Gaming, on my Tesla, more likely than you think

Have You tried to flash the titan gpu bios?

1 Like

I’ve been having issues getting my K40M to actually show up in the list when selecting a GPU in the Graphics settings. I’m using a GTX980(non-ti) on my MSI X99 Gaming Pro Carbon on a 6850k. I’ve verified the hardware is working just fine, both cards on a straight shot to the CPU on x16 links (Both have switchable pair of x8 links). The tesla card correctly jumps up to 8Gt/s when put under load, set my frequency etc.

Gonna have a dig through the win32-videocontroller documentation and see if I cant work something out.

I doubt it could potentially be my Seagate XP6500 causing issues, I popped it out just in case and no dice.

1 Like

DDU all drivers and install the geforce first then Tesla and the try the registry hack

3 Likes

Well other than being unable to find any documentation, I set the hyrbid key to 2 just for giggles and now it shows up as a low-power option.

:man_shrugging: Sure I’ll take it

But I’ll knock everything out with DDU, I noticed the odd extra video adapter identifier kicking around.

2 Likes

You are a beast giga, a gigadrill of a beast.

5 Likes

After hours of pulling my hair out I worked it out, I kept having issues with the geforce driver crapping out the tesla driver along with the order windows had discovered the cards.

Specifically for my setup with a 980 and a Tesla K40M.

  1. Install your drivers. Installed the Tesla driver first in safemode so it appears first in the list in the display adapters class.
  2. Go into regular windows, install geforce driver.
  3. Add “EnableMsHybrid” = 1 to your tesla in “0000”, delete the adaptor entry.
  4. Set your non-tesla gpu to have “EnableMsHybrid” = 2 in “0001”

This is what has given me the entries for “Power saving GPU: NVIDIA GeForce GTX980” and “High performance GPU: NVIDIA Tesla K40M”, anything just wouldnt work.

I assume this just an edge case for my setup.

4 Likes

I think I might know why that is, there are two nvidia GPUs and so they’re grouped sequentially, this is actually a huge break through

3 Likes

You end up effectively running a weird combo driver where the tesla card is actually using the inf supplied from the geforce driver (all because of windows being shite), its been awful trying to get it displayed in that bloody menu

Edit: and it shows up in the Phyx menu correctly too!

2 Likes

I think my k40 did around 1125mhz core

2 Likes

Still fiddling around but hit that just fine, got an extra 30 fps in warthunder (around 100 ish)
I haven’t ramped up my front fans that feed it yet, on the todo list.

1 Like

I had the both the “rocket thruster” pulling air and the front fan pushing it through

1 Like

I couldn’t find a good solution for my grid k2 cooling yet. No purchasable solutions anyways. I’m going to take some measurements and see if I can’t find a suitable centrifugal blower to add to the front of it in some fashion. Because I need something either in the rear like yours @GigaBusterEXE or low profile on the interior.

Local libraries might have a 3D printer, or I can ship you one

2 Likes

I’ll check locally. See what’s avalible out here.

1 Like

Pulled the trigger on a K40 when the video hit YT lol. Having trouble getting over 745mhz. Running CMD as admin… Getting these errors.

1 Like

Errors are fine, now adjust your afterburner core clock, then do the supported clocks
The do nvidia-smi -ac memory,core
Where memory is what it says in supported clocks and core what it says you can boost to

It’s really weird way but it works

1 Like

Like gigabuster said, using the command to specify a fixed clock works fine. The auto-boosting is only on the later cards if I recall correctly.

Heres my silly little cocopops ceral box fan duct running off of a 140mm fan, I have the fan set to change speed according to the K40M core temperature.

I wanted to get my 4.2TB Seagate Nytro XP6500 in here too (goes with the theme of jankyness) but I dont have enough physical space because of my heatsink mod for the ROC silicon. Moved it into my debian server and instead use a SMB multi-channel (5xGigabit) to access it now.

I have thought about getting a hold of arctic aftermarket GPU coolers and making it fit but again physical space is an issue. All will be remedied with Zen 3 threadripper, hurry up AMD!

7 Likes

Not silly at all. Nice work getting the air channeled. :+1:

1 Like

Noticed this on my setup when I was adjusting some settings

Anyone else now get a integrated option in the Nvidia control panel? after setting this here I no longer have to rely on the windows control panel, everything goes straight to the tesla for D3D while all other desktop processes go to my GTX980. (See image blow, ID 0 being the 980, ID 1 being the Tesla)

Edit: Got this in my context menu now too

Works a treat, normally furmark would default to the 980 but here it is launched through the context menu. Im quite happy with this setup its exactly what I wanted.

4 Likes

Thanks man. I guess I was needing a system reboot or did something out of order because it works great now. I got the 1100 config with 3125 memory working. Just rigged up some temporary cooling to do game tests. :smiley:

Also, I think I found a typo in the original post. Bottom one would need to be +250 and -ac 3125,1125

6 Likes