Info [Toms, Anand] AMD EPYC Benchmarks

[Justinbaileyman]

Its no big deal I am just a cheap skate LOL. No really though I was thinking that getting 4 sticks now and 4 sticks later would net me faster 3200 speed sticks since there is a shortage right now.
Maybe I will just get the 2933 sticks...
This is all very frustrating to me!!

 

[maddie]

Its no big deal I am just a cheap skate LOL. No really though I was thinking that getting 4 sticks now and 4 sticks later would net me faster 3200 speed sticks since there is a shortage right now.
Maybe I will just get the 2933 sticks...
This is all very frustrating to me!!

Why did you get a Rome system? What's your use case? No biggie, just interested.

 

[Justinbaileyman]

Why did you get a Rome system? What's your use case? No biggie, just interested.

Media encoding and conversion, compression , file transfer , video editing , audio editing ect..
This 7302p will run circles around threadripper and probably the upcoming 3950x which I am also buying in the coming days or weeks but that will be used for other things such as gaming and media viewing..

 

[Topweasel]

Media encoding and conversion, compression , file transfer , video editing , audio editing ect..
This 7302p will run circles around threadripper and probably the upcoming 3950x which I am also buying in the coming days or weeks but that will be used for other things such as gaming and media viewing..

It should't "run circles" around a 16c TR3. They are the same CPU, with TR3 dies picked for max clock speeds. They might be really close if both have similar die configs (4 Chiplets vs. 8 chiplets). If the EPYC is using more dies then the clocks will be measurably lower.

The only advantages the EPYC would have on TR, is memory channels and PCIe lanes. If you aren't using all of either of them then it should be a wash with the TR holding a clock advantage.

 

[Justinbaileyman]

It should't "run circles" around a 16c TR3. They are the same CPU, with TR3 dies picked for max clock speeds. They might be really close if both have similar die configs (4 Chiplets vs. 8 chiplets). If the EPYC is using more dies then the clocks will be measurably lower.

The only advantages the EPYC would have on TR, is memory channels and PCIe lanes. If you aren't using all of either of them then it should be a wash with the TR holding a clock advantage.

Not true the Epyc rome die is the second generation which will be in the upcoming threadripper 3000 series which hasnt even come out yet.. Beside I was talking about vs. first and second gen threadripper.
so yes my Epyc will pretty much completely destroy any first or second gen threadripper core for core that is.
I a not worried about clock speed as 3.3Ghz is plenty fast for the kind of work I do and if not I can always use PBO which actually works on the server side of things I hope vs. the high end/ main stream platform.
With the huge ipc gain plus 128mb of lvl 3 cache vs 32mb on threadripper and pcie-4.0 vs 3.0 for threadripper its a no brainer that Epyc will be faster.
But this really isnt a, "this is better then that thread" so sorry if we are getting off topic guys.

Edit: sorry again guys , to answer your other question which I missed at first was, "why buy the Epyc Rome system in the first place"?? The answer is easy.. Total upgrade path to 128c/256t cpu if I ever want or need such a super monster cpu.
in a year or two maybe even three prices will drop and I will swoop in like a vulture to pick off the little tidbits left by huge server farm upgrades which will allow me to snag up cheap huge core server cpu's that are being sold off on say ebay or what ever..

 

[Topweasel]

Not true the Epyc rome die is the second generation which will be in the upcoming threadripper 3000 series which hasnt even come out yet.. Beside I was talking about vs. first and second gen threadripper.
so yes my Epyc will pretty much completely destroy any first or second gen threadripper core for core that is.
I a not worried about clock speed as 3.3Ghz is plenty fast for the kind of work I do and if not I can always use PBO which actually works on the server side of things I hope vs. the high end/ main stream platform.
With the huge ipc gain plus 128mb of lvl 3 cache vs 32mb on threadripper and pcie-4.0 vs 3.0 for threadripper its a no brainer that Epyc will be faster.
But this really isnt a, "this is better then that thread" so sorry if we are getting off topic guys.

Sorry I miss read the 3950 as being the TR3 1950/3950 Replacement. You are absolutely right.

 

[Justinbaileyman]

Oh no probelm.. yeah Epyc will be faster then 3950x/x570 platforms and 2950x/x399 platform but not the upcoming threadripper 3000 series. I am expecting that to be awesome since they are pretty much the same die's except clock speeds will be much much higher for threadripper 3000.
I would totally buy a Threadripper 3000 right now if it were out but they are still in the rumor stages and nothing is set in stone yet as far as I know.

 

[nicalandia]

yeah Epyc will be faster then 3950x/x570 platform

Your 16C/32T Epyc 7302p will not be faster than a 16C/32T Ryzen 3950X.

 

[Justinbaileyman]

Your 16C/32T Epyc 7302p will not be faster than a 16C/32T Ryzen 3950X.

Prove it!! Epyc 7302p is a generation if not 2 generation ahead of 3950x!!

 

[nicalandia]

Prove it!! Epyc 7302p is a generation if not 2 generation ahead of 3950x!!

Really?

Amd Epyc 7302P: 16C/32T(Rome, Zen 2 Core uArch)

CCD Compute Units: 8
Cores per CCD: 2
Total L3 Cache: 128

Base Clock: 3.00 Ghz
Max Boost Clock: 3.2 Ghz

TDP: 155W



Amd Ryzen 9 3950X: 16C/32T (Matisse, Zen 2 Core uArch)

CCD Compute Units: 2
Cores per CCD: 8
Total L3 Cache: 64

Base Clock: 3.5 Ghz
Boost Clock: 4.7 Ghz

TDP: 105W


From that point you can see that 3950x will have the best binned Compute Units, the Epyc will have 6 cores disabled per Compute unit, and while Epyc will generally fare better at Server/Data Center type of workloads due to 8 channel memory, many HEDT and Mainstream Apps will do just fine with Ryzen fast dual channel memory.

 

[Justinbaileyman]

Really?

Amd Epyc 7302P: 16C/32T(Rome, Zen 2 Core uArch)

CCD Compute Units: 8
Cores per CCD: 2
Total L3 Cache: 128

Base Clock: 3.00 Ghz
Max Boost Clock: 3.2 Ghz

TDP: 155W



Amd Ryzen 9 3950X: 16C/32T (Matisse, Zen 2 Core uArch)

CCD Compute Units: 2
Cores per CCD: 8
Total L3 Cache: 64

Base Clock: 3.5 Ghz
Boost Clock: 4.7 Ghz

TDP: 105W


From that point you can see that 3950x will have the best binned Compute Units, the Epyc will have 6 cores disabled per Compute unit, and while Epyc will generally fare better at Server/Data Center type of workloads due to 8 channel memory, many HEDT and Mainstream Apps will do just fine with Ryzen fast dual channel memory.

The max boostclock for 7302p is 3.3Gghz not 3.2ghz and ryzen 3950x only has 24 pcie lanes vs 128 on Epyc.
The 3950x only has 64mb L3 cache vs 128mb of L3 for Epyc
The 3950x only has 2 channels DDR4 support for a max of 2933 with 4 sticks with anything greater then 8GB per stick or up to 4200 with only 2 stick supposedly while Epyc has 8 channels with a Max speed allotment of DDR4 3200 with a max cap of 16 stick or more depending on the motherboard at full speeds with higher memory speed to be introduced later in the life cycle so says the rumors.
SO whats your point? The Epyc is far far far superior to the 3950x otherwise Epyc would not exist in the first place!!
Also I will have both 3950x and 7302p and plan to pit them against each other to see who the real king of kings will be in the 16c/32t market.
Beside it seems to me you kind of have a personal agenda against the Epyc CPU's, you mind me asking why this is??

 

[nicalandia]

Also I will have both 3950x and 7302p and plan to pit them against each other to see who the real king of kings will be in the 16c/32t market.

I wish to thank you in advance for doing such a test(Ryzen vs Epyc) Core vs Core, many people would love to have Epycs in Workstation duties and we don't see any type of HEDT/WorkStation test made on thme often.

Due to CCD(CCX/CCX) arrangements and High OC(stock) I believe the 3950X will reign supreme, but I will be very happy if proven wrong

 

[Justinbaileyman]

I wish to thank you in advance for doing such a test(Ryzen vs Epyc) Core vs Core, many people would love to have Epycs in Workstation duties and we don't see any type of HEDT/WorkStation test made on thme often.

Due to CCD(CCX/CCX) arrangements and High OC(stock) I believe the 3950X will reign supreme, but I will be very happy if proven wrong

Well hopefully we will get to see what is what if AMD would ever release the darn 3950x!!
Here we are in Sept. already and no word yet of when it is to be released..

 

[nicalandia]

Well hopefully we will get to see what is what if AMD would ever release the darn 3950x!!
Here we are in Sept. already and no word yet of when it is to be released..

I agree, it's going to be after the 10th and hopefully before the 30th.

 

[lobz]

Nobody really pays list prices for xeons. Also in the TCO of the whole server incl. software the cpu plays a minor role. Of course yeah you fare better with epyc this time but in general CPU price isn't that important in the big picture.

I always hear this exact bullcrap. If this was true, the CPU prices could also be 10 times as high, big picture guys. However, they are not.

 

[DrMrLordX]

I always hear this exact bullcrap. If this was true, the CPU prices could also be 10 times as high, big picture guys. However, they are not.

Texas Boosts U.S. Science with Fastest Academic Supercomputer in the World

Frontera, at the Texas Advanced Computing Center, will power discoveries by the nation’s top computational scientists

news.utexas.edu

Frontera has something like $160 million worth of Cascade Lake CPUs in it. Apparently they didn't pay that much for the entire system. Intel did some deep discounting to get those chips in there.

 

[StefanR5R]

Amd Epyc 7302P: 16C/32T(Rome, Zen 2 Core uArch)

CCD Compute Units: 8
Cores per CCD: 2
Total L3 Cache: 128

It is possible that the 7302P is configured with 4 CCDs, not 8.

 

[nicalandia]

It is possible that the 7302P is configured with 4 CCDs, not 8.

Yes, that makes sense

2+2 CCD(2 core per CCX) will allow for the full L3 Cache of 32 MB per CCD.

 

[nicalandia]

Red Hat sets new performance records on AMD ROME processors

Red Hat sets new performance records on AMD ROME processors

As x86_64 hardware sees improvements in the area of CPU die size reduction, Red Hat has continued to work with our partners to better enable world class performance. Recently, AMD released its ROME line of processors, known as the AMD Epyc 7002 Series. With a die size of 7nm, AMD has achieved...

www.redhat.com

You must add your own commentary to third party content when posting in the CPU board. No dropping links, quoting articles, or posting images or videos without your own commentary.

AT Moderator ElFenix

 

[Justinbaileyman]

Hope that performance mirrors or leaks over for users of Ubuntu like me..
Well actually I use a custom modified version of Ubuntu Mate but you know what I am getting at!!

 

[Asterox]

 

[Panino Manino]

What Linus Sebastian does with a 64 core Epyc?
Tests Cinebench and Blender then proceeds to play games...

 

[StinkyPinky]

He did say it was not the full review, just a bit of a play around.

 

[tomatosummit]

He's playing it pretty well. His audience isn't going to be interested in how fast it can crunch through a simulation, they want video games and overclocking.
Him doing a software rendered game is pretty funny and gets people talking about how powerful the cpu actually is, an interesting video for his fans and a positive look for epyc.

 

[mopardude87]

I remember how excited i was to have a 2.4Ghz Q6600 back in 2008. Mind blowing to see 32x the threads at higher clocks. The cinebench results were so insane it has made me lose the will to even bother running cinebench ever again on my 8700 non k. Not that i use cinebench much outside of testing thermals.

The max memory support is more then everyone know i know and their hard drive storage capacity combined which includes mine! When he stuffed it with 256gb all i thought was ok he could do better then that! I just assume had nothing even close he could have used to even pushed to 200gb.