#2

[M344587487]

Originally Posted by lavalamp

And my least favourite. I'll never buy another gigabyte product as long as I live.

There's always brands to get turned off to due to bad experience, a lot of the time it's bad luck but sometimes there is merit to the accusations. The community suggests Gigabyte have poor bios on otherwise decent AM4 products and I have experienced that first hand, was that your problem?

[chalsall]

Originally Posted by M344587487

There's always brands to get turned off to due to bad experience, a lot of the time it's bad luck but sometimes there is merit to the accusations.

Absolutely. A sample size of one does not a heuristic make.

Personally, I'm somewhat constrained in my immediate COTS options.

I've found Gigabyte and MSI to be fairly good for functioning to spec (taking into proper allowance for the small percentage of DOI you'll find from any supplier).

I don't have the time to drill down, but one MB brand I tried to use once claimed to support both iGPU and discrete GPU at the same time, even though it didn't actually work in the field (yes, the CPU supported video).

(I won't go into the "discussion" that resulted in, between myself and my supplier... )

[Xyzzy]

https://www.gskill.com/product/165/1...V128GB-(4x32GB)

https://pcpartpicker.com/product/rJk...200c16q-128gvk

The memory we chose was not on our motherboard's QVL. The memory's QVL didn't list our motherboard either. But, we got lucky! We now have 128GB of memory running at 3200MT/s!

We originally planned to run just 2×8GB sticks of really fast (expensive) memory but our benchmarking told us that there was not much performance benefit to this. So we went the other direction and filled up all of the memory slots with cheaper and denser sticks.

On the first boot the memory registered at its 2133MT/s JEDEC rating and then all we had to do was enable XMP. Super easy!

This memory is plain. It has no RGB lights but we are not into that stuff anyways.

We paid $429.99 + $41.71 sales tax plus $9.04 shipping for this part.

[lavalamp]

Originally Posted by M344587487

but sometimes there is merit to the accusations

I had a computer with a gigabyte board, which once was a moderately high end mini-ITX board, but was somewhat dated by this time, though still under warranty.

One day it turned off and wouldn't turn back on again. Upon investigation I saw nothing wrong, and the symptom was typical of a PSU protection cut off. So after unplugging the PSU for a few seconds and reconnecting it, it turned back on.

Great success! Then the CPU 12V power connector caught fire. Frantically I yanked the power and blew the fire out. I think I see why the PSU cut out the first time.

Upon contacting and relaying all of this to gigabyte and requesting an RMA code, I was informed that since I had purchased the board in the US, I would need to contact US support, despite the fact that gigabyte is a global company. So I contacted the US support and they informed me I would have to ship it to them ... in the US. There was no compromising to let me ship it to a UK location, as if there was ever any chance they were going to do anything except throw it in the bin on arrival.

The shipping at this point would literally cost more than buying a new motherboard. They were completely disinterested that it burst into flames and had a rigid, unhelpful customer support structure.

Compare this to Crucial, I had bought a set of memory in the US, and since moved back to the UK. A year or so later the computer it was in was crashing which I determined to be due to memory errors, confirmed with memtest. Contacted support, no issues at all, returned the sticks and got a working set back. The new set even had the same US identifying codes and such in the serial numbers.

I know that things break, and I don't expect every product to be perfect (though actual fire is a bit concerning). But when dealing with support Gigabyte were appalling and Crucial were excellent. For that reason I'm never buying Gigabyte again, but I'll go out of my way to buy Crucial SSDs and memory.

tl;dr Gigabyte boo, Crucial yay.

[Xyzzy]

So far we have been unsuccessful in getting OpenCL working with our 5700XT.

We previously got OpenCL working before in build #1 using two "pro" AMD 5500 cards.

We suspect that our "amateur" 5700 is blacklisted from working.

"ROCm" OpenCL definitely doesn't work.

[M344587487]

Originally Posted by Xyzzy

So far we have been unsuccessful in getting OpenCL working with our 5700XT.

We previously got OpenCL working before in build #1 using two "pro" AMD 5500 cards.

We suspect that our "amateur" 5700 is blacklisted from working.

"ROCm" OpenCL definitely doesn't work.

If you've already tried the pro drivers, recent kernels and pleading, the last thing I'd try are the various OpenCL packages on Arch. They are the same libraries you've already tried but packaged differently for a different environment, worth a shot. After that, self-flagellation by manual compilation. I'm led to believe that OpenCL support might as well be called broken, unless something's changed you may be banging your head against a wall for a while yet.

[ewmayer]

@M344587487: I similarly was unable to get a working OpenCL install on the Radeon 540 GPU in the Intel Nuc I recently bought (mainly for its being the only compact-format Intel PC with an avx-512-capable CPU):

ewmayer@ewmayer-NUC8i3CYS:~$ clinfo
Number of platforms                               1
  Platform Name                                   AMD Accelerated Parallel Processing
  Platform Vendor                                 Advanced Micro Devices, Inc.
  Platform Version                                OpenCL 2.0 AMD-APP (3137.0)
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd cl_amd_event_callback 
  Platform Extensions function suffix             AMD

  Platform Name                                   AMD Accelerated Parallel Processing
Number of devices                                 0

Do you think The Arch libs might work here? If so, a link would be appreciated.

[M344587487]

Originally Posted by Xyzzy

So far we have been unsuccessful in getting OpenCL working with our 5700XT.
...

ROCm 3.7 has been released, although there is no official support for Navi there is a rumor that it may work. There's just as good a chance that they're mistaken so don't get too excited: https://www.phoronix.com/scan.php?pa...adeon-ROCm-3.7

Originally Posted by ewmayer

@M344587487: I similarly was unable to get a working OpenCL install on the Radeon 540 GPU in the Intel Nuc I recently bought (mainly for its being the only compact-format Intel PC with an avx-512-capable CPU):

ewmayer@ewmayer-NUC8i3CYS:~$ clinfo
Number of platforms                               1
  Platform Name                                   AMD Accelerated Parallel Processing
  Platform Vendor                                 Advanced Micro Devices, Inc.
  Platform Version                                OpenCL 2.0 AMD-APP (3137.0)
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd cl_amd_event_callback 
  Platform Extensions function suffix             AMD

  Platform Name                                   AMD Accelerated Parallel Processing
Number of devices                                 0

Do you think The Arch libs might work here? If so, a link would be appreciated.

I don't know, I recommended Arch as a last ditch effort that involves a lot of messing about for a small chance of success. It's wishful thinking perhaps, my logic is simply that the distro is rolling release so the environment is more up-to-date, and it's a different packaging system with different maintainers targeting a different environment potentially using fresher source code. To give it a try you'd need to dual boot into Arch and try the various OpenCL AUR packages, you may be able to get away with putting Arch into a VM and doing GPU passthrough but that just introduces another chunk of ways for it to fail: https://wiki.archlinux.org/index.php/GPGPU#OpenCL

[Xyzzy]

Help!

Is this normal?

The case we bought has a plate that plugs into the mains. You install the power supply inside the case and then connect it to this plate. This is because the power supply is mounted deep inside the case so you can't reach the power button on it.

We took a DVM and checked the pins on the plate. If we check the continuity of the pins, then the ground matches the ground but the hot/neutral pair is reversed. That seems odd.

Then we tested if we get voltage with the plate plugged into the mains. Whether we turn the switch on or off we get 125V when we test from the left (neutral?) to ground.

Safe?

It appears that the switch does nothing.

In the mean time we have removed the plate and snaked in the mains cord directly to the power supply.

In the attached picture we have matched the "pins" by color and identified which pair delivers the 125V all the time. The white line shows the path from the wall outlet to the plate to the plug that goes into the PSU.

[henryzz]

Assuming AC it won't matter which way around the pins are as long as earth is in the correct place.
Never come across that sort of design before.

[PhilF]

Unbelievably, it is indeed wired backwards. You can confirm that by measuring the continuity from end to end on a standard AC cable.

Do NOT use it for 120V! Hot and neutral reversed is very dangerous. OTOH, if this is for a 240V installation then it indeed won't matter, but if that is the case the plug doesn't meet code because it is a 120V plug.

I can't believe they would let something like that out of the factory, when obviously it should not have passed even the most elementary QA test. If they have put a bunch of them out that are mis-wired in a similar manner they are open for some lawsuits.