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!

 

[szrpx]

This is only guessing as neither has been actually released, but I would probably bet on Intel in networking and telecommunication - QuickAssist really helps for some loads, especially if you are power constrained (or aim at efficiency) as accelerators are usually burning less power for same job.

At the same time, who knows what kind of surprises in this area will Genoa bring. Also ARM based SoCs are taking their piece of pie, so it's gonna be hard for Intel in coming years.

Thing is, Quickassist works on AMD systems too.

 

[Timmah!]

Seriously, unless they sacrifice a blue eyed virgin blonde the moment she hits puberty, they are not getting lucky any time soon. Anyone hoping for an Intel resurgence should be aware that AMD isn't sitting still. Resting on their laurels is not something they can do because they have a lot more to lose than Intel. They will continue to maintain their architectural lead, unless TSMC falters really badly and Intel finally manages to get back on track on their process roadmap.

Intel does make great products still (Alder Lake is honestly a surprise for me, especially the P-cores) but AMD has shown that they can fight fire with fire while using less gas

Maybe. Or maybe they just need process node as small and efficient as TSMCs one, and maybe that next one of theirs (4?) will be the one. And then they will be able to produce comparable products to AMD. Can we guess how much power would Alder/Raptor Lake draw if it was produced on TSMC 5nm instead of Intel crappy 10nm node?

 

[igor_kavinski]

Can we guess how much power would Alder/Raptor Lake draw if it was produced on TSMC 5nm instead of Intel crappy 10nm node?

The real question is, can Intel engineers do that on someone else's node? Problem with Intel is, they think they do better than anyone else, even when it's a misstep. They are so smug about their abilities that their overconfidence cost them years and years of being stuck on 14nm. They refused advice from outsiders with proven track records and even drove them out (Jim Keller). I cannot see any miracles happening unless they launch a real product on Intel 4 that I can actually buy. Even then, they would be equivalent to TSMC 7nm and not 5nm. Unless, they think their 7nm is so superior that it can give TSMC 5nm a run for its money.

And oh, they have yet to find success with Intel 7 in the server space (when is SPR coming out again?)

 

[Timmah!]

The real question is, can Intel engineers do that on someone else's node? Problem with Intel is, they think they do better than anyone else, even when it's a misstep. They are so smug about their abilities that their overconfidence cost them years and years of being stuck on 14nm. They refused advice from outsiders with proven track records and even drove them out (Jim Keller). I cannot see any miracles happening unless they launch a real product on Intel 4 that I can actually buy. Even then, they would be equivalent to TSMC 7nm and not 5nm. Unless, they think their 7nm is so superior that it can give TSMC 5nm a run for its money.

And oh, they have yet to find success with Intel 7 in the server space (when is SPR coming out again?)

All that is true, what i am saying though, things can be very different in 3 years time. Before Zen 1 dropped, AMD catching up to Intel looked almost impossible as well. Yet here we are today.
Intel SPR issues might be more chiplet/tile related, as its basically their first product of that kind, or maybe something about security. Does not need to be process related.

 

[igor_kavinski]

Before Zen 1 dropped, AMD catching up to Intel looked almost impossible as well. Yet here we are today.

Very true.

 

[Exist50]

I cannot see any miracles happening unless they launch a real product on Intel 4 that I can actually buy. Even then, they would be equivalent to TSMC 7nm and not 5nm.

Intel 4/3 is comparable to TSMC 5/4, not N7...

 

[igor_kavinski]

Intel 4/3 is comparable to TSMC 5/4, not N7...

Really? So Intel 7 is comparable to TSMC 7nm? What explains the starkly different temperature characteristics?

Intel Core i9-12900K Review - Fighting for the Performance Crown

The Intel Core i9-12900K is Intel's flagship processor for the Alder Lake architecture. In our testing, we saw fantastic gaming performance from this new processor. Not only low-threaded tests have improved, the 12900K can even beat AMD at highly threaded workloads.

www.techpowerup.com

5950X base 3.4 GHz / boost 4.9 GHz
12900K base 3.9 GHz / boost 5.2 GHz

Only 500 Mhz higher on base and 300 MHz higher on boost results in a 35 degrees higher temp for ADL? Has anyone done a temperature comparison at iso-base/boost speeds for these two?

 

[Exist50]

Really? So Intel 7 is comparable to TSMC 7nm? What explains the starkly different temperature characteristics?

Intel Core i9-12900K Review - Fighting for the Performance Crown

The Intel Core i9-12900K is Intel's flagship processor for the Alder Lake architecture. In our testing, we saw fantastic gaming performance from this new processor. Not only low-threaded tests have improved, the 12900K can even beat AMD at highly threaded workloads.

www.techpowerup.com

5950X base 3.4 GHz / boost 4.9 GHz
12900K base 3.9 GHz / boost 5.2 GHz

Only 500 Mhz higher on base and 300 MHz higher on boost results in a 35 degrees higher temp for ADL? Has anyone done a temperature comparison at iso-base/boost speeds for these two?

I do hope that you're not seriously claiming that the temperature two starkly different pieces of hardware can reach is indicative of anything about the process...

 

[igor_kavinski]

I do hope that you're not seriously claiming that the temperature two starkly different pieces of hardware can reach is indicative of anything about the process...

Not claiming. Just trying to understand. So Alder Lake is a bad design architecturally that is making Intel 7 look bad?

 

[nicalandia]

AMD 7900X vs Intel 7900X.... Fight...!!!

 

[Exist50]

Not claiming. Just trying to understand. So Alder Lake is a bad design architecturally that is making Intel 7 look bad?

I'll continue this in more detail in a more appropriate thread if you want, but suffice it to say that temperature is a very complicated combination of process, architecture, design, boosting behavior (firmware, power limits), floorplanning, packaging, cooler selection, and probably other things I'm forgetting. There're simply far too many variables to derive any useful conclusion about the process metrics from temperature alone.

If anything, density has been scaling faster than power has, which makes thermals a bigger concern on the newer, better nodes.

 

[fkoehler]

And AMD, IBM, Nuvia, Qualcomm are all just going to sit on their butts and wait for Intel to catch up?

Intel does seem to spend on deep research, so there is a real possibility they could have something pop in the R&D labs, photonics maybe.
But thats hoping for a hail mary vs steady tick-tock cadence forwards by TSMC.
Its just as likely that Intel ends up like Kodak.

Maybe. Or maybe they just need process node as small and efficient as TSMCs one, and maybe that next one of theirs (4?) will be the one. And then they will be able to produce comparable products to AMD. Can we guess how much power would Alder/Raptor Lake draw if it was produced on TSMC 5nm instead of Intel crappy 10nm node?

 

[PJVol]

So Alder Lake is a bad design architecturally that is making Intel 7 look bad?

Temps is mainly the function of power and its dissipation area.
There's always compromises between power and performance, so I see the main design challenge is to lower leakage currents (or dI/dt), as well as to minimize the switching activity for a certain workload. The latter is more relevant, since it's a major contributor to chip's energy efficiency (leaving leakage power loss as overhead)
Could it be Intel did prioritized performance in favor of energy efficiency? Or may be E-cores turned out to be less effective than they deemed to?

 

[IntelUser2000]

Even then, they would be equivalent to TSMC 7nm and not 5nm. Unless, they think their 7nm is so superior that it can give TSMC 5nm a run for its money.

You do make it sound like you are claiming even though you say you aren't, just saying.

Process is a tool, a very important tool but one regardless. Let's compare it to cooking. You can have same tools, and same ingredients, but the amount and quality of food that can be cooked vary vastly right?

We speculated for years the Cove design team needs major improvements because they are behind the industry. We can also see that by comparing theirs to Intel's small core team.

 

[biostud]

In most cases performance/watt would be the most important factor when comparing processors or process nodes. Intel even did that until they started losing

For me it is the best way to see who has the "best" node/design.

It is important very in laptops and servers, but maybe not that important in a gaming rig.

How many would choose a RPL based laptop over a zen4 based, if everything else is equal?

 

[nicalandia]

X Shapped Thermal paste Applied...

 

[JustViewing]

Isn't the purpose of thermal paste is to fill the air gap and not to completely block metal to metal contact?

 

[eek2121]

That is fairly accurate, providing you can keep the chip cool.

Note: I'm not talking about having to increase PPT, I mean just keeping at boosting at 230W.

The 7700X has a 142W PPT does it not? Only the dual chiplet products are 230W.

Intel 4/3 is comparable to TSMC 5/4, not N7...

Intel 4 (unless they water it down) is actually comparable to TSMC N3. Slightly better even.

Really? So Intel 7 is comparable to TSMC 7nm? What explains the starkly different temperature characteristics?

Intel Core i9-12900K Review - Fighting for the Performance Crown

The Intel Core i9-12900K is Intel's flagship processor for the Alder Lake architecture. In our testing, we saw fantastic gaming performance from this new processor. Not only low-threaded tests have improved, the 12900K can even beat AMD at highly threaded workloads.

www.techpowerup.com

5950X base 3.4 GHz / boost 4.9 GHz
12900K base 3.9 GHz / boost 5.2 GHz

Only 500 Mhz higher on base and 300 MHz higher on boost results in a 35 degrees higher temp for ADL? Has anyone done a temperature comparison at iso-base/boost speeds for these two?

The base clock of the 12900k is actually 3.2ghz. The base clock of the 13900k is 3ghz.

As far as your question is concerned, it is mostly an Apples to Oranges comparison. Intel 7 does appear to be mostly comparable to TSMC N7. If I had to guess, I would say TSMC has a slightly superior process, but this is based on publicly released products.

Zen 3 was also heavily optimized for performance, so it is possible that Intel simply did not care to optimize their products as much as AMD.

That is why both Meteor Lake and Arrow Lake will be important for Intel, and why AMD should absolutely not fall asleep on the job. Meteor Lake will be on Intel 4, so AMD won’t have a process advantage. Instead it will come down to “who has the better design?”

Competition is glorious.

 

[DrMrLordX]

Isn't the purpose of thermal paste is to fill the air gap and not to completely block metal to metal contact?

There will be gaps with something bad like air in them. You want TIM in those gaps. The thinner the layer, the better, but you need TIM in there. It effectively blocks all metal/metal contact. But that isn't necessarily a bad thing.

 

[Kaluan]

Intel 4 (unless they water it down) is actually comparable to TSMC N3. Slightly better even.

Neither of those nodes are available out in the wild and in the case of Intel, def not in the hands of any 3rd party.

How you can say such thing is beyond me. Nodes are more than just about density and Intel can parade 2-2,4x as much they want, we know Intel 4 will be garbage. Why else would they need to follow it up/replace it with Intel 3 immediately after?

I also think some people need to look up when was the last time a Intel node was used in a low power or power efficient design (or worse, in a successful/innovative one)... I can wait. Even Samsung probably beats them here.

Last but not least, they seem more expensive than TSMC too. Even tho capacity may not be a problem for them. Intel Alchemist anyone?


Also, what's up with focusing on what Pat's saying? I don't trust 8 out of 10 things Lisa says, despite her great track record and non-petty attitude. Let alone those 2 other clowns.

 

[uzzi38]

The 7700X has a 142W PPT does it not? Only the dual chiplet products are 230W.

Yeah, but you also get half the area for thermal dissipation between CCD and IHS...

 

[TESKATLIPOKA]

I hope this wasn't posted yet.
AMD Ryzen 5 7600X tested in Cinebench R23 with AMD Core Performance Boost disabled

	R5 5600x(Link)	R5 7600x (CPB Off)	R5 7600x (CPB On)
CB R23	11225	13003	14767
Power consumption	~76 W ?	60 W	110 W
Performance gain	100 %	116 %	132 %
Power draw	100 %	79 %	145 %
Power efficiency	100 %	147 %	91 %

If this will hold true even for retail version, then as I thought, If you want max performance you will have to sacrifice some efficiency.
But 110W in CB R23 is better than I expected based on the rated 105W TDP(142W PPT), I expected 120-125W.
If you don't need max performance, then efficiency is much much better than Zen3.

P.S. If someone has 5600x, then please post his package power during CB R23. I am not sure, If 76W is accurate.

edit: removed the clockspeed.

 

[deasd]

As you see, when enabling CPB it has ~16% clock uplift(5456/4705) but only has 13.56%(14767/13003) R23 score gain. I would rather wait for the BIOS being more stable. Also the temperature is not good in this leak from VCZ. (too high voltage profile lead to throttling?)

 

[inf64]

As you see, when enabling CPB it has ~16% clock uplift(5456/4705) but only has 13.56%(14767/13003) R23 score gain. I would rather wait for the BIOS being more stable. Also the temperature is not good in this leak from VCZ. (too high voltage profile lead to throttling?)

5456Mhz is ST clock gain, not MT . Everything looks ok score wise, all core boost is great, around 5.3Ghz.

 

[nicalandia]

If this will hold true even for retail version, then as I thought, If you want max performance you will have to sacrifice some efficiency.

The Performance/Efficiency Champs for Zen4 will be on Servers and Mobile segments where Intel can't just push the cores due to Space and TDP Constrains.