This is a great suggestion I will do a table.
Can someone check for me if memtest86 works 100% on this board either v1 or 2. Mine gets stuck on startup “collecting memory controller info…”
R2.0 is OOS now. Strange.
Newegg was the only retailer that seemed to be selling this R2.0 board. Early last week, I placed an order for one. At the time, the Newegg website said “Release Date 10/28/2022”. It did not say “pre-order”, I was able to add to cart and order normally. I had assumed Newegg had stock and was going to ship on 10/28.
Come 10/28, I noticed mine wasn’t shipped yet, so I called in to Newegg to see about the delay. They said they’re not shipping the boards yet. Not only that, they haven’t even received any stock from the distributors! When I asked when stock would be available, they said they have no concrete date. Could be 1 month, could be 6 months from now. Essentially I was told my order was only a “pre-order” and I had to wait indefinitely… I have all the other parts, I was just waiting on this board to complete my build.
Man, what a bait & switch from Newegg! Why put “Release Date 10/28/22” and accept orders if they don’t even have any stock to release on that day? I don’t think you can call it a “pre-order” if there’s no concrete release date. At this point it’s a Kickstarter campaign. 
I think after I called in, they changed their system to stop accepting new orders they can’t fulfill anytime soon, hence why it now says “Out of Stock”.
I talked with a local computer retailer and they said they could special order the board for me. But it won’t come in until end of Dec or Jan 2023. At least they were being honest with me about their timeline. Not sure what to do at this point, need a board, but there’s no good viable options for sWRX8 anymore. Will a Threadripper Pro 5000 series CPU work with a generic socket SP3 board for EPYC?
Maybe ASRock realized how much backlash it created lol
No — it will physically fit and the pinout is largely the same, but the design intentions and enabled functionality are different. Threadripper is designed for use with a chipset, while EPYC is chipsetless and has additional management ports; the memory controllers have different signaling enabled; etc. The combination won’t boot out of the box and is not supported by anyone.
I have seen some reports of people hacking bits of firmware together to get particular 3000 series TR Pros to boot on particular EPYC Rome boards, just for experimentation’s sake, but that’s about it.
1 Like
You can try to find someone mod the bios for you. I’ve heard story about people using 3975WX on H12/H11SSL. Maybe I am wrong. But it’s just add a AGESA uefi module to the bios. (easier said then done.) But actually if you have a progremmer and tinker about UEFI tools.
Right now it seems that ASUS is still the best one to go. This is a workstation, Full tower or 2U/4U. So EEB is not issue.
As for thunderbolt, just buy a titan ridge and short the pins, it works.
As for VGA and IPMI eth, probably just eat it. Don’t use them often, but it works.
7T83 7V83 7763 is actually cheaper than 5995WX (sometimes even cheapter or similart to 5975WX) if you know where to find it and risk a bit.
stucks for very long (maybe 10-15mins). Try to disconnect all things and try. Was able to run the test eventually.
The Supermicro MBD-M12SWA-TF says “Socket sWRX8/SP3”. Does that mean it supports both EPYC and Threadripper? So the other way around works, sWRX8 motherboards can take EPYC CPUs?
The Asus board as it seems to have known bugs, at least according to this forum. It’s also quite wide for an EATX (330mm). My case only supports EATX up to 285mm. The Asrock board is only 270mm.
No, the two product lines are not considered interchangeable.
I don’t know for certain why Supermicro lists it that way, but my guess would be the mechanical compatibility for accessories: SP3, TR4 / SP3r2, sTRX4 / SP3r3, and sWRX8 / SP3r4 are all LGA 4094 with the same mechanical specifications. So e.g. Supermicro’s heatsink part matrix categorizes everything under SP3, even though they’ll fit any of those sockets. (Note there’s often an airflow orientation difference between board lines, so you’ll want to pay attention to that when choosing coolers from any vendor.)
Is the Marvell 10GBe going to be a problem if I plan to use Proxmox and virtualization?
I was looking to get a WRX80 motherboard that supports PBO, afaict the only options are the MSI and the ASRock boards, but MSI PBO doesn’t seem to work yet. That said I don’t think I can actually find any ASRock WRX80 r1 or r2 boards currently for sale.
The asus one has 8 sata ports of which 4 are shitty asm just avoid that. If you need more sata ports, buy a u.2. to sata and set U.2 mode in bios from pcie to sata to get another 4 good sata ports. Other than that, the board is good build.
1 Like
The marvel 10gbe does not support sriov. So you have to use macvtap or bridged device for vm virtual nics, which relies on CPU to steer the traffic. While sriov is a hardware feature.
Well, you can buy a mlnx nic for cheap it supports sriov. wrx80 has too many pcie slots. Also your cpu is going to powerful enough to steer the traffic. But sometimes sriov makes your setup cleaner and preserves sanity when manage them.
I have rev 1 but with the 6.06 bios (released 10/27), I can swap the onboard video to enabled (instead of auto) and get video on both the vga and pcie ports. this means kvm via ipmi works perfectly. I haven’t tested loading a linux console fb via kvm while X is running (I’m still doing ram stability testing) but I expect it to work fine. you should also be able to get kvm access if you leave the console fb enabled and just enable the onboard vga, without actually plugging in a monitor via vga.
older versions of this bios forced you to choose between pcie video and vga but the newest version doesn’t allow you to disable pcie graphics and allows you to force onboard video on which in practice enables them both.
on a different note, has anyone found that PBO is kinda broken on this board? I have a decent AIO on the cpu (5975x) and it clocks 4.3ghz all core at full load. if I enable PBO, nothing changes. if I increase the PBO scalar/PBO limits, the voltage increases but I’m still capped at 4.3ghz all core and 4.5ghz single core. if I add a 200mhz offset to the frequency, the all-core frequency hits 4.5ghz but the single core frequency doesn’t go up to 4.7ghz. why is the frequency capped like this? a curve optimizer offset of -30 works perfectly which means my cpu has quite a bit of headroom for higher clockspeeds – but I can’t use it because of the board/PBO. my CPU gets to 55C max so there’s a lot of headroom going unused.
lastly, I’m having a beast of a time stabilizing 256GB of 3600MT/s, non-ecc ram (I had it from a previous build and thought I’d save a few bucks as I was already pushing my budget with the cpu). what I’ve learned, as there’s basically no information on google that I can find:
cpu: Threadripper Pro 5975x (Zen 3)
motherboard: Asrock WRX80 Creator
memory: G.Skill Ripjaws V 32GB x 8 (F4-3600-C18Q-128GVK, 04266XS820M, Hynix MJR [I think?])
- this board is very sensitive to cpu and memory seating. it took me several attempts to get the board running the ram at 2666 stably.
- the xmp profile will yield lots of data corruption errors when the cpu is loaded. if you test the memory without loading the cpu (i.e. memtest), you may or may not find this corruption. I’m consequently using GSAT (Google’s stressapptest) to validate stability.
- dual rank, 8 channel memory configurations are very hard on the memory controller. the only way I could get 3200 or 3600 MT/s stable was to raise the VDDP voltage to 1.05V, SOC voltage to 1.2V, IOD voltage to 1.15V, and CCD voltage to 1.10V, with the ProcODT to 68.6, and the DATA_BUS at:
- RTT_NOM = RZQ/2
- RTT_WR = RZQ/1
- RTT_PARK = RZQ/1
but this RTT_PARK overlap causes memory training issues some of the time so I’m going to try setting it to RZQ/3 to ensure retrain consistency; dunno if it’ll affect stability. I also had to set the CAD_BUS as follows:
- ClkDrvStr: 120 Ohm
- AddrCmdDrvStr: 24 Ohm
- CsOdtDrvStr: 24 Ohm
- CkeDrvStr: 24 Ohm
ClkDrvStr at 60 Ohms also works when GDM is enabled – however, I’ve found disabling GDM improves performance by ~4% so I’m hoping I can keep it disabled and command rate at 1T.
I’ve also found that the write related secondary timings need to be kept pretty loose in order to avoid data corruption on these sticks:
- tWR - keep this at tCL; it won’t come down to 12 like on most other configurations.
- tWTRs/tWTRl - I haven’t been able to get any stability if these come down at all from their default values of 5/14.
SCLs do fine at 4/4 and the tRRDs/tRRDl/tFAW at 4/4/16 works perfectly.
the big oof is CAS latency. I can’t get the OS to boot at CL18 or even CL20. even CL22 requires 1.4V to the memory. I suspect that if I could find stability at a lower VDDP and ProcODT, this would come way down. luckily, CAS latency doesn’t seem to affect either bandwidth or memory latency very much as the initial setup time in a long sequences of accesses is kind of irrelevant.
I’m also still working on finding the right AddrCmdSetup time – 61-63 are all pretty stable but if I can dial this in exactly right, I have a bit more room to experiment with lower voltage configurations.
I haven’t completed this testing by any means - it takes a very long time and there’s very little documentation online about how to stabilize 8 channel, dual-rank configurations. so I’ll keep updating this info.
all that said, I’m never paying money for non-ecc RAM ever again. this has taken me two weeks to stabilize the god damned xmp profile.
1 Like
This only works if the M2 slots are not populated, right? As I understand,the U2 connectors and the M2 slots share the same bandwidth. So you can only use one or the other. Or does running in “sata” mode on the U2 connector avoid this bandwidth sharing issue?
for asus board one m.2 shares u.2, the other two m.2 is independant. (iirc) check manual.
According to common chipset design perspective, it will still disable the shared m.2. if you switch to sata, but have no time to verify.
1 Like
*The U.2_1 slot shares bandwidth with M.2_2, U.2_2 slot shares bandwidth with M.2_3. When M.2_2 slot is being populated, U.2_1 slot will be disabled. When M.2_3 slot is being populated, U.2_2 slot will be disabled. Each U.2 slots can support up to 4 SATA devices via a transfer cable. the cable is perchased seperately.
But anyway, let’s get back on track, as this thread is about the ASRock board.
I contacted ASRock USA and the sales team told me that the 1st shipment of the WRX80 Creator R2.0 boards will arrive in the 2nd week of November. Which I presume is next week?
Newegg will then need to get the boards from their distributors, process, and then ship to customers based on order date. I’m guessing that’ll take another 1-2 weeks, so we may get the boards by end of November. (Assuming enough boards arrive to fulfill all backorders). I’m also keeping an eye on Amazon and other retailers to see if they pop up there.
1 Like