56 Cores in ONE SYSTEM! - HOLY $H!T

[nicklmg]

Buy Intel Xeon CPUs on Amazon: http://geni.us/sLXzn

Buy Intel Xeon Platinum 8180 on Amazon: http://geni.us/9AT5OE4

 

 

[vanished]

28 CORECPUS!?

 

Shouldn't there be a space?

[nicklmg]

2 minutes ago, Ryan_Vickers said:

28 CORECPUS!?

 

Shouldn't there be a space?

Good catch  

[LienusLateTips]

https://www.amd.com/en/products/epyc-7000-series-2-socket-models

Quote

AMD EPYC™ 7601

 

# of CPU Cores:

32

 

Base Clock:

2.2GHz

 

Max Boost Clock:

3.2GHz

 

Default TDP / TDP:

180W

 

[vanished]

1 minute ago, JDE said:

https://www.amd.com/en/products/epyc-7000-series-2-socket-models

 

AMD can't afford to send them free CPUs "just cuz"

[Streetguru]

So when are yall getting an EPYC server to play with? Because it sounds like a new series, Clash of the TItans, each companies top end product Vs the other.

 

Also are you guys going to get Radeon SSGs in for 8k video editing?

[GodAtum]

that motherboard has no 10GBpe port

[ignaloidas]

I challenge you to find a prime number with this with good old Prime95. You get cookies if you find one!

[Daniel644]

putting every video editor on a single motherboard sounds like a BAD IDEA, imagine how fucked you'd be if that thing crashed.

[David89]

Meh. Another one of those Intel Videos. 56 Cores are great and such, but they are getting actually boring. I'd really like to see EPYC in one of those Videos.

 

Although i get, that there is no Motherboard on AMDs side, that can compete with that. Still. A comparison between the Opteron Piledriver and EPYC would also be...uh...Epic.

[johnnydoe9]

Been thinking of getting an update on my rig so I can play some games I've had my eyes on for a while, but this is so damn far ahead in performance that I can't even think of something to do with it. Like will you run a couple different benchmarks on it at once idk. 

[NoxiousOdor]

Screw the cpus, I am a huge fan of xeons but I have been so excited to see this motherboard. I hope they end up cooling it properly because you can OC and it is super coolio

[Tysonr]

Would totally be interested in seeing video about the Xeon W CPUs mentioned at the beginning. 

[zENjA]

Some of my customers run Dell R940 with 4 CPUs, Mellanox Connect-X 5 and MSN2700-CS2F as their hyper convertged setup in my Datacenter, but back to topic.

Build you own Nvidia Grid solution by allocate GPU performance dynamicly to the user. And at night then the editors aren't working use it as an in/output render system.

Or do Remote editing on the vGPUs with RemoteFX of the MS Server 2012/2016 with Tesla M10's. ALike Morten from My PlayHouse did it:

And than combine it with 40G/100G Networking with Mellanox SX1012 (or faster) and ConnectX 3 / 5 cards .

[slurpderp]

You should try deep learning benchmark and compare it to GPUs.

[Luscious]

YAWN... all show and no go.

 

Wake me up when you build an ACTUAL WORKING SYSTEM with this - you know - inside a case, with water blocks on the CPU's instead of those handicapped 140W 2U ding dongs you KEEP SHOWING.

 

That motherboard is seriously DOA without water blocks for the CPU and ASUS knows it. Socket compatibility aside, the blocks also need to be able to handle 205W heat output that an 8180 can and will do at full tilt, which no block design on the market is currently capable of. Then there's the third aspect of actively cooling the naked RAM to put some airflow near the VRM's as well.

 

Tackle those hurdles first.

 

Honestly, as much as I would like to see a beast dual CPU system like this run overclocked and water cooled, the demand for it in the real world just isn't there. Companies have tried water cooled servers before and the products have flopped. Air cooled workstations are much cheaper, simpler to build and require minimal maintenance to run compared to a custom loop. Don't hold your breath for an Epyc water block either - those no-name alibaba/ebay knock offs have no place near a $10K CPU investment.

[The Gaige]

BUT, can it play Crysis? 

[zooks97]

Hey I'm not sure if I'm saying anything new here, but I'm interested in seeing more scientific benches when these crazy builds get posted. In my research group, we use a bunch of codes, but I'm most familiar with quantum espresso, which I don't think supports GPU acceleration, so seeing how it performs on these high core count CPUs would be really cool. Getting good benchmarks for QE and other free, open source codes like LAMMPS, GROMACS, ABINIT, and more would be pretty novel as far as I can tell and potentially useful for the development of the tools and for researchers looking to buy new hardware for their labs.  

 

It looks like LAMMPS has a page dedicated to benchmarks here: http://lammps.sandia.gov/bench.html, and it looks like it could use info on anything from the last couple years.

Quantum espresso has some here: http://www.quantum-espresso.org/benchmarks/

GROMACS: http://www.gromacs.org/About_Gromacs/Benchmarks

 

Even though most of us have access to supercomputing resources, I think using some of these to stress new processors and GPUs (if they're supported) could be useful advice, would provide a decent challenge, and could give some nice animations if you choose your simulation well.

[GDRRiley]

5 hours ago, Ryan_Vickers said:

AMD can't afford to send them free CPUs "just cuz"

I am sure if they asked supermicro they would send over a unit with 32 core CPUs in a 2U or 4U form factor.

[vanished]

Just now, GDRRiley said:

I am sure if they asked supermicro they would send over a unit with 32 core CPUs in a 2U or 4U form factor.

Maybe... we can hope   One thing I want to see is a dual EPYC 7601 system with both overclocked to 4 GHz

[GDRRiley]

Just now, Ryan_Vickers said:

Maybe... we can hope   One thing I want to see is a dual EPYC 7601 system with both overclocked to 4 GHz

I would drive one up from CA to CA if supermicro didn't want to ship it. I live only a few miles away from their headquarters  

[indrora]

Seconding @zooks97 on getting a set of benchmarks from GROMACS and friends. NASA's Supercomputing group has a series of benchmarks: https://www.nas.nasa.gov/publications/npb.html -- You'll need to compile them to get them running: They're built to be optimized heavily for the *specific* chip you're running them on. Go find a computational mathematics graduate at a local university and have them help you.)

 

The frustrating part is that there's no "good" benchmark that isn't... vaguely synthetic. You could blow the entire Folding@home world out of the water with something like this (Seriously, FAH would love you for running this for a day!) but for real world benchmarks, you're going to need some... subjective tests.

 

I don't know if there's anywhere you can get 100MP/200MP RAW DNGs for play with in software light Aftershot/Lightroom/UFRaw. Tools like the RED production workbench might be fun to play with. These situations, where you have a metric fuckton of compute power to toss at the problem are great when you're dealing with these sorts of highly parallelizable but very intensive problems. (in a side note: Can LMG get in touch with Phase One and review their *achromatic* -- yes, black and white -- 100MP IQ3 medium format? Maybe a Hasselblad? That'd be cool)

 

Figured I'd nitpick the windows file search comment: It only kicks in after four characters if you haven't let the prefetch cache catch up -- which can take a few days in some cases. It's also case sensitive (*yup, I know, It's dumb!*) in those cases -- it follows Wikipedia rules. None of the prefetch caching happens if you tweak undocumented registry values (or don't have an MSA attached to the account), from what I've encountered.

[Thiago Miro]

What about gaming with 16 monitors

[vanished]

2 minutes ago, Thiago Miro said:

What about gaming with 16 monitors

 

[Thiago Miro]

Just now, Ryan_Vickers said:

 

I watched this video. =)

I'd like to see a game running at least 30 fps in 16 monitors.