2 available but they're $625! Gigabyte Motherboard MZ31-AR0 AMD EPYC E-ATX 1920x1200 DDR4 SATA PCIE USB Retail | eBay
Add a $1K CPU and $1700 for 24C?
Add a $1K CPU and $1700 for 24C?
Gigabyte EPYC MOTHERBOARD FOR SALE
- Thread starter MiniKnight
- Start date
-
- Tags
- epyc
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
We actually have this in a server in the lab. I believe Anandtech and Tom's will be reviewing the board as well.
How soon will the review here be posted? I am closely reviewing the Gigabyte literature to understand whether this board makes any sense for a workstation concentrating on C++ compilation. Presently, the literature is unclear whether populating all memory slots is the best performance. Any memory slot usage suggestions?
Best performance is realistically going to be populating 8 of 16 memory slots.
I do have a fairly strong bias towards recommending EPYC as a server platform over a workstation platform. EPYC clock speeds max out around 3.0GHz which is low for single-threaded UI elements and etc.
The plus side is that the motherboard has the integrated GPU so you can run KVM over IP and access from a notebook wherever you are.
Thank you for your quick reply. I am also comparing Epyc versus Threadripper 1950X as a workstation configuration. It may very well be that the Threadripper platform makes the most sense for C++ compiling. I quickly read the MSI X399 Gaming Pro Carbon review. Are any other X399 boards in the review pipeline?
I think I can answer the review pipeline question... I do have a Gigabyte Designare EX here now and ASUS Zenith Extreme coming.
Like I have said before, I currently run on a ASUS Z10PE-D16 WS, 2x E5-2699v3 and 256GB RAM, ASUS Strix GTX1080 Ti.
Patricks comment on slow clock speeds on these types of processors makes sense but it has been something I have gotten used to. When I do run on these single socket systems I am impressed about how responsive they are, makes for a great user experience.
My plan is to switch over to a TR based system here very soon because of the speedy clocks of TR. I have been running the above WS for many years now and feel the new rig bug biting me LOL.
Like I have said before, I currently run on a ASUS Z10PE-D16 WS, 2x E5-2699v3 and 256GB RAM, ASUS Strix GTX1080 Ti.
Patricks comment on slow clock speeds on these types of processors makes sense but it has been something I have gotten used to. When I do run on these single socket systems I am impressed about how responsive they are, makes for a great user experience.
My plan is to switch over to a TR based system here very soon because of the speedy clocks of TR. I have been running the above WS for many years now and feel the new rig bug biting me LOL.
@William is working on a few. I have been using the MSI X399 with a 1950X and the combo has been working ok.
Another option is a Xeon E5-2600 V3/ V4 used workstation. Pricing is fairly comparable and that is a well supported configuration.
At the office we are currently testing a Z10PE. I unfortunately cannot recall whether it is the D8 or D16. To put it simply, we are having unexpected performance issues with the E5-2650 V4. Most likely there are several BIOS items to modify. I also expect not having all memory slots occupied on the Z10PE is contributing too. In your Z10PE experience, have you noticed whether following the manual usage table for less than populating all slots, needs to explain the performance effects? I have come across a post that the solution to poor performance with the D16 version is to sell all items, including the motherboard. Then use the money on something with fewer memory slots.
When you go to higher DPC configurations, you also need to look at items like memory ranks. As you move up to higher numbers, most chips will lower memory clocks thereby lowering bandwidth.
EPYC has more memory channels, but it also four NUMA nodes per socket. As you lower memory frequency by adding additional DIMMs, AMD Infinity fabric slows and you lose performance.
With Threadripper, you have higher clocks, fewer NUMA nodes and DIMM channels, but you can also run at higher DIMM speeds.
EPYC has more memory channels, but it also four NUMA nodes per socket. As you lower memory frequency by adding additional DIMMs, AMD Infinity fabric slows and you lose performance.
With Threadripper, you have higher clocks, fewer NUMA nodes and DIMM channels, but you can also run at higher DIMM speeds.
With out knowing what performance issues you are having its hard to say.
My board is one of the first released and I have updated the BIOS to the latest. It is old enough that the onboard battery started going out which caused some major issues. I replaced the battery and everything went back to normal.
I run my rig on default settings.
On platforms like this if you disable EIST, Speed Step and other features like that you loose Turbo. The best answer is to keep all power saving features on and let the processors handle Turbo etc.
The D8 and D16 have different board layouts, D16 has more memory slots. Best bet is to look at the manual to check which slots to use if you do not fill them all up. My system has all slots filled.
My board is one of the first released and I have updated the BIOS to the latest. It is old enough that the onboard battery started going out which caused some major issues. I replaced the battery and everything went back to normal.
I run my rig on default settings.
On platforms like this if you disable EIST, Speed Step and other features like that you loose Turbo. The best answer is to keep all power saving features on and let the processors handle Turbo etc.
The D8 and D16 have different board layouts, D16 has more memory slots. Best bet is to look at the manual to check which slots to use if you do not fill them all up. My system has all slots filled.
It looks like a spreadsheet comparing the Epyc and Threadripper memory timing versus throughput is on tomorrow's task list for the office. Then the same table will have to list a couple Xeon models too.
It is possible a default setting in a new Z10PE is different than the original board. I find that the D16 manual indicates Performance in the BIOS | CPU Advanced PM Tuning is not the default.
Thank you for the memory rank information. Now I know the reason for rank appearing on the various approved memory lists.
I am very thankful you guys run the same tests. It is much easier to determine what makes the best fit in various configurations.
It is possible a default setting in a new Z10PE is different than the original board. I find that the D16 manual indicates Performance in the BIOS | CPU Advanced PM Tuning is not the default.
Thank you for the memory rank information. Now I know the reason for rank appearing on the various approved memory lists.
I am very thankful you guys run the same tests. It is much easier to determine what makes the best fit in various configurations.
Thank you for the test. I will check with the office system builder and confirm. I do not remember hearing the AIDA64 test suite is installed. Do you know if the trial version will show the same detail as you have? With Broadwell based Xeons, the speed is supposed to be faster, correct?
I have re-read the Epyc 7401P review. Based on the Linux kernel 4.4.2 compile results, it looks like a single socket Epyc is expected to have better performance than a single socket Threadripper 1950x.
On a related topic, I confirm on Memory QVLs on Gigabyte and MSI, there simply is no ECC memory out there at 3000 MHz. The highest I can find is a 16GB per DIMM DDR4 Samsung. Have people been successful overclocking ECC memory?
I am looking forward to the Gigabyte MZ31-AR0 review.
On a related topic, I confirm on Memory QVLs on Gigabyte and MSI, there simply is no ECC memory out there at 3000 MHz. The highest I can find is a 16GB per DIMM DDR4 Samsung. Have people been successful overclocking ECC memory?
I am looking forward to the Gigabyte MZ31-AR0 review.
Im not sure ECC exists above 2667 mhz for unbuffered (which is what you need on TR).
Overclocking with ECC memory is quite a rare thing to want to do - are you sure you need ECC?
I didnt bother for my TR build as I'm using it as a workstation - not a server, so went for some 2800 mhz memory (memory errors are quite rare these days - relatively speaking).
I thought the risk to my workload is quite low, as though I do production builds on my machine, they get tested before being distributed anyway, so the chances of a memory error while doing a compile, which would not be caught by testing (if the build even completed following the error) are low.
Overclocking with ECC memory is quite a rare thing to want to do - are you sure you need ECC?
I didnt bother for my TR build as I'm using it as a workstation - not a server, so went for some 2800 mhz memory (memory errors are quite rare these days - relatively speaking).
I thought the risk to my workload is quite low, as though I do production builds on my machine, they get tested before being distributed anyway, so the chances of a memory error while doing a compile, which would not be caught by testing (if the build even completed following the error) are low.
Yes I agree the chances are quite low. I work on a team that heavily emphasizes high quality releases. A compilation error on the production build system from non ECC memory would cause issues internally and externally. However, non ECC use on a typical workstation would likely continue to be acceptable.
Perhaps this is a silly question, but would you not be better off just getting a proper server for build tasks for the team?