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!

 

[nicalandia]

Here you go @nicalandia, i know you like GB5 runs

7950x PBO CO tuned

ST= 2407
MT = 26241
View attachment 68762

I also have higher MT scores, but lower ST then..
View attachment 68763

Here it is on It's Glory..

The fastest 7950X to date..!

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 7950X processor.

browser.geekbench.com

And against the fastest 13900K posted so far.

ASUS System Product Name vs Z690 Taichi Razer Edition - Geekbench

 

[igor_kavinski]

Who here is brave enough to try it?

What say you, Det0x? Do copper nanoparticles sound interesting as a coolant?

 

[repoman0]

What ever happened to the phase change refrigeration setups that nerds were running back in the Pentium 4 and Socket 939 days? I remember wanting one of those in high school for those few hundred extra MHz, luckily I settled for building a custom water loop rather than a compressor running 24x7. That would have really kept the friends and girls at bay even more than my water cooling did.

 

[DrMrLordX]

What ever happened to the phase change refrigeration setups that nerds were running back in the Pentium 4 and Socket 939 days?

They exist. They're expensive and require loud compressors or multi-stage compressor setups to function, and the peak heat flux from the heat source (read: CPU) can't be that high either. The more power these CPUs shed, the less-attractive phase gets.

 

[Markfw]

Zen 4 performance update. So I am running primegrid right now. It running on 16 threads of the 32 for the 7950x. It using avx-512 according to the logs. Its running the "units" at 2:05 to 2:40 depending on the nature of the unit. My 5950x's (and others on the project) are running ~3.5 hours to 4 for the same units, no avx-512, but using other code (like FMA3). Both are running about 50% load, which is correct for 16 of 32 threads loaded.

Here is the killer....
5950x is using 90% of 142 watts @65c and at 3100 to 3300 speed. (not sure why the low speed, 100% stock)

7950x is using 38% of 230 watts, or 60 watts CPU and 13 watts SOC (according to Ryzen master) and 3300 mhz. My kill-a-watt verifies this power usage(minus the 3090TI@100% stock 450 watts)

But its running 95c !!! So, its running 50% faster throughput (on this specific load with avx-512) at ~1/2 the power usage, but at 95c.

This is all stock setting for both. I realize this is just one specific set of software running, but my god, what a difference, and on the good side.

Oh, cooling, 240mm AIO for the 5950x and 420 AIO for the 7970x.

If this is not interesting, I don't know what is !!!!

 

[RnR_au]

Intel and partners are still selling their peltier solutions;

Intel Cryo Cooling tech debuts in EKWB & Cooler Master products

Intel's sub-ambient tech uses a thermoelectric cooler plate, custom firmware and software.

hexus.net

Haven't come across reports of the system being ported to AMD sockets/cpus.

 

[Hitman928]

Zen 4 performance update. So I am running primegrid right now. It running on 16 threads of the 32 for the 7950x. It using avx-512 according to the logs. Its running the "units" at 2:05 to 2:40 depending on the nature of the unit. My 5950x's (and others on the project) are running ~3.5 hours to 4 for the same units, no avx-512, but using other code (like FMA3). Both are running about 50% load, which is correct for 16 of 32 threads loaded.

Here is the killer....
5950x is using 90% of 142 watts @65c and at 3100 to 3300 speed. (not sure why the low speed, 100% stock)

7950x is using 38% of 230 watts, or 60 watts CPU and 13 watts SOC (according to Ryzen master) and 3300 mhz. My kill-a-watt verifies this power usage(minus the 3090TI@100% stock 450 watts)

But its running 95c !!! So, its running 50% faster throughput (on this specific load with avx-512) at ~1/2 the power usage, but at 95c.

This is all stock setting for both. I realize this is just one specific set of software running, but my god, what a difference, and on the good side.

Oh, cooling, 240mm AIO for the 5950x and 420 AIO for the 7970x.

If this is not interesting, I don't know what is !!!!

Sounds like either there's a contact issue between the waterblock and the IHS for the 7950x, or this particular code creates a massive hotspot in the execution units of the cores. I'm inclined to believe it's a bit of both.

 

[Markfw]

Sounds like either there's a contact issue between the waterblock and the IHS for the 7950x, or this particular code creates a massive hotspot in the execution units of the cores. I'm inclined to believe it's a bit of both.

well, it seems to live at 95c, load or no load. I was very careful putting the compound on and torquing the screws, its an Artic Cooling 420 AIO, the nest rated AIO.

Remember, AVX-512 can run very hotspotty.

 

[PJVol]

This is all stock setting for both

What's the default EDC limit for 7950x and max EDC reported in both tests?

 

[Markfw]

What's the default EDC limit for 7950x and max EDC reported in both tests?

I should probably know this, but whats EDC ?

Wait, I found it. Its in Ryzen master. 39% for the 7950x and 96% for the 5950x.

Edit2: default ? Is that max ?

140amp for 5950x and 225 amp for 7950x.

Edit 3: 42% for the 7950x and 95% for the 5950x USED,.

 

[LightningZ71]

While those are solid numbers, I can't avoid thinking that some hand tuning might further improve your results. I suspect that a tiny bit of negative voltage offset might make a notable impact on both processors.

If you are holding at 95c, which is the target for Zen4, with wattage numbers that low, and you are confident in your cooling installation, a slight negative voltage offset could allow higher clocks at the temperature preset max and get you under 2 hours average.

 

[Kocicak]

7950x is using .... 60 watts CPU and 13 watts SOC (according to Ryzen master) ...
But its running 95c !!!
cooling 420 AIO

Something is wrong here. Even the smallest AMD stock air cooler should be able to cool the CPU consuming 73W without any problem. Is not your AIO broken? Did not you forget the protection foil on the block or something like that?

 

[RnR_au]

Something is wrong here. Even the smallest AMD stock air cooler should be able to cool the CPU consuming 73W without any problem. Is not your AIO broken? Did not you forget the protection foil on the block or something like that?

Definitely something is wrong.

 

[Doug S]

Probably some sort of bug in reporting temperature.

 

[Det0x]

I should probably know this, but whats EDC ?

Wait, I found it. Its in Ryzen master. 39% for the 7950x and 96% for the 5950x.

Edit2: default ? Is that max ?

140amp for 5950x and 225 amp for 7950x.

Edit 3: 42% for the 7950x and 95% for the 5950x USED,.

On asus motherbard ? Set LLC 5 ("recommended for oc") instead of "auto" and see if that helps

 

[Markfw]

FIRST I want to thanks everyone here for saying "something is wrong" After a zillion builds, I did the worst thing ever. I forgot to take the tape off the head for the AIO.

So... NOW a lot changed. Still 95c. BUT, NOW I am running 5.1 ghz, and vcore went up from .9 to 1.2. And watts went from 55 to 150.

Now what remains to be seen is what happens to performance. It was already kicking butt at 3100. So now at 5100, how good is it.

Its 11 pm at night, but I could not sleep thanks to your advice, so I fixed it. I am going to bed, and will report in the morning.

Getting too old (will be 69 soon)

Edit: The one thing I will say, is that I learned something. At 55 watts, its kicks butt compared to every CPU out there, including Alderlake. (my 12700F) At 150 watt ? Well, I will let you know in the morning, but lets just say I know its the efficiency king at least.

 

[poke01]

FIRST I want to thanks everyone here for saying "something is wrong" After a zillion builds, I did the worst thing ever. I forgot to take the tape off the head for the AIO.

So... NOW a lot changed. Still 95c. BUT, NOW I am running 5.1 ghz, and vcore went up from .9 to 1.2. And watts went from 55 to 150.

Now what remains to be seen is what happens to performance. It was already kicking butt at 3100. So now at 5100, how good is it.

Its 11 pm at night, but I could not sleep thanks to your advice, so I fixed it. I am going to bed, and will report in the morning.

Getting too old (will be 69 soon)

Edit: The one thing I will say, is that I learned something. At 55 watts, its kicks butt compared to every CPU out there, including Alderlake. (my 12700F) At 150 watt ? Well, I will let you know in the morning, but lets just say I know its the efficiency king at least.

hey we all make mistakes. I'm glad you fixed it.

 

[Markfw]

Update. I am getting at a minimum, 20% more performance out of the chip. It was beating every chip I had, and now its just killing them. 1 hour 50 minutes, instead of the quickest by this chip of 2:20, and 3+ hours for a 5950x.

 

[inf64]

Glad you sorted it out Mark. Now you can maybe play a bit with power limits and max. temp tweaking (ie. set a lower than 95C limit in BIOS). This should bring the power down with a smallish performance impact.

 

[Kocicak]

After a zillion builds, I did the worst thing ever. I forgot to take the tape off the head for the AIO.

Well, at least you got the silicone nicely burnt-in.

 

[Sable]

FIRST I want to thanks everyone here for saying "something is wrong" After a zillion builds, I did the worst thing ever. I forgot to take the tape off the head for the AIO.

So... NOW a lot changed. Still 95c. BUT, NOW I am running 5.1 ghz, and vcore went up from .9 to 1.2. And watts went from 55 to 150.

Now what remains to be seen is what happens to performance. It was already kicking butt at 3100. So now at 5100, how good is it.

Its 11 pm at night, but I could not sleep thanks to your advice, so I fixed it. I am going to bed, and will report in the morning.

Getting too old (will be 69 soon)

Edit: The one thing I will say, is that I learned something. At 55 watts, its kicks butt compared to every CPU out there, including Alderlake. (my 12700F) At 150 watt ? Well, I will let you know in the morning, but lets just say I know its the efficiency king at least.

Der8auer delidded one and it dropped 20 degrees. AMD used a thicker heatspreader to maintain compatibility with AM4 coolers.

 

[moinmoin]

And watts went from 55 to 150.

I am getting at a minimum, 20% more performance out of the chip.

If you are conscious of power consumption as a first easy step you could consider using ECO mode, the 105W TDP variant of that should be a good fit for your use case without diminishing the improvement you are seeing.

 

[Markfw]

If you are conscious of power consumption as a first easy step you could consider using ECO mode, the 105W TDP variant of that should be a good fit for your use case without diminishing the improvement you are seeing.

I have not figure out where that is in bios yet. Thats what I want. ASRock bios.

 

[nicalandia]

I have not figure out where that is in bios yet. Thats what I want. ASRock bios.

Also check what is the performance with AVX-512 Disabled, also if you could just disable the AVX-512 All together and run Cinebench R23 and see if there is any type of changes be it power consumption or performance or both.

 

[Daneden]

FIRST I want to thanks everyone here for saying "something is wrong" After a zillion builds, I did the worst thing ever. I forgot to take the tape off the head for the AIO.

So... NOW a lot changed. Still 95c. BUT, NOW I am running 5.1 ghz, and vcore went up from .9 to 1.2. And watts went from 55 to 150.

Now what remains to be seen is what happens to performance. It was already kicking butt at 3100. So now at 5100, how good is it.

Its 11 pm at night, but I could not sleep thanks to your advice, so I fixed it. I am going to bed, and will report in the morning.

Getting too old (will be 69 soon)

Edit: The one thing I will say, is that I learned something. At 55 watts, its kicks butt compared to every CPU out there, including Alderlake. (my 12700F) At 150 watt ? Well, I will let you know in the morning, but lets just say I know its the efficiency king at least.

This CPU is so good you can run it with an insulator between the die and the cooler and you still kick ass with it 😂