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!

 

[gdansk]

So you re stating that the 7950X will have way lower perf/watt than a 5950X, that s your "maths", or rather your sound...

For once and for all :

1.35 x the perf at 1.25 x the perf/watt mean 1.35x the perf at 1.35/25 = 1.08x the power.

Is that so difficult to understand..?.

Simply put my thesis is that the CB23 score per watt of 16 Zen3 cores at ~142W may be superior to 16 Zen 4 cores at ~230W.

 

[Ranulf]

Plenty of AMD loyalists won't mind paying that, to support AM5. Value conscious buyers have to look at Intel or AM4. AMD knows the initial production batch will sell out quickly. They will leverage that to get the maximum profits they can.

If these prices are correct then it would be to get the initial market as you say and then when RL is released they have room to drop prices especially during the holidays and when they release the 3D model or models next year.

Initial reviews should be more amusing though if true.

 

[Exist50]

If these prices are correct then it would be to get the initial market as you say and then when RL is released they have room to drop prices especially during the holidays and when they release the 3D model or models next year.

Initial reviews should be more amusing though if true.

I'm sure if that strategy works. My 2 cents is that for anyone who's both willing to choose between the two and able to wait a month (most people), it makes sense to do so to see how everything shakes out. So to get the early buyers, pricing needs to be conservative enough that buyers believe there isn't significant room to undercut. I think these prices are just simply what AMD believes makes sense for themselves. Assuming these prices are real.

 

[eek2121]

Unsure how anyone thinks an Intel chip on an inferior node will have the same or better perf/watt than the AMD chips. This quite simply will not be the case. Stop looking at power limits. Look at what the chip will consume running the benchmark in question. AMD is on a custom N5 process at TSMC. Intel’s process is slightly inferior to N7.

Note that we likely won’t see real numbers until September.

This was in a post of theirs.

"Both the 15th September and the 27th September dates were reported by us first, so in the case AMD decides to proceed with sales on either of the two dates, it will be a confirmation of what we reported."

I beat Wccftech. I called the launch date as the last week of September several months ago. 😂

Is it possible that AMD is holding a speculated 7800X as a 10 core / 20 Thread release at a later date to better match up with the Intel 6+8/20 thread product in the market? While I don't think that N5 yields are so bad that there will be a lot of CCDs that have three under-performing from spec cores, it may be worth it from a marketing perspective. A 2 CCD, 5 cores each, product would have a lot of cache per core (a bit over 7MB if including L2) to keep performance rather high.

I am inclined to believe this is a possibility. N5 has awesome yields however, better than N7, so they would likely have to disable good working cores or wait until they have enough inventory to launch such a product.

 

[gdansk]

Unsure how anyone thinks an Intel chip on an inferior node will have the same or better perf/watt than the AMD chips. This quite simply will not be the case. Stop looking at power limits. Look at what the chip will consume running the benchmark in question. AMD is on a custom N5 process at TSMC. Intel’s process is slightly inferior to N7.

13900K is likely to be behind in perf/watt but it won't be as bad as 12900K vs 5950X. In embarrassingly parallel workloads like CB23 Intel now has more cores. This and voltage changes means they can clock the E cores lower into a better area of their efficiency curve.

AMD switching nodes doesn't make up for the core count deficit and as a result their perf/watt advantage will actually decrease despite Intel being on an old node.

 

[Exist50]

Unsure how anyone thinks an Intel chip on an inferior node will have the same or better perf/watt than the AMD chips.

Who's claiming that? Being more efficient is one thing; being twice as efficient is quite another.

 

[TESKATLIPOKA]

Simply put my thesis is that the CB23 score per watt of 16 Zen3 cores at ~142W may be superior to 16 Zen 4 cores at ~230W.

It most likely will be, but that's to be expected, 62% higher PPT(230/142) won't translate to the same increase in performance, but that's not a very fair comparison to begin with.
As I showed in my previous post, Zen 3 at 5 GHz needs 3.1-3.6x more power than at 3775 MHz, power efficiency goes horribly down.
If we compared Zen 3 vs Zen 4 at the same TDP, then Zen 4 will win in power efficiency.
If we compared Zen 3 vs Zen 4 at the same performance, then Zen 4 should win by a larger margin, because It would need lower clocks.

BTW, I have to wonder, how will Intel manage to allocate power to each type of core.
8 E-cores at 3.7GHz need ~50W -> 16 E-cores will need 100W, so ~150W will be left for 8 P-cores. That wouldn't be a problem, If the clocks were lowered, but from what I saw from leaks, both P-cores and E-cores should be clocked higher.

 

[jpiniero]

If these prices are correct then it would be to get the initial market as you say and then when RL is released they have room to drop prices especially during the holidays and when they release the 3D model or models next year.

I think Zen 4 on AM4 is the backup plan if sales underwhelm. The pricing on that would be interesting.

 

[Joe NYC]

Simply put my thesis is that the CB23 score per watt of 16 Zen3 cores at ~142W may be superior to 16 Zen 4 cores at ~230W.

That would be unexpected.

I think the first 26,196 pts of CB23 will be using less power on 7950x than 5950x, but the additional points may dip deep into the power budget.

So who knows, it may be within the realm of possibilities, but I think unlikely.

 

[gdansk]

That would be unexpected.

I think the first 26,196 pts of CB23 will be using less power on 7950x than 5950x, but the additional points may dip deep into the power budget.

So who knows, it may be within the realm of possibilities, but I think unlikely.

Both at ~142W Zen 4 will be way ahead of Zen 3 in pts per watt. But AMD's performance per watt metric was from even lower clock rates than that (think Genoa).
The final 88W will scale horribly. Giving them maybe only 15% more performance instead of the 60% increase in power it is. And they're doing it because
1) they specified the socket to handle it so they can
2) Intel has more cores
3) AMD/TSMC want to show they have the highest clock rates

 

[Joe NYC]

Initial reviews should be more amusing though if true.

In initial reviews, Zen 4 will beat any corresponding ADL and Zen 3 model in performance and price performance.

We will see what happens when Raptor Lake is released...

 

[pakotlar]

I am not surprised.
Zen 4 won't be cheap. You will need new MB, DDR5 and a new CPU, at least this platform will be with us a few CPU generations.


This thread was closed for a good reason by esquared, so why do you have a need to continue? Please, just stop. BTW this is not meant only to you. Just a few hours left until announcement, so let's wait patiently. Thank you.

Stop what? I asked why Mark thought 13900K would vastly underperform in PPW compared to 7950X, the topic of this thread, and provided the rumoured PPT/TDP’s for AMD and Intel’s top parts to give context for my question. This was on topic at least as much as his comment.

He responded with a factually incorrect statement and called me a troll, which I found hurtful.

Why encourage the bullying and put down my calm on topic discourse?






Yet again, you continue to ignore my warning.


esquared
Anandtech Forum Director

 

[Joe NYC]

Both at ~142W Zen 4 will be way ahead of Zen 3 in pts per watt. But AMD's performance per watt metric was from even lower clock rates than that (think Genoa).
The final 88W will scale horribly. Giving them maybe only 15% more performance instead of the 60% increase in power it is. And they're doing it because

Agreed, that will be the question. How efficient Zen 4 will be using up that last available power envelope. At which point things start to go off the cliff. But that's the problem all processors have.

So it is about where you place the cut off. and how much inefficiency you are willing to tolerate.

After Intel pushed the inefficiency envelope quite high, AMD finally said: "Hold my beer"

 

[Karnak]

https://twitter.com/x/status/1564289007772508161

All posts in CPUs require personal commentary. Not just a link. See http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...f-current-rules-rules-updated-5-4-20.2332179/.

Asministrator allisolm

 

[Joe NYC]

https://twitter.com/x/status/1564289007772508161

10% smaller die size, to about 75.4 mm2 for Zen 4 Durango CCP?

 

[exquisitechar]

https://twitter.com/x/status/1564289007772508161

Great article, as always.

AMD’s client desktop processors remain the weakest in their line-up compared to Intel. Outright performance for the money lags behind while being much more difficult to cool than Intel’s offerings. Ryzen 7000’s 5nm process significantly increases production costs and risks eroding margins as it is difficult to increase ASPs commensurately in such a competitive environment. Mandatory DDR5 on the new AM5 platform will also have some impact on uptake due to a much higher base platform cost.

Basically exactly my thoughts on Raphael. For anyone who has a modern CPU, this generation presents an extremely unappealing potential upgrade, especially when the long awaited Zen 5 is coming next. Expensive platform, expensive CPUs, expensive RAM, expensive cooling, unimpressive performance.

 

[TESKATLIPOKA]

Stop what? I asked why Mark thought 13900K would vastly underperform in PPW compared to 7950X, the topic of this thread, and provided the rumoured PPT/TDP’s for AMD and Intel’s top parts to give context for my question. This was on topic at least as much as his comment.

He responded with a factually incorrect statement and called me a troll, which I found hurtful.

Why encourage the bullying and put down my calm on topic discourse?

esquared already closed this thread, I don't want him to do It again, that's why I said to stop. I have no problem with you or anyone else on this forum.
If you have a problem with other user(s) or moderator(s), then resolve or complain directly to esquared.






Do not play moderator on this forum. If you have an issue with a post or poster, user the Reported post function.


esquared
Anandtech Forum Director

 

[tamz_msc]

https://twitter.com/x/status/1564289007772508161

Huh, if it is limited by thermals, then there better be advances in the interface between the die and IHS - like Intel does with the i9-K chips. Also, that article doesn't really mention the major area where the transistor budget was spent in my opinion - increasing clock speeds. AVX 512 is a part of the investment, but it isn't as big as people think it is.

 

[Saylick]

https://twitter.com/x/status/1564289007772508161

Gotta love a good Angstronomics article (is that redundant given how consistent they've been?) these days. WCCFTech could learn a thing or two from them, chiefly in how to keep an article concise and to the point. More than half of the words in every WCCFTech articles are just links to their other, overly wordy articles.

 

[Exist50]

Ouch. They seem pretty pessimistic. I'm curious how the thermal thing will pan out in reviews, and how the density impacts cooler selection.

 

[Zucker2k]

Also, if thermals are going to be a major hinderance that doesn't bode well for Zen4 X3D.

I'm still treating all these as rumour until official AMD numbers drop in a few more hours.

Edit: This is also going to mean AMD basically pulling an Intel by clocking their chips so high, at the expense of power and thermals in order to remain competitive against the shear throughput of Intel's e-cores. I have a feeling once RPL is released, skeptics are going to develop a better appreciation for the upgraded Gracemont cores in RPL.

 

[inf64]

Yeah, it seems that things became bleak really fast. If the chips can perform well only under the best cooling solutions then this brings cost up even more versus Zen 3 and intel offerings. Let's hope the performance is there.

 

[Karnak]

Well tbh I don't see a problem in "the better the cooling the better it performs". That's also nothing new. Slap a waterblock on your NV GPU and it'll clock higher due to lower temps.

 

[TESKATLIPOKA]

I am glad Phoenix Point is a monolith + limited TDP.

 

[Joe NYC]

Well tbh I don't see a problem in "the better the cooling the better it performs". That's also nothing new. Slap a waterblock on your NV GPU and it'll clock higher due to lower temps.

AIO coolers are now pretty standard for gaming set ups.

But it will be interesting to see how it gets tested in independent reviews.