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!

 

[Markfw]

A single Run of MT Cinebench is really not applicable to people that actually use their computer for work. Hence a 10 minute CB Run the Intel CPUs lose more performance due to thermals and power budget when compared to AMD

Exactly my point. Those who are "Intel supporters" and not realistic will use ANY excuse to put AMD down, like if the 13900k wins in a one minute task that is single threaded. This would be my definition of a troll.

And also, someone who does the same, but is an AMD supporter, but the the other conditions apply would also be a troll. Anyone who skews the facts to fit their own agenda.

 

[DrMrLordX]

It might turn out to be true, but i would not put much trust into this video. Its clearly one of those made, when there is actually nothing new to share, but you gotta still produce some content.

I am always willing to consider the worst possible outcome. Plus even if we ignore MLID, based on AMD's statements, it does look like a Nov/Dec launch for Raphael-X is just wishful thinking. Their language is ambiguous.

That is because it uses actual workloads.

Sorry but if a benchmark can't make the CPU do anything but idle then it's a yawner for me.

Maybe, but Cinebench is not ultimate CPU benchmark.

It's a good sign of fp performance + SIMD up to AVX2 (at least as far as R20 is concerned) with cache/memory subsystem somewhat de-emphasized.

A single Run of MT Cinebench is really not applicable to people that actually use their computer for work. Hence a 10 minute CB Run the Intel CPUs lose more performance due to thermals and power budget when compared to AMD

Alder Lake lose more than 10% MT performance on a 10 minute loop as compared to less than 1% of the AMD top of the line. This will translate linearly to Raptor Lake.

That's assuming they use a normal tau. Not every board does by default. That's been happening since the 9900k.

 

[Exist50]

So the people for whom the short duration performance is more important are trolls?

Remember, caring about anything more than power-limited, embarrassingly parallel HPC workloads is trolling, but claiming that Zen 3 is 4x as efficient as Golden Cove is merely an unbiased contribution.

 

[tamz_msc]

A single Run of MT Cinebench is really not applicable to people that actually use their computer for work. Hence a 10 minute CB Run the Intel CPUs lose more performance due to thermals and power budget when compared to AMD

Alder Lake lose more than 10% MT performance on a 10 minute loop as compared to less than 1% of the AMD top of the line. This will translate linearly to Raptor Lake.

So say a stock 16C/32T 7950X beats/matches(by a 1-2% at stock) a 13900K/S, You know that for work the 7950X will actually perform about 10% better due to thermals/Power budget and that Intel is already pushing those CPUs to the limit.

View attachment 66551

What makes you think that "workstation" use only implies long duration workloads?

 

[nicalandia]

What makes you think that "workstation" use only implies long duration workloads?

People that actually do work with their computer would like to do things as fast as possible because Time=Money. Most of those Rendering jobs take longer than 10 minutes, heck some will take hours




Yes, some task only take a few minutes.

 

[Tuna-Fish]

The other thing I want to see is what memory speed they are using in all these tests, both the 13900k and the 7950x. I would love to see matched memory at 6000-6600 speed testing on those 2.

Note that matched memory is probably always an unfair comparison. AMD can do 1:1 speeds with fabric at ~3000MHz, while Intel can only do 2:1 at practical DDR5 speeds. So let's say you have a very good kit that can do 6600. If you force both to run at 6600, you are hurting AMD because they have to go to 2:1 for that speed, and 10% extra memory speed is in no way worth the latency and internal bw hits. If you force both to run at 6000, you are hurting Intel because they gain nothing from dropping down and lose some bandwidth.

In the end, imho the only sensible comparisons are either with both at maximum officially supported, at peak performance with oc, or when compared at isodollar (and then you should include cpu & mb & ram prices).

 

[tamz_msc]

People that actually do work with their computer would like to do things as fast as possible because Time=Money.

This argument holds only for a subset of people who are looking into buying these kind of processors. Let's not pretend that the whole world is full of 3d-artists running Blender and Cinema4D.

 

[Exist50]

People that actually do work with their computer would like to do things as fast as possible because Time=Money. Most of those Rendering jobs take longer than 10 minutes, heck some will take hours

View attachment 66554


Yes, some task only take a few minutes.

If you're doing V-Ray and you care about speed, you would be buying an Nvidia GPU. Comparing CPU performance in tasks where a CPU would actually be used seems like the only sane approach.

 

[nicalandia]

This argument holds only for a subset of people who are looking into buying these kind of processors. Let's not pretend that the whole world is full of 3d-artists running Blender and Cinema4D.

I agree.

But in any event, these CPUs(7950X and 13900K/S) weather balanced or Balls to the Wall will give many people(Enthusiasts or freelancers) the chance to have a Workstation Class CPU that have the same grunt of Current Workstation CPUs like the 32 Core Ice Lake and ThreadRipper Pro CPUs

 

[biostud]

I don't think AMD will announce much about vcache SKUs until nearly ready for availability.
It'd tank sales. Which gamer would buy a 7800X knowing the 7800X3D is coming in 3 months?

Hence the reason that there is no 7800X.

 

[Markfw]

Remember, caring about anything more than power-limited, embarrassingly parallel HPC workloads is trolling, but claiming that Zen 3 is 4x as efficient as Golden Cove is merely an unbiased contribution.

First, running 32 separate DC tasks is not parallel. Second, ONE time in ONE case, the 12700K was taking twice as much power as the 5950x and doing half the DC work. In that case it was 4x as efficient. But you will keep spouting that until doomsday, just to make yourself look self-important.

Edit: and using that term "embarrassingly parallel" is stupid IMO. If a task can run multiple parallel threads to accomplish a task, it runs them parallel, and the task is efficiently coded.

 

[Exist50]

First, running 32 separate DC tasks is not parallel.

I'm not sure if you're joking, but that is indeed a parallel workload...

Second, ONE time in ONE case, the 12700K was taking twice as much power as the 5950x and doing half the DC work.

And what a fascinating coincidence that not a single person could substantiate your claim, including a supposed ~100W PL2 violation, and twice that for PL1. It's truly a mystery... Nor should we forget the time you accused people of being trolls for reading a posted link when you did not.

 

[Markfw]

I'm not sure if you're joking, but that is indeed a parallel workload...

And what a fascinating coincidence that not a single person could substantiate your claim, including a supposed ~100W PL2 violation, and twice that for PL1. It's truly a mystery... Nor should we forget the time you accused people of being trolls for reading a posted link when you did not.

Yes, running 32 at a time they are parallel, but the cases everyone keeps talking about are ones where there are 32 threads doing one task, like CB23.

As for the other "violations" I was using a kill-a-watt. No idea why the motherboard allowed that. Talk to ASUS.

And what does this have to do with Zen4 or EPYC Genoa ?? Just you trolling, backing up tamz_msc's trolling.

 

[Exist50]

And what does this have to do with Zen4 or EPYC Genoa ?? Just you trolling, backing up tamz_msc's trolling.

You yourself brought up the topic of different workloads, but now responding to that point is trolling? Yeah, I think my point is made.

 

[Carfax83]

In memory sensitive apps Zen 4 should easily eclipse the Golden/Raptor Cove.

Why do you think so? By all accounts, Raptor Lake has significantly improved memory performance via expanded caches, faster L3 cache and a better memory controller.

Not saying you're wrong, but I don't think we have enough information to go off of yet concerning Zen 4's memory performance, especially relative to Raptor Lake's.

 

[Markfw]

You yourself brought up the topic of different workloads, but now responding to that point is trolling? Yeah, I think my point is made.

Your first post on this subject is post 7.653 in response to tamz_msc. Thats where you accuse me (without naming), here:
Remember, caring about anything more than power-limited, embarrassingly parallel HPC workloads is trolling, but claiming that Zen 3 is 4x as efficient as Golden Cove is merely an unbiased contribution.

Thats was the trolling and crap about me. and that stupid term "embarrassingly parallel"

 

[Exist50]

Your first post on this subject is post 7.653 in response to tamz_msc

You brought up the topic at 7648. Now you're trying to claim that anyone who responds about the same topic is trolling. So which is it?

Thats was the trolling and crap about me. and that stupid term "embarrassingly parallel"

If case you weren't aware, that's an industry term.

https://en.wikipedia.org/wiki/Embarrassingly_parallel

 

[Markfw]

You brought up the topic at 7648. Now you're trying to claim that anyone who responds about the same topic is trolling. So which is it?

If case you weren't aware, that's an industry term.

https://en.wikipedia.org/wiki/Embarrassingly_parallel

Yes, I brought up the fact that trolls will try and justify their position talking about single core or lightly threaded tasks, and tamz_msc responded as a troll would, then you responded to him, supporting his position.

As for the term being an industry term, not sure about that, but its is recognized by wiki as a common phrase. BUT, it does talk about a task that can be broken up into smaller tasks (threads), not what I was talking about in my 32 task example.

 

[Abwx]

Why do you think so? By all accounts, Raptor Lake has significantly improved memory performance via expanded caches, faster L3 cache and a better memory controller.

Not saying you're wrong, but I don't think we have enough information to go off of yet concerning Zen 4's memory performance, especially relative to Raptor Lake's.

Node deficit cant be overcomed, for the same CB score the soon to come 13900K will use roughly 50% more power than the 7950X, if anything Intel s decision to allow 350W short duration power is an indication that this is what is necessary to match a 7950X@230W.

 

[Markfw]

Node deficit cant be overcomed, for the same CB score the soon to come 13900K will use roughly 50% more power than the 7950X, if anything Intel s decision to allow 350W short duration power is an indication that this is what is necessary to match a 7950X@230W.

I think the 7950x is 170 watt. The socket is 230. But we will see when official benchmarks come out. Also, the 13900k can only hold that power for short periods, after that performance degrades.

Regardless, the 13900k is short term, and the 7950x is long term for high performance mode.

 

[Exist50]

Yes, I brought up the fact that trolls will try and justify their position talking about single core or lightly threaded tasks, and tamz_msc responded as a troll would, then you responded to him, supporting his position.

So you bringing up the very topic you call trolling, apropos of nothing, is somehow not trolling, but responding to that is? Incredible...

 

[Abwx]

I think the 7950x is 170 watt. The socket is 230. But we will see when official benchmarks come out. Also, the 13900k can only hold that power for short periods, after that performance degrades.

Regardless, the 13900k is short term, and the 7950x is long term for high performance mode.

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.

 

[Asterox]

Looks like it's just a 2021 ES

View attachment 66546

Yes, but do we now what is its cpu frequency(in Cinebench R20 test) as R7 7700X ES CPU?If we use this R5 7600X ES model as example.

- ES vs retail CPU, or the difference in Singlethread and multithread All core CPU frequency

https://twitter.com/x/status/1561971310443794432

 

[Carfax83]

Node deficit cant be overcomed, for the same CB score the soon to come 13900K will use roughly 50% more power than the 7950X, if anything Intel s decision to allow 350W short duration power is an indication that this is what is necessary to match a 7950X@230W.

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.

 

[nicalandia]

This has nothing to do with what I was asking. @inf64 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.

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. ...