Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[eek2121]

The truth does not get old. Only for those who need to mislead.

Who exactly is being misleading here? The folks spinning hyperbole. There isn’t any issue with what AMD has shown thus far. Folks getting stuck on the 15% part of greater than 15% and the TDP nonsense. They are missing the fact that these chips are seeing large multicore gains and will likely be the most efficient chips in terms of perf/watt that you can buy.

They take AMD’s messaging to say “minimal performance gains with much higher power consumption” rather than what it really is: “Our new chips not only perform great, but they scale very well.”

Part of that is due to AMD’s poor attempt at messaging. Most of it is due to users having the inability to look at the demos and information for what it is: a preview of an unreleased product. Bias probably also plays into it as well.

 

[tomatosummit]

Sadly none of above aplies to Zen4 vs Raptor lake discussion and esp the part of it where AMD 6 and 8 CPUs will have same priced counterparts with 4-8 extra efficiency cores.

Do you have a source on this or is it coming from where the sun doesn't shine?
When were we transported into a world where both cpu makers were very vocal on their planned skus specifications, prices and performance they will offer and had no ability to change all of those upto the moment of launch and even afterwards should they choose to do so.
IF it's big core matching big core sku cost for cost then it's because customers aren't buying into the extra e cores so they're considered a free addition for a cpu, like an igpu was until the F series.

 

[deasd]

Charlie said Genoa's spec scales back a bit, very likely lower clocks than original plan

AMD scales back Genoa a tiny bit

It looks like AMD is scaling back a little on Genoa’s specs for the moment.

www.semiaccurate.com

he guess it's the delay of a product from Intel that make AMD conservative

https://twitter.com/x/status/1530181614680129536

 

[JoeRambo]

Do you have a source on this or is it coming from where the sun doesn't shine?

It is in leaks and rumours, nothing is concrete before official announcement. But if You think that Intel will have 8P + 16E for top of the line product, and will not increase E core counts for other SKUs? Well I also happen to sell NFT bridges, are You interested?

Full list of Intel 13th gen Raptor Lake SKUs leak, Intel apparently looking to hit a world record 5.5 GHz turbo and retain single-threaded dominance with the Core i9-13900K - NotebookCheck.net News

 

[Hans Gruber]

Charlie said Genoa's spec scales back a bit, very likely lower clocks than original plan

AMD scales back Genoa a tiny bit

It looks like AMD is scaling back a little on Genoa’s specs for the moment.

www.semiaccurate.com

he guess it's the delay of a product from Intel that make AMD conservative

https://twitter.com/x/status/1530181614680129536

That article was BS about Intel, they didn't say what was delayed. I am guessing the delay will be the Intel ARC GPU's again. I think they are going on 2 years now.

 

[jpiniero]

That article was BS about Intel, they didn't say what was delayed. I am guessing the delay will be the Intel ARC GPU's again. I think they are going on 2 years now.

That's because the meat of the article is behind the paywall.

 

[lobz]

This has become such an absurd thread again. It's like many of you would say Robert Hallock is going on an unprecedented press frenzy of explaining things about a teaser of a product that hasn't even been announced solely to defend a leaker you don't like.

 

[positivedoppler]

I'm still not sure what's the difference between TDP and PPT. So PPT is the amount of power the socket can draw, why is that different than the "Thermal Design Power?"

 

[tomatosummit]

It is in leaks and rumours, nothing is concrete before official announcement. But if You think that Intel will have 8P + 16E for top of the line product, and will not increase E core counts for other SKUs? Well I also happen to sell NFT bridges, are You interested?

Full list of Intel 13th gen Raptor Lake SKUs leak, Intel apparently looking to hit a world record 5.5 GHz turbo and retain single-threaded dominance with the Core i9-13900K - NotebookCheck.net News

I meant this
>"AMD 6 and 8 CPUs will have same priced counterparts with 4-8 extra efficiency cores."
The actual core counts will not matter, if they have the performance prices will be adjusted to suit.
Intel and AMD are playing a game, amd has just revealed a card about zen4 and rumour mills are in overtime.

 

[lobz]

That article was BS about Intel, they didn't say what was delayed. I am guessing the delay will be the Intel ARC GPU's again. I think they are going on 2 years now.

It was not BS, he just lives off subs so peasants like us don't get the exact info - although it's not really hard to guess.

Everything is on track at Intel, just in an alternate timeline where Jesus died 2 years later.

 

[JoeRambo]

The actual core counts will not matter, if they have the performance prices will be adjusted to suit.

Erm, isn't it the crux of my argument. That AMD has 6 cores, Intel has 6 cores PLUS 8E cores. If we ignore DDR5, TDPs, PPT, die areas, processes, prices and so on, question is, how AMD can sell their 6C and not look bad in throughput benchmarks? The answer is -> they need to clock the parts as high as possible, potentially giving up power efficiency.

 

[ryanjagtap]

Erm, isn't it the crux of my argument. That AMD has 6 cores, Intel has 6 cores PLUS 8E cores. If we ignore DDR5, TDPs, PPT, die areas, processes, prices and so on, question is, how AMD can sell their 6C and not look bad in throughput benchmarks? The answer is -> they need to clock the parts as high as possible, potentially giving up power efficiency.

That would depend on the V/f curve of the N5P node AMD is using, isn't it? We don't know that yet so we can't know if they are forgoing efficiency yet.

 

[tomatosummit]

Erm, isn't it the crux of my argument. That AMD has 6 cores, Intel has 6 cores PLUS 8E cores. If we ignore DDR5, TDPs, PPT, die areas, processes, prices and so on, question is, how AMD can sell their 6C and not look bad in throughput benchmarks? The answer is -> they need to clock the parts as high as possible, potentially giving up power efficiency.

OR the potential 40% performance uplift makes up for it OR the $200 cpu is now an 8core cpu OR zen4 has 10 and 14core cpus to further diversify the stacks.
Claiming your opinion as fact is the problem.
Even if I think you're half right and it's another year of $250 rrp 6coreX, if it doesn't sell then the market will adjust. If it does sell then why does amd care if they lose out against some e-cores in compile/render benchmarks.

 

[JoeRambo]

That would depend on the V/f curve of the N5P node AMD is using, isn't it? We don't know that yet so we can't know if they are forgoing efficiency yet.

That is what i include in "unknowns", just like exact % performance increase, price etc.

What i am worried about is that we have what I call "marketing retard mark*" embedded in this preview:

5950x is 142W PPT CPU
7950x could be 230W PPT CPU

That would be 62% more peak power. Obviously we don't know, it might be using 100W full core 5.5Ghz CB23 and this 230W PPT is reserved for some future SKU that runs AVX512 y-Cruncher @ 5.5Ghz.
But AMD might also be pushing V/F curve too far and in doing so ruining efficiency for their SKUs to win benchmarks.

Do note that it's impossible to stop good CPU design on leading TSMC node: enthusiasts can take 7950x, underclock it to where V/F is more sensible and enjoy best efficiency and great performance @ say 4.75ghz all core. So what if it looses to 255W equally retarded 13900K by several cb23 points?

 

[TESKATLIPOKA]

The Threadripper parts run at decent clocks, unsure what you are blabbing on about. All of the Zen 3 Threadripper parts top out at 4.5ghz. AMD could push higher, if needed, but at the expense of power. I brought it up to show exactly how far behind Intel is. AMD can squeeze 64 cores into a 280W TDP. Intel can’t.

Threadripper has 4.5GHz a boost, but you won't reach It with all cores within 280W TDP, so why do you even mention It? 4.5GHz would require ~ 456W just for the cores(7.125W/core at ~4.5GHz).

If I leave 256W from 280W TDP to 64 cores, It would mean 4W per core and that would mean probably ~3-3.2GHz clockspeed. Is that supposedly a decent clockspeed or just a very conservative one?
Intel could put 64 cores into 280W TDP(32P + 32E cores). Yes, clocks would be quite low, comparable to what i7 12900 has at PL2 65W(see my above post), but still doable unlike what you said. On the oher hand, size would be a big problem, because unlike Threadripper It would be a monolith. In the end this is a still pretty bad comparison in my opinion.

Writing off Zen 4 because they are possibly releasing a halo SKU that matches Intel’s power limits is just silly. Most, if not all Zen 4 chips will be at much lower TDPs and power limits. You do you, however.

Please enlighten me where exactly am I writing Zen4 off?
I mentioned they will go for max performance without regard for efficiency, at least with the model with 170W TDP.
That doesn't mean there won't be any models with lower TDP or that RaptorLake will be more efficient.
BTW I am eagerly anticipating Phoenix APU(Zen4) based laptop, which will most likely end up as my next system.

 

[coercitiv]

Erm, isn't it the crux of my argument. That AMD has 6 cores, Intel has 6 cores PLUS 8E cores. If we ignore DDR5, TDPs, PPT, die areas, processes, prices and so on, question is, how AMD can sell their 6C and not look bad in throughput benchmarks? The answer is -> they need to clock the parts as high as possible, potentially giving up power efficiency.

@tomatosummit made an important observation by flipping the PoV, which you may want to contemplate on: AMD only needs to sell a 6 core CPU, whereas Intel needs to sell a 6+8 CPU.

IF it's big core matching big core sku cost for cost then it's because customers aren't buying into the extra e cores so they're considered a free addition for a cpu, like an igpu was until the F series.

That's the conundrum Intel will face soon enough with their efficiency cores. I don't remember if I posted this or it simply remained a draft in some thread, but I was joking last week that I'll upgrade to Raptor Lake if they keep offering the marketing cores for free. The E-cores no longer take an insignificant amount of space on the desktop dies, especially if they appear in the smaller die as well. They need to start earning a paycheck or else...

That would be 62% more peak power. Obviously we don't know, it might be using 100W full core 5.5Ghz CB23 and this 230W PPT is reserved for some future SKU that runs AVX512 y-Cruncher @ 5.5Ghz.
But AMD might also be pushing V/F curve too far and in doing so ruining efficiency for their SKUs to win benchmarks.

They will probably do this no matter how well they stack up against RPL at lower PPT. Intel opened the gates of hell and customers said 'whatever', so AMD will definitely do this on some 16-core SKU to show Intel how lava soup tastes like.

At least it will be fun to watch.

 

[Markfw]

There is so much GUESSING going on here from the small amount of information provided, its hilarious. I am not even going to comment about Zen4 yet, aside from the fact that I think they will be great.

 

[PJVol]

BTW I am eagerly anticipating Phoenix APU based laptop, which will most likel

+1
I hope Phoenix G parts do exist in their roadmaps for the future office PC upgrade.
As for desktop, for me personally what was spilled in a preview itself and the Q/A that followed convinces me even more to wait for the AM5 to mature, DDR5 and MBs prices to settle and build a new home PC on zen5 platform.

 

[coercitiv]

aside from the fact that I think they will be great.

is that a guess?

 

[yuri69]

Moore's Law Is Dead said 7-9% IPC

Gotta love them Youtuberz.

MLID timeline of Zen 4:
* May 2021 - IPC 20% (while sources telling being notably higher!)
* May 13th 2022 - IPC 15-24%
* --- May 22nd 2022 --- AMD shows a single benchmark pointing to 5-10% IPC
* today - suddenly IPC goes 7-9%

 

[JoeRambo]

@tomatosummit made an important observation by flipping the PoV, which you may want to contemplate on: AMD only needs to sell a 6 core CPU, whereas Intel needs to sell a 6+8 CPU.

I actually had a though of incoming very funny "battles" in forums, one with very ironic twist on AVX512 for both sides:

We will see influx of forum member posts, where Zen4 side will be championed using some quite heavy FP benchmark, preferably with AVX512 codepath. I think Blender fits this rather well, not sure about AVX512 being enabled in benchmarks.
And this will be matched by equal influx of forum member posts, where RPL will be compared using just CB23, as that is obviously where E cores are giving good help too.

I think given what we already know, for that hypothetical matchup of 6 vs 6+8E we can easily have situation where AMD has no trouble matching and winning Blender, but Intel is dominating them in CB23.
We will find that certain members were in fact running "real world" rendering and fluid dynamics sims all their life (when not idling on their 64C Threadrippers that is) and others were running equally "real world" rendering too.

Combined with obvious suspects of DDR5 prices, sudden focus on PCIE5 M.2 drives i see fun times ahead in forecast for near future.

 

[maddie]

Who exactly is being misleading here? The folks spinning hyperbole. There isn’t any issue with what AMD has shown thus far. Folks getting stuck on the 15% part of greater than 15% and the TDP nonsense. They are missing the fact that these chips are seeing large multicore gains and will likely be the most efficient chips in terms of perf/watt that you can buy.

They take AMD’s messaging to say “minimal performance gains with much higher power consumption” rather than what it really is: “Our new chips not only perform great, but they scale very well.”

Part of that is due to AMD’s poor attempt at messaging. Most of it is due to users having the inability to look at the demos and information for what it is: a preview of an unreleased product. Bias probably also plays into it as well.

I was replying to this from JoeRambo "Bringing 12900K "efficiency" to every ADL discussion is getting old as well."

 

[itsmydamnation]

I'm still not sure what's the difference between TDP and PPT. So PPT is the amount of power the socket can draw, why is that different than the "Thermal Design Power?"

Ignoring overclocking and other controls within PBO,
think of one as instantaneous and one as power of time.

 

[TESKATLIPOKA]

+1
I hope Phoenix G parts do exist in their roadmaps for the future office PC upgrade.
As for desktop, for me personally what was spilled in a preview itself and the Q/A that followed convinces me even more to wait for the AM5 to mature, DDR5 and MBs prices to settle and build a new home PC on zen5 platform.

I don't see why there won't be a Phoenix G for Desktop, but personally I am more interested only in mobile chips.

 

[JoeRambo]

I think of PPT as part of AMDs way of doing turbo, a limit package power draw can't cross no matter what. Kinda PL2 but infinite in time and with complicated dependencies on actual core load and temperatures. So 6C can already chew full PPT of 142W, but 16C workload might hit just 110W package power due to other limits.
Due to these algorithms that are heavily influenced by marketing desire to win benchmarks, the efficiency of larger chips is completely ruined at low-mid load. You get gems like 4-5 core workload using 20W per core for 300mhz gain versus 8C using 12W per core and so on.

That is all stock obviously and in early Zen3 era, MB manufacturers realized that they can lie to chip about how much juice it is actually receiving and that would result into more clocks (and even worse efficiency at wall socket).
Crazy just how huge AMD's advantage was vs Skylake iterations on 14nm for them to pull such gems and still win in efficiency