it apples-to-apples what do you think for frequency and score?