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

[Ajay]

Lets see, I think Intel showing off their roadmap this month is going to make AMD share more info, however vague it may be.

I hope so. Getting sick of the 'cones of silence'.

 

[eek2121]

https://twitter.com/i/web/status/1449331716976435203

Isn't this the 128 Core CPU.
Seems ExecuFix is saying it is Zen4 based.

But from leaked manual you can see

View attachment 51513

CCD: 0 to 11 (12 CCDs)

From Family 19H PPR


View attachment 51515

Either this is a 96 Core, or it is a new model which was not part of the leak.

I was looking at this earlier, and I remember seeing this a while ago. FYI, this SKU was supposed to pop out after Genoa. From what I understand, current plans are to launch around the time that Zen 4 desktop chips launch (roadmaps haven't been leaked, but the original leak had this coming out in July/August). This may possibly refer to the 24-32c Zen 4 desktop chip which may, or may not launch, however (as opposed to 16 cores or less).

AMD has been both closing up leaks (because they can) and moving stuff around (due to competition), so I don't actually know what to expect.

 

[ahimsa42]

"AMD Ryzen 7000 ‘Raphael-H’ Next-Gen & High Performance Laptops Rumored To Feature Up To 16 Zen 4 Cores"

AMD Ryzen 7000 'Raphael-H' Next-Gen & High Performance Laptops Rumored To Feature Up To 16 Zen 4 Cores

AMD's next-generation high-performance Ryzen laptops featuring the Raphael-H CPUs are expected to rock up to 16 brand new Zen 4 cores.

wccftech.com

 

[DrMrLordX]

That's interesting. We barely have any info on Raphael and they're already teasing a laptop version of it with an -H designation? What are they really up to?

 

[moinmoin]

Intel has been pushing higher core count desktop dies into laptops for quite some time, so it makes sense for AMD to follow suit with their already more efficient desktop products there. Stumbling block so far had been the lack of iGPU in the desktop MCMs which is what Raphael was said to change. So an -H designation of Raphael would be an almost boring consequence of all that.

 

[Timorous]

That's interesting. We barely have any info on Raphael and they're already teasing a laptop version of it with an -H designation? What are they really up to?

Do you need a portable workstation? Why settle for big.Little when you can go big or go home!

 

[DrMrLordX]

Do you need a portable workstation? Why settle for big.Little when you can go big or go home!

Also probably AMD's attempt to keep up with M1X in the mobile segment.

 

[leoneazzurro]

It depends what the "mobile segment" is. M1pro and Max are priced way too high for attacking the market where AMD is. Not speaking abut the fact that Laptop gaming is a solo competition fo rWindows machines.

 

[DrMrLordX]

It depends what the "mobile segment" is. M1pro and Max are priced way too high for attacking the market where AMD is. Not speaking abut the fact that Laptop gaming is a solo competition fo rWindows machines.

All Mac hardware is overpriced. The M1X lappies look targeted at the high-end professional set, so there you have it. Putting what is essentially a desktop chip into a laptop form factor (which is what AMD would appear to be doing with a Raphael -H product) would be the best way to compete with that, assuming too many people become interested in Apple for their hardware.

Still not sure how AMD is going to reconcile the SERDES link overhead in a mobile unit but hey, maybe they have something different in mind.

 

[leoneazzurro]

Quite frankly, AMD could easily create something like the M1Max, after all they already created the Xbox and PS5 APUs, a year ago. And they did because there were customers asking for that. In PC laptop world, unfortunately, this did not happen yet. Maybe Apple will create a trend also for the WIndows laptops, who knows?

 

[Ajay]

As for as prices, these Macs aren’t just DTRs, they are really professional grade workstations in a laptop form factor, so the prices, while high aren’t as bad as they seem in that context. I don’t know if we even have pro grade laptops like this in the x86 world.

 

[Ajay]

Quite frankly, AMD could easily create something like the M1Max, after all they already created the Xbox and PS5 APUs, a year ago. And they did because there were customers asking for that. In PC laptop world, unfortunately, this did not happen yet. Maybe Apple will create a trend also for the WIndows laptops, who knows?

Yes, except the consoles eat much more power- so, actually, no . It is an interesting idea though - certainly console SoCs would be much closer to being pro grade workstation, if put into a laptop.

 

[dr1337]

Yes, except the consoles eat much more power- so, actually, no

on 7nm without slc and much faster vram

 

[BorisTheBlade82]

@DrMrLordX
The SERDES is what also keeps me thinking. So maybe Zen4 could finally be the time they at least on mobile could use something like Info-LSI (TSMCs Version of EMIB, but better). This would improve Interconnect power efficiency by around 1000% and would make chiplets in the mobile space viable. But the leaks suggested that they seem to stay on the organic package Interconnect at least on the Desktop. So for me it is rather improbable that it was technically feasible with one and the same die. But I keep my fingers crossed.

@leoneazzurro
If you mean that AMD could stick a large GPU into a SoC and feed that with high bandwidth memory, then clearly yes. If you mean them to reach the same power efficiency, then definitely no. The margin is much too big for them to catch up within one generation.

 

[Ajay]

on 7nm without slc and much faster vram

Fair point, so, time to step up AMD . Seriously, in a time of GPU shortages, getting a console like SoC into a PC, even with limited upgradability, might be able to find a large enough market among PC gamers.

Oh, and while Zen doesn’t have SLC, they do have very large L3 caches. Different strokes and all..

 

[leoneazzurro]

@leoneazzurro
If you mean that AMD could stick a large GPU into a SoC and feed that with high bandwidth memory, then clearly yes. If you mean them to reach the same power efficiency, then definitely no. The margin is much too big for them to catch up within one generation.

Don't forget the node of advantage for M1X (which does not feed out of thin air, however, you can easily see that battery life went down with battery size being the largest seen in a laptop). But consoles are not optimized for power. Rembrandt will get 12 CU and 8 Zen3+ cores in 18W, on N6. So it's not too far fetched to think about a 50-60W APU on N5, with four times that graphic power (48CU), Infinity cache to help with bandwidth, and the same Zen3+ cores, with a slight bump in frequency. How does that compare to a M1Max?
Don't get me wrong, Apple made a great piece of technology and for its target it is a wonderful part. But I don't think that AMD's (or Intel's, which had management problems instead of lack of technical capability) are idiots, they gave proof of being able to create good products as well.

 

[BorisTheBlade82]

Badly. As stated in the Apple M1 Thread their efficiency margin is 500%. None of the things you stated will change that in a significant way.

 

[leoneazzurro]

Badly. As stated in the Apple M1 Thread their efficiency margin is 500%. None of the things you stated will change that in a significant way.

I think you are too much in the marketing hype. Such an APU would have better graphics power than M1Max and probably worse CPU power, but that could be tuned, and power draw would be similar, because if you think M1Max works with 15W power, well.. you're plainly wrong. And what does even "efficiency margin 500%" mean, and what is the base, in which applications, and again you are comparing a year's old APU on a worse process node, and in an environment AMD does not fully control because hardware configuration other than the APU itself are from OEMs. This makes no sense.

 

[DisEnchantment]

These Console GPUs are very narrow, they traded Efficiency for die area/cost and the consoles sell for less than 500 USD
On top of this, they are meant to run at high clocks to increase the throughput of the command processor, geometry processor etc which is conducive for the game pipelines for which they are designed for.
They are made to run games, not for compute throughput. If you have some RDNA2 GPU, try downclocking them, they are very efficient too.
What is even the comparison? matrix math, triangle output, ray tracing, tensor operations, nobody knows
Also, N5P over N7 is a 43% efficiency gain, which is a lot.

Also why bring Apple PR in this thread, there are lots of such threads.

 

[DisEnchantment]

Nothing will change in the Windows/Linux ecosystem if no OEM wants to step up their game. If they need, AMD/Intel/QCMM whoever will make the HW.
Windows ecosystem will be on the receiving end, if MS does not take matters in their own hands.
I am kind of happy about the prospect of seeing the likes of Dell losing customer base.
If left to the OEMs, the likes of Dell will be more than happy to hamstring competition as long as they get some kickbacks/short term gains.
I recollect someone on here debating against me why it is not possible to have a thin and light laptop for engineering work and I should take whatever brick I can get , and here we are everyone debating about how the PC ecosystem lagged behind.

I recollect some days ago, that Hallock mentioned it would be better (for AMD) to go with a single core architecture instead of big.LITTLE because of complexities to be addressed in SW.
The scheduling issue with Win11 is an example how hard it is for AMD (or Qualcomm) to design something else.
For the Windows ecosystem at least.
Another one is the Hyper-V support on Ryzen.
I am sure none of the AMD SEV features will ever come to Windows (Server)

At least on Linux side there is more freedom, you can propose changes to the kernel based on the HW features you develop and that is what is happening with the countless Arm vendors, AI accelerator, x86 etc.
Steam Deck is a prime example.

 

[BorisTheBlade82]

@leoneazzurro & @DisEnchantment
Neither am I on a hype train nor did I bring PR to this thread. I just replied to some over optimistic person believing that a Zen3+ in some new process would be enough to overcome the gap to Apple.
It still gives me headaches that people do not realize how far ahead Apple is. And it is lying there in plain sight since November 2020. Anandtech have published several articles in order to explain this. Firestorm as a ST core is as fast as the best Intel and AMD have to offer while at the same time only using 1/5 of the power. After that everything else is just a matter of scaling.
AMD and Intel need to change the very foundations of their designs in order to catch up.

 

[moinmoin]

Also probably AMD's attempt to keep up with M1X in the mobile segment.

It isn't.

Still not sure how AMD is going to reconcile the SERDES link overhead in a mobile unit but hey, maybe they have something different in mind.

I doubt it needs to reconcile that. Even with the overhead a Raphael-H will already be more efficient than the direct competition. And the direct competition so far has been DTR chips of the likes of 10980HK, 11980HK and likely 12980HK and 13980HK unless Intel changes the model names.

Apple definitely is the competition to beat in efficiency (I've stated that years ago already). But with Zen 4 AMD won't have a direct answer to any M1 variant, and also doesn't need one unlike Intel which wants to win back Apple as a customer.

 

[eek2121]

@leoneazzurro & @DisEnchantment
Neither am I on a hype train nor did I bring PR to this thread. I just replied to some over optimistic person believing that a Zen3+ in some new process would be enough to overcome the gap to Apple.
It still gives me headaches that people do not realize how far ahead Apple is. And it is lying there in plain sight since November 2020. Anandtech have published several articles in order to explain this. Firestorm as a ST core is as fast as the best Intel and AMD have to offer while at the same time only using 1/5 of the power. After that everything else is just a matter of scaling.
AMD and Intel need to change the very foundations of their designs in order to catch up.

The new Apple chips are not as nearly as efficient as you are being lead to believe. All of the performance IS due to a node advantage. You can somewhat guesstimate where Zen 3 would be if it were on 5nm, but as I have stated before, it matters not. Macs are closed systems. Apple does not need to support desktop DDR4 or 20 PCIE 4 lanes. They don’t need to maintain compatibility with 20 year old instruction sets. Most of the software of the world does not run on a Mac of any variety.

AMD chips are a completely different beast.

Back to Zen 4, I think in the long run, we will see 8, 12, and 16 core H variants. With ADL-S and it’s successor bringing 10+ cores to mobile, they will need high end chips to stay in the game.

 

[BorisTheBlade82]

@eek2121
Should I maybe point that I have never ever mentioned M1Pro/Max specifically. I am talking about the Firestorm core of Apple Silicon in general. Because therefore I have hard numbers (AKA facts). Apple is 500% more efficient while 5nm brings smaller 150% - end of story, not?

@moinmoin
IMHO that might be overly confident as TGL-8C is not that bad in comparison to Cézanne. While on the other hand Vermeer is really bad, efficiency wise wrt ST/light load.

 

[leoneazzurro]

Should I maybe point that I have never ever mentioned M1Pro/Max specifically. I am talking about the Firestorm core of Apple Silicon in general. Because therefore I have hard numbers (AKA facts). Apple is 500% more efficient while 5nm brings smaller 150% - end of story, not?

Quite franlky, not. Link these data, the conditions in which these data are taken, and look also not at the ST score (easily to throw completely off by measuring different things) but at the MT score, which is more representative of the real performance of the cores.