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!

 

[maddie]

Zen3 Memory subsystems is actually much better/faster than Golden Cove(As far as L1/L2/L3 Cycle/ Latencies). Alder lake had many advantages over Zen3, Like Speed and DDR5

Zen4 has effectively reach parity with Golden Cove/Raptor Cove in Speed and DDR5. Hence it will take the lead in Memory sensitive apps like games, compression/decompression etc. ...

It also seems that the DDR5 implementation will be better.

 

[Abwx]

This has nothing to do with what I was asking. @inf64 was talking about memory sensitive apps. Do you honestly think CB is memory sensitive?

Memory sensitive apps are games, compression/decompression etc.... I want to know why he thinks Zen 4 will "easily eclipse" Raptor Lake in those sorts of applications.

Games and loads such as decompress/compress use integer instructions, dunno what Zen 4 will do in thoses matters but it should do well because it s designed for servers loads, and thoses ones are integer instructions based.

That being said what i said about Cinebench will be accurate for thoses tasks as well, 7ZIP should be improved by the same >35% ratio while games will benefit from hugely increased RAM bandwith that was a limiting factor with Zen 3, as demonstrated by the 5800X3D.

 

[HurleyBird]

The only problem with your post is that MLID claims Zen 4 with Vcache is 30% faster than Zen 4 without Vcache. So the comparison is not versus Zen 3 Vcache parts. This puts Zen 4 on a whole other performance level.

To be precise, he claims there's an internal benchmark showing that, not that it's indicative of a general performance uplift. Given the Zen3D numbers on the same benchmark, I'd take it with a grain of salt, although it may be more indicative of gaming performance than general, assuming it's real to begin with.

That said, again assuming the benchmark is real, it does call into question why Zen 4 benefits so much more. After all, Zen 4 has a much better memory subsystem so one would think it would tip the other way. If this is true, and not some weird edge-case outlier, then 2nd gen v-cache must have some extra special sauce besides working at higher voltages.

 

[Carfax83]

Zen3 Memory subsystems is actually much better/faster than Golden Cove(As far as L1/L2/L3 Cycle/ Latencies). Alder lake had many advantages over Zen3, Like Speed and DDR5

Zen4 has effectively reach parity with Golden Cove/Raptor Cove in Speed and DDR5. Hence it will take the lead in Memory sensitive apps like games, compression/decompression etc. ...

You haven't been keeping up with Raptor Lake leaks. Raptor Lake has a significantly improved memory subsystem and memory controller. You guys honestly think Intel was not going to improve the weakest part of Alder Lake? LOL!

I think you guys are in for a rude awakening.

 

[Markfw]

You haven't been keeping up with Raptor Lake leaks. Raptor Lake has a significantly improved memory subsystem and memory controller. You guys honestly think Intel was not going to improve the weakest part of Alder Lake? LOL!

I think you guys are in for a rude awakening.

First, I think the weakest part of Alder lake is the e-cores. Second, even if you are right, its a question of who improved it more. Well, in a month or so we will see.

 

[eek2121]

Yeah I don't know about that, whenever I see a benchmark have results up for something like a 5950X varying from 9K to 17,5K points I can't think of anything but how garbage it is.

Particularly when almost all other benchmarks out there give much more consistent and clear results... and can also be ran in sustained mode, a feature which GB is sorely lacking (given it's mobile SoC focus you'd think throttling/anti-artificial boosting measurements would also be part of it's test suite but no... we're supposed to belive a 30s benchmarks is the be-all end-all of precise performance gauging lol)

And even in ideal scenarios, it's misleading at best. Take 5800X v 12600K results, 5800X typically scores 10xxx points, 12600K scores 11xxxx for DDR4 and 12xxx for DDR5. But taking a quick look at cumulative application performance on review outlets like TPU or ComputerBase, 5800X sits in front of 12600K by a few percent, not 10-20% behind, as GB would make newbies belive.

If anything, I feel like it doesn't get ENOUGH criticism. Still, I hope v6 might pleasantly surprise me one day.

Yeah, GB sucks because the variability of results. You bench your rig, go on to look at the comparable results and uh-oh, they are all-over the place, so you have no clue, whether your rig is working as expected or not. Completely misses the point then.
Cinebench on other hand might only be single specific kind of workload, so it should definitely not be taken be all end all either, but at least, from what i could remember, it was always fairly consistent, with the results.

ANY benchmark that makes results public will have this issue. It isn’t the benchmark that is the problem, but rather the users.

Want me to show my CBr23 results while doing a blender render?

 

[tamz_msc]

They state >35% better perf and >25% better perf/Watt in Cinebench R23.

Using Computerbase CB score wich is 26193 pts this imply at least 35360 pts and power being at most (142 x 1.35)/1.25 = 153.36W.

FTR Computerbase measurement show the 5950X using 14W less (at the main) than a 5900X or a 5800X, so the actual power of the 16C is about 130W, making the estimation above even more easy to get by AMD.

You're forgetting that according to rumors the 7950X is 170W TDP, while the comparison you're making assumes it'll be 105W TDP.

A better comparison would be the latest leaked 7700X against the 5800X, which is supposed to be 105W TDP, the same as 5800X. According to vidoecardz the 5800X scores 6119 pts in R20(R23 is the same except for the scoring system and apple silicon support), while the 7700X scores 7701. 7701/6119 = 1.259. So AMD's claim of >25% PPW improvement in Cinebench is valid.

However, the comparison you made is simply wrong. The 7950X isn't going to score 35-36000 points while consuming 150W PPT.

 

[TESKATLIPOKA]

A better comparison would be the latest leaked 7700X against the 5800X, which is supposed to be 105W TDP, the same as 5800X. According to vidoecardz the 5800X scores 6119 pts in R20(R23 is the same except for the scoring system and apple silicon support), while the 7700X scores 7701. 7701/6119 = 1.259. So AMD's claim of >25% PPW improvement in Cinebench is valid.

It's a pity we don't know, If TDP was 105W or less(more).
I will think optimistically and say TDP was lower.
Hopefully these gains will be also true for mobile(<=55W) and not just for desktop.

 

[inf64]

Greymon is hinting at 5.6Ghz all-core Turbo for 7600X?

https://twitter.com/x/status/1562707438885048320

 

[Gideon]

Greymon is hinting at 5.6Ghz all-core Turbo for 7600X?

https://twitter.com/x/status/1562707438885048320

It is almost certainly an overclock.

Even then, running 5.6 all-core on a SKU that's not the best binned one, is excellent news! So hopefully it's true.

 

[leoneazzurro]

It's a pity we don't know, If TDP was 105W or less(more).
I will think optimistically and say TDP was lower.
Hopefully these gains will be also true for mobile(<=55W) and not just for desktop.

Especially as the 3700X and 5700X parts were 65W parts and not 105W parts (these were the 2700X and the 3800/5800X). If this is true also for Zen4, and the 7950X is indeed a 170W part, it could lead to interesting results.

 

[inf64]

And another tweet from Retired Engineer (translated Digitimes article):

https://twitter.com/x/status/1562609696976302088

 

[uzzi38]

It is almost certainly an overclock.

Even then, running 5.6 all-core on a SKU that's not the best binned one, is excellent news! So hopefully it's true.

Seriously - the main bottleneck for Raphael is cooling. I really can't stress this enough.

 

[biostud]

At the investor meeting the zen4 with 3D cache was set to launch this year, but do you believe this to be changed or can launched also be a paper launch?

 

[Abwx]

However, the comparison you made is simply wrong. The 7950X isn't going to score 35-36000 points while consuming 150W PPT.

What is wrong is your inability to understand what >35% better perfs and >25% better perf/watt mean and that it s the number for the 7950X vs 5950X as stated by AMD.

Edit : The 5950X run at 130W in Cinebench, that s 10% below the 142W PPT, 153W happen to be just 10% below 170W PPT...

If the improvements stated by AMD were at 170W PPT vs 142W PPT then it would score 39188 pts...

 

[Vattila]

What is wrong is your inability to understand

Educational tip: Rather than attack the pupil, patiently show him the calculation (again, in more detail).

 

[Abwx]

Educational tip: Rather than attack the pupil, patiently show him the calculation (again, in more detail).

The calculations were shown, it s just that the guy is desperatly hoping for them to be wrong..

They state >35% better perf and >25% better perf/Watt in Cinebench R23.

Using Computerbase CB score wich is 26193 pts this imply at least 35360 pts and power being at most (142 x 1.35)/1.25 = 153.36W.

Here another way to get to the same numbers.

According to AMD 5NP has 2 x the perf/watt of a vanilla N7.

At 4GHz, wich is the frequency of a 5950X in Cinebench, A 7950X would use at most 75W and have 10% better throughput/Hz.

Assuming a cubic relationship for power/frequency from 4 to 5GHz power would be about 150W at the latter frequency and perf would be 1.1 x 1.25 = 1.375.

That bode well with AMD statement of >35% better perf at >25% better perf/watt.

Edit : At 230W it should score 10% better than at said 153W and run at the demoed 5.5GHz, that s assuming a biquadratic relationship for power/frequency from 5 to 5.5GHz.

 

[Asterox]

It is almost certainly an overclock.

Even then, running 5.6 all-core on a SKU that's not the best binned one, is excellent news! So hopefully it's true.

It doesn't have to be, or no one can claim that the "final rumored spec sheets are absolutely correct".

R5 7600X

- 4.7ghz stock

- 5.3ghz singlecore

Zen 4/retail final samples maybe have significantly higher cpu frequency than expected.

 

[Timmah!]

However, the comparison you made is simply wrong. The 7950X isn't going to score 35-36000 points while consuming 150W PPT.

Still, one can dream

 

[DrMrLordX]

Node deficit cant be overcomed,

It can be somewhat mitigated by refinements. There's an open question as to whether Intel can squeeze any more performance or perf/watt out of their 10nm node family.

At the investor meeting the zen4 with 3D cache was set to launch this year, but do you believe this to be changed or can launched also be a paper launch?

If you read the language carefully, it says something like Raphael-X will be "introduced" this year. So paper launch is possible, or it could just be a demo.

 

[Abwx]

It can be somewhat mitigated by refinements. There's an open question as to whether Intel can squeeze any more performance or perf/watt out of their 10nm node family.

The numbers are more or less known, to get the same CB score as a stock 7950X a 13900K will use 65% more power.

That s the most favourable outcome for Intel and that s assuming that AMD s >25% perf/watt improvement occur at 153W for the 7950X (vs 142W for the 5950X), if the improvement is referenced at 170W PPT vs 142W PPT then the gap will be even more wider.

 

[insertcarehere]

I love how some people here can make seemingly ironclad conclusions about the perf/w of the 7950x when there's been next to no leaks about the perf or the power consumption of such product...

 

[PJVol]

Greymon is hinting at 5.6Ghz all-core Turbo for 7600X?

I don't get, why 7600X and not 7900X?
There are 12 clocks in that tweet - hence 12 cores. Fmax for the 7900X rumored to be 5.6Ghz.
If proper cooling applied it very well could hit Fmax at all cores, being unrestricted on PPT/TDC

 

[inf64]

Greymon55 again :

https://twitter.com/x/status/1562751802059399168

76 23 1 19xx
77 23 1 20xx

R23 CB, 1T, 7600X ~1950 pts?
R23 CB, 1T, 7700X ~2050 pts?

If 7950X hits 5.8Ghz it should score ~2150 pts in R23 1T

I don't get, why 7600X and not 7900X?
There are 12 clocks in that tweet - hence 12 cores. Fmax for the 7900X rumored to be 5.6Ghz

He replied that it's a 12T part. 7900X has 24T. It's likely an all-core stable OC for 7600X

 

[PJVol]

He replied that it's a 12T part

Well, then it's OC indeed.
Being properly cooled 5600X can easily pass all CB's MT running all cores 150-200Mhz above stock.
So if 5.6Ghz (+300Mhz) doesn't require exotic cooling, then it could indicate conservative boost clocks due to TDP constraints.

High clocks costs too much power. PPT of 5600X is TDP+20%, if they keep this ratio, 7600X should have this limit around 125W.
So 65% more power.