meina222
-
Posts
449 -
Joined
-
Last visited
-
Days Won
5
3 Followers
About meina222
meina222's Achievements
Senior Member (3/3)
149
Reputation
-
@Ploddles - you won't be the only one waiting, mate. I can't afford to spend any time on this before the weekend - I am sure going from like 11.1 to 11.4 and from OC 6.3 to 6.8 on a different BIOS (due to resizeBAR) will take me longer than installing the new beta itself.
-
@Ploddles - what version of OC are you using? Do you mind sharing your latest config plist so I can diff with mine as I have not touched the Hack in months due to selling my 5700XT. Time to try rebooting this project with the 6900XT.
-
Good I kept my 6900XT. Time to move it from my mini ITX back into the Threadripper but before that I need to catch up with all the MMIO BIOS changes and OC updates. Does anyone know if Big Sur 11.4 requires version of OC later than 6.3? I think that's when I stopped updating it.
-
@fabiosun, didn't know how to include image in IM. Lmk, and I can benchmark this puppy.
- 78 replies
-
- 1
-
-
- rx 6900 xt
- rx 6800 xt
- (and 4 more)
-
Finally got a chance to retest the 3090 FE (single GPU as I don't have studio but used free version 17.1). 66 Blur nodes - 27-28fps 6 TRN nodes - 31-32 fps.
- 78 replies
-
- 1
-
-
- rx 6900 xt
- rx 6800 xt
- (and 4 more)
-
@jncasey I have not checked the MMIO map, but I am 100% it did as my PCI id's changed and I upgraded from a non-standard BIOS f4h, which had already different maps compared to f4c and f4l. Also f4p is coming with its own annoying issues such as increased boot time (cycling longer through codes, especially the dreaded 9A about which there is a thread on the Gigabyte forum) and also I use Proxmox - now when rebooting the host after a VM that owned a USB controller shuts, the board gets stuck on 9A and refuses to post back (reset would not work, one needs to shut off power to get that USB to initialize). I raised an issue to Gigabyte. For many, this 9A code became a no-post issue even w/out Proxmox based on the feedback on the Gigabyte board. Resize BAR is not enabled on the 3090s as far as I can tell. This, however, is more likely NVIDIA's fault - they do not list TRX40 and Threadripper Castle Peak as being added to Resizable Bar support - seems the driver itself has to support this. https://www.nvidia.com/en-us/geforce/news/geforce-rtx-30-series-resizable-bar-support/#:~:text=New+GeForce+RTX+graphics+cards,graphics+cards+starting+in+March.?ncid=afm-chs-44270&ranMID=44270&ranEAID=TnL5HPStwNw&ranSiteID=TnL5HPStwNw-14DiO6CW5Ks.pbuyh7AETA In short: I would not upgrade to f4p and I am considering rolling back to f4h (depending on what Gigabyte says - I don't have much faith in them releasing a quick fix, their BIOS development is terribly slow).
- 3,970 replies
-
- 1
-
-
- amd vanilla patches
- amd kernel patches
- (and 3 more)
-
Despite enabling above 4G and Resize BAR to Auto, nVidia control panel still claims Resizable BAR is a 'No' . So did Gigabyte really enable anything? p.s. Looks the fault maybe Nvidia's https://www.nvidia.com/en-us/geforce/news/geforce-rtx-30-series-resizable-bar-support/#:~:text=New+GeForce+RTX+graphics+cards,graphics+cards+starting+in+March.?ncid=afm-chs-44270&ranMID=44270&ranEAID=TnL5HPStwNw&ranSiteID=TnL5HPStwNw-14DiO6CW5Ks.pbuyh7AETA Why not include support for Threadripper and TRX40?
- 3,970 replies
-
- 1
-
-
- amd vanilla patches
- amd kernel patches
- (and 3 more)
-
😞 24fps using 66 blur nodes. However I noticed my 2nd GPU was only 5% core usage and when I tried to enable 2, it told me I am being too cheap.
- 3,970 replies
-
- 2
-
-
- amd vanilla patches
- amd kernel patches
- (and 3 more)
-
Ultimately I managed to finish the install by passing the target nvme to my Windows VM and running the initial install from the installer iso from there, then finishing on the bare metal itself. What a mess. @fabiosun Yes, I decided to keep the 2 3090 FE cards I got a few weeks back at MSRP. Was debating re-selling one or both, but decided to do some GPU compute learning. Set them up in SLI (not sure if that helps with any apps), but took me a while to install them as I didn't want any water cooling and struggled to find a case to run them in Slot 2 and 4 (settled on a BeQuiet 900 Pro in inverted layout and some minor mods - picture is upside down for some reason). I wanted to keep my x16 slot free for my 4x4x4x4 raid NVMEs. Was a challenging re-do of my build. Finally done with Windows baremetal - what benchmarks would you like to see?
- 3,970 replies
-
- 3
-
-
-
- amd vanilla patches
- amd kernel patches
- (and 3 more)
-
@fabiosun, So this 'solution' almost worked for me, but just in the end during the "updating" (after copying all files) phase it still failed with the same error and I could not boot from the target drive still. This trick certainly did something as before the installer could not even start copying the installation files. Other people suggest that one needs to unplug all drives but the target nvme and try the install, but I am not disassembling my entire system because of this. I even tried to install from an old SSD instead of USB as the installation drive - same issue. This is an almost laughable situation - how is one supposed to do a clean Windows install? Do they even test these chipsets before they release them? The irony is Gigabyte officially supports only Windows - if you ever try to point out you have Linux, they will not answer your concerns, yet Linux is the one system that works reliably on this board, except for the USB re-initialization issues when released by VMs.
-
BIOS f4p is a major PITA for the Designare. I can no longer reboot my Proxmox host w/out hanging w BIOS code 9A. I need to power cycle the PC by turning it off (reset does not work). Something with the release of USB controller by guest VM does not play nicely with the re-initialization of BIOS. Total garbage, lots of people are complaining by f4p on the Gigabyte US forum. To top if off (not related to f4p as the same happens on f4h) - I was shocked to find out today I am unable to install Windows 10 bare metal. I never had Windows 10 outside of a VM attempted on this motherboard, so I wanted to try my dual GPU 3090 setup for ML (I have pretty much stepped away from MacOS for the time being till Radeon 6000 cards get drivers) and shockingly the setup failed - both latest Win 10 iso and 2004 versions. This person describes the same problem plus a proposed crazy solution, which almost worked but not quite. I can't believe I cannot install Windows in 2021 on a PC chipset that has been out for more than a year now. Gigabyte, AMD and Microsoft suck 😠. https://answers.microsoft.com/en-us/windows/forum/all/windows-10-fresh-install-fails/f9614ae4-6dbd-4715-9a81-8586b16dfaf7
-
@Driftwood, I noticed this limitation a while ago - on the Gigabyte BIOS-es this causes D4 PCI resource allocation error and refuses to boot. I never researched the exact topology/PCI lanes, but the TRX40 chipset is definitely not as good as the Intel Xeon ones for many PCI devices. On this topic, as of now, WRX80 is available, which should have 7 x16 PCI slots, but unfortunately requires you to also swap your Threadripper for a Threadripper Pro.
-
You can run your memory more stably at 3200mhz CL16 and tight timings than 3600mhz CL18 and get better performance as the IF won't be overclocked beyond spec. Threadripper sweet spot is 3200mhz.
-
@Jaidy, This is a memory corruption error. At some point one of your processes (Google Chrome Helper) requested to allocate memory, which the kernel tried to handle and failed a sanity check. If this is not reproducible it could be due to issues with memory instability or just bad luck. It is hard to know for sure what's happening based on the error alone. Cosmic rays, random memory errors or overclocking instability are possible. You are running 256G of memory at 3600mhz. This is a very ambitious overclock on such size with the Threadripper platform. Since your memory has no error-correction, I wouldn't be shocked if you get memory errors but you just get lucky enough they don't cause kernel or application panics. Linus Torvalds famously (in his usual 'colorful' style) commented on the lack of unbuffered ECC memory and how important he thinks it is to avoid random occasional crashes or even malicious attacks: https://www.phoronix.com/scan.php?page=news_item&px=Linus-Torvalds-ECC I would recommend you scale back your overclock to 3200mhz (you don't lose much of performance on this platform) which matches the officially stable IF of 1600mhz of Threadripper. I have a hard time running 256GB stable at 3600mhz even though my memory is rated (same brand).
-
@Arrakis - wow - so I don't how long it took to load the clip, but it was definitely more than a minute - I left it alone and minimized the window and when I checked 20 min later it was there. This is really bad in practice. So on the Windows VM it's averaging only 21 fps after a long run. It only uses 190w which is about 60-70% of the cards TDP and card doesn't get hot at all.
