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

Vattila · Oct 6, 2019

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!

Thala · May 25, 2022

ondma said:
I see what you are saying, but essentially one metric is looking at rate of performance, while the other is looking at time to complete the task. So you can say it is 46% faster at rate of performance, but the time to complete the task is only 31% less. If it makes you feel better, I will re-phrase as so;
The rate of performance is 46% faster, but the time saved, compared to the original time, is only 31%. (Mathematically, it comes down to whether the calculation uses the smaller or larger number for the denominator.)

@ Inteluser: BTW, your first two examples are absurd. No cpu can complete a task in 0 seconds (the rate of performance would be infinite), and Car B is not faster than car A.

The math is extremely simple. We are looking at 2 metrics, speed metric and time metric. The speed metric is work/time. So you see both metrics are inversely proportional. Now when putting 2 measurements into relation you can relate them in terms of time or in terms of speed - like time0/time1 or speed0/speed1. However since we already concluded that both metrics are inversely proportional, their relations must be inverse too (speed0/speed1=time1/time0). Point in case 31% less time == 0.69 time == 1/0.69 speed == 1.45 speed == 45% faster.

PJVol · May 25, 2022

majord said:
If blender benchmark was measured in pixels per second or something , there wouldn't be so many confused people.

Yep, that's why I think it measures performance in aggregated samples per minute, starting from 3.0.1.

ondma · May 25, 2022

Thala said:
The math is extremely simple. We are looking at 2 metrics, speed metric and time metric. The speed metric is work/time. So you see both metrics are inversely proportional. Now when putting 2 measurements into relation you can relate them in terms of time or in terms of speed - like time0/time1 or speed0/speed1. However since we already concluded that both metrics are inversely proportional, their relations must be inverse too (speed0/speed1=time1/time0). Point in case 31% less time == 0.69 time == 1/0.69 speed == 1.45 speed == 45% faster.

Well put. I still would argue that the time metric is more important, because, why does one buy a new cpu? To do a task in less time. That is just my opinion though, either calculation is mathematically correct.

itsmydamnation · May 25, 2022

Man this thread is insane,

The best way to answer is it 31% or 45% is simple.

How does Intel/AMD/whoever else show performance uplift compared to $X. is the new product the reference point or is $X.

its almost always $X right?

So 45% it is.

done , dusted , speak no more!

stupid but consistent is way better then smart but inconsistent, ask me how i know!

My preferred programming language is PERL

Thala · May 25, 2022

ondma said:
Well put. I still would argue that the time metric is more important, because, why does one buy a new cpu? To do a task in less time. That is just my opinion though, either calculation is mathematically correct.

The issue is really not what metric is more important. Thing is a duration or time-span is a distance metric - and is never used in conjunction with "faster" - if you use the term "faster" you imply that you refer to a speed metric - and you have to convert the numbers accordingly.

nicalandia · May 25, 2022

itsmydamnation said:
Man this thread is insane,

The best way to answer is it 31% or 45% is simple.

How does Intel/AMD/whoever else show performance uplift compared to $X. is the new product the reference point or is $X.

its almost always $X right?

So 45% it is.

done , dusted , speak no more!

stupid but consistent is way better then smart but inconsistent, ask me how i know!

My preferred programming language is PERL

Can you please Quite with that nonsense?

SteinFG · May 25, 2022

Please, let's just move on from percentages on blender. "Ryzen 7000 completed the task 31% faster" is the quote from a computex keynote. Yes, they could've said "Ryzen 7000 completed the task in 31% less time" or "Ryzen 7000 is 1.45X faster in this task", or bla bla, whatever.

They just had two numbers and wanted to say "look, ours is smaller, which is good". Some people discuss this math problem more than those who came up with that slide.

itsmydamnation · May 25, 2022

nicalandia said:
Can you please Quite with that nonsense?

echo reply.

nicalandia · May 25, 2022

CAN WE PLEASE GET A MOD HERE TO CLEAN THIS THREAD OF THIS MATHEMATICAL NONSENSE? Would be nice to start handing out Warnings to Offenders

itsmydamnation · May 25, 2022

nicalandia said:
CAN WE PLEASE GET A MOD HERE TO CLEAN THIS THREAD OF THIS MATHEMATICAL NONSENSE? Would be nice to start handing out Warnings to Offenders

Show me on the dolly where "relative reference points" touched you?

Hans de Vries · May 25, 2022

Thala said:
The issue is really not what metric is more important. Thing is a duration or time-span is a distance metric - and is never used in conjunction with "faster" - if you use the term "faster" you imply that you refer to a speed metric - and you have to convert the numbers accordingly.

Yes this correct of course:

Faster implies "More per unit of time" so: More kilometers per hour or More instructions per second

Therefor the correct answer is 46% faster.

Vope45 · May 25, 2022

Hans de Vries said:
Yes this correct of course:

Faster implies "More per unit of time" so: More kilometers per hour or More instructions per second

Therefor the correct answer is 46% faster.

This is the correct answer

jamescox · May 25, 2022

Markfw said:
All of this is off-topic. I replied to one user who thinks Alder Lake rules number one in everything, and Zen 4 is coming late. Now you are arguing about Alder lake vs Zen3. DROP IT !

We do seem to suddenly have a lot of people in the Zen 4 speculation thread saying how great intel processors are and how AMD is a disappointment or they should be super worried about what Intel is cooking up. I have worked at tech companies where the marketing guys would get on forums to spread FUD about their competitors. This usually fades quickly after whatever announcement triggered it, so I probably shouldn’t have bothered reading this thread for a while.

For me, the only disappointment is how little information we have. It is quite amazing how quiet AMD has managed to keep things. The >15% on preproduction silicon with no clock speeds specified seems to be a bit of a red herring. Since it is on an application that is likely almost completely insensitive to most of Zen 4’s improvements, this doesn’t really give us much of any real information. How fast was a single thread running on this test if multithreaded apps are hitting 5.2 to 5.5? Was it actually clocking up to the expected boost or a lower clocked preproduction chip? The clock speed improvements demonstrated will give it a large boost over Zen 3, even if it isn’t any more efficient. A lot of applications like the larger L2 cache, so I suspect AMD is cleaning up the last remaining areas where they trail Intel. I don’t know if massive changes to the pipeline is reasonable until Zen 5. They could have some higher latencies on the larger cache, but I think that is unlikely due to the move from 7 nm to 5.

As for actual Zen 4 speculation, the package photo is interesting. I suspect that there is some advanced packaging technology in use here (RDL or something), but the cpu die being so close together (looking like a single unit) doesn’t make that much sense. With the way Zen CCXs work, all communications go to the IO die for cache coherency traffic. It is likely too far from the IO die for a silicon bridge between the IO die and the cpu die. You don’t want long interconnect routing or daisy chaining. They aren’t oriented the way they would be on a Genoa die and it would only make sense to connect them together if they can act as if they are a single CCX. The idea makes me wonder if Bergamo might be implemented like that, using 16 separating die in groups of 2. This would be very similar to the rumors on the gpu side. It might also just be space saving with no underlying architecture change.

The chunky lid is still a bit of a mystery also. I have wondered if some high end models might have an integrated vapor chamber or a low power integrated peltier cooler. It definitely seems to be much thicker than necessary.

Timmah! · May 25, 2022

Anyway, i want to know more about that SP6 Socket. Only know found out about that.

jamescox · May 25, 2022

uzzi38 said:
Nah, the future is one of close competition, thankfully. The more competition, the better, and thankfully I don't think we're going to be let down in that regard.

Well, I'm talking about the desktop market when I say that, I don't want to take any bets on server.

There is almost no way for the market to remain stable. Things need to be decided on far in advance, but things change quickly enough to sometimes ruin plans. I would rather not go back to an Intel monopoly anytime soon, so I think AMD still needs to gain some more market share and mind share. If not for AMD, Intel would likely still be in trouble. Intel wouldn’t have accelerated their roadmaps as much so they would look even worse compared to ARM competitors. I read a phoronix review a while back with AMD, Intel, and some ARM server processors. Intel was in third place in some of them. It will be interesting to see how that changes with Genoa and then with Bergamo vs. ARM server processors.

jamescox · May 25, 2022

Thibsie said:
SP6 - Epyc

Do we have anything more than a rumor though? If it exist, it probably isn’t going to be out for quite a while or we probably would have seen something more.

nicalandia · May 25, 2022

jamescox said:
We do seem to suddenly have a lot of people in the Zen 4 speculation thread saying how great intel processors are and how AMD is a disappointment or they should be super worried about what Intel is cooking up.

I am actually pretty impressed with the Multi Threaded Performance Boost, 45% Faster than the 12900K on Blender would mean that the 13900K with it's 24 Cores have a Daunting task ahead

nicalandia · May 25, 2022

Now that I think about it. This is not the first time AMD does this.

Also in this case the ThreadRipper is actually 42% faster than the $20,000 2S Platinum set up. This time it just happens to be on a Ryzen 7000 CPU, so it's more popular than ThreadRipper

Markfw · May 25, 2022

Timmah! said:
Anyway, i want to know more about that SP6 Socket. Only know found out about that.

Edit: I thought you were talking the AM5 socket. SP^ is EPYC. It will work just like SP3, but few more pins.

From what I read (here) it is compatable with AM4 coolers, and the picture looks like its a straight bolt-on.

Saylick · May 25, 2022

nicalandia said:
Now that I think about it. This is not the first time AMD does this.

View attachment 62114

Also in this case the ThreadRipper is actually 42% faster than the $20,000 2S Platinum set up. This time it just happens to be on a Ryzen 7000 CPU, so it's more popular than ThreadRipper

LOL. Shows you how much attention (or rather anticipation) people are investing into AMD. No one gave a crap about the correct percentages when that slide dropped, yet everyone now is spending more time talking about math than actual PC hardware.

Hans de Vries · May 25, 2022

Timmah! said:
Anyway, i want to know more about that SP6 Socket. Only know found out about that.

It was posted here on this thread: Same size as the current SP3 (Napels/Rome/Milan) package. Looks like it also has 8 DIMMS. Maybe on request of motherboard/system builders to make it easier to fit it into existing systems.

jamescox · May 25, 2022

nicalandia said:
I am actually pretty impressed with the Multi Threaded Performance Boost, 45% Faster than the 12900K on Blender would mean that the 13900K with it's 24 Cores have a Daunting task ahead

If it can maintain ridiculously high all-core clocks then it will completely dominate in multi-threaded applications. A Threadripper part would be spectacular, if we actually get one.

jamescox · May 25, 2022

Hans de Vries said:
It was posted here on this thread: Same size as the current SP3 (Napels/Rome/Milan) package. Looks like it also has 8 DIMMS. Maybe on request of motherboard/system builders to make it easier to fit it into existing systems.
View attachment 62115
View attachment 62117

Thanks. I think I did see this a while ago. 8 channels is a little odd considering Genoa at 12, but it would be easy to do by just connecting 2 channels per quadrant. There is no guarantee we get this as a regular Threadripper part though.

lobz · May 25, 2022

Hans Gruber said:
Every reputable publication has said Alder Lake is ahead of Zen 3. The only CPU in Zen 3 that can hold it's own against Alder Lake is the 5950x. Those are not my words but from publications. I can look them up but I am going by memory. Zen 4 is a year behind the original release schedule.

Intel CEO's said that AMD was in it's windshield and they survived and planned to put AMD in their rearview mirror. I think Zen 4 will be fine or good enough. The problem is the 7nm Intel process. Their silicon/cores are more dense than TSMC. Intel has already announced they are rapidly speeding up their timeline for releasing products.

The roadmap is not looking good for AMD because Intel has already caught up. Zen 4 will put them back in front for the short term. Intel 7nm means the core count advantage AMD has had will be gone. A major IPC (Raptor Lake) increase from Intel cuts into the core count lead on Zen 4.

My concern is that Meteor Lake (7nm) will arrive well ahead of Zen 5. AMD had a good run. It's not over yet but the future is not looking very bright for AMD. I was hoping that AMD would have a better showing in the GPU arena. So many miners here, I am not one of them. I don't believe in wasting energy chasing pipe dreams.

what a whole bunch of nonsense

lobz · May 25, 2022

Timorous said:
How much faster is an 8.83ms frame time vs a 16.67ms frame time?

Duh?

It's obviously half as faster!

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

Senior member

Golden Member

Senior member

Attachments

Platinum Member

Platinum Member

Golden Member

Diamond Member

Senior member

Platinum Member

Diamond Member

Platinum Member

Senior member

Member

Senior member

Golden Member

Senior member

Senior member

Diamond Member

Diamond Member

Moderator Emeritus, Elite Member

Diamond Member

Senior member

Senior member

Senior member

Platinum Member

Platinum Member