Fermi support ending in January

[Mark Rose]

The news is on AnandTech: https://www.anandtech.com/show/12624...bit-os-support

It's critical fixes only from now until January, then support will be completely dropped.

But I guess one could still use old drivers. Anyone want to buy some GTX 580s? 😁

[EdH]

Originally Posted by Mark Rose

... Anyone want to buy some GTX 580s? 😁

Would my stone age machines run them?

Actually, I have no idea as to their worth monetarily or to my hobbies, but then again, I'm using a laptop with an arch 1.2 NVidia, that I run with revision 979 of Msieve from time to time. It's getting to be a pain to reinstall the old CUDA package every time I update the OS.

[LaurV]

Owning 4 GTX580 all water cooled, all working. Two Asus DCII cards, with EKWB coolers, and two unknown (they could be Palit or MSI, no idea, but they are Nvidia ref-design, maybe they are Nvidia) with Saphire coolers (all 4 are full block).
Not selling.
Just boasting only.

[retina]

Originally Posted by Mark Rose

The news is on AnandTech: https://www.anandtech.com/show/12624...bit-os-support

It's critical fixes only from now until January, then support will be completely dropped.

But I guess one could still use old drivers. Anyone want to buy some GTX 580s? 😁

If your current card is still working fine then I see no real reason to feel obliged to continually update the software/drivers. All the media pressure to always be on the latest and greatest isn't necessary IMO. If what you have works for you then keep using it.

I'm still using FF 3.6.28, and it still works for me, I see no reason to jump onto the continuous upgrade nonsense and have everything regularly break, or find the interface has been "modernised", or whatever. Yes, I know FF isn't hardware, but the process is the same. No need to change it if it still works.

[kladner]

Originally Posted by retina

If your current card is still working fine then I see no real reason to feel obliged to continually update the software/drivers. All the media pressure to always be on the latest and greatest isn't necessary IMO. If what you have works for you then keep using it.

I'm still using FF 3.6.28, and it still works for me, I see no reason to jump onto the continuous upgrade nonsense and have everything regularly break, or find the interface has been "modernised", or whatever. Yes, I know FF isn't hardware, but the process is the same. No need to change it if it still works.

[kriesel]

Originally Posted by retina

If your current card is still working fine then I see no real reason to feel obliged to continually update the software/drivers. All the media pressure to always be on the latest and greatest isn't necessary IMO. If what you have works for you then keep using it.

Amen to that.

Going to the latest CUDA level, or to the latest driver version, can cost performance. (Testing GTX480s showed maximum performance around CUDA 6.5, for example.)

The gotcha is if running a new card and an old card in the same system, eventually the support for the old card limits CUDA level, while the new card may require a level higher than that. Or the level required for the new card reduces performance of the old card. And we don't get to install multiple driver versions for NVIDIA to be active at the same time for different NVIDIA cards.

On the AMD based RX550, I've seen the February 2018 driver reduce performance by about 5% compared to the April 2017 driver in gpuowl V1.9. However, gpuowl 2.0 wouldn't run with the April 2017 driver and did with the February 2018.

All the above is on Windows 7 64-bit.

There's also total cost of ownership to consider; older cards can have much lower throughput per kilowatt-hour.

[VictordeHolland]

Some are already beginning to feel the effects:

http://mersenneforum.org/showthread.php?t=22866
http://mersenneforum.org/showthread.php?t=23214

One of my cards will also be affected:

lspci | grep -i nvidia
01:00.0 VGA compatible controller: NVIDIA Corporation GF110 [GeForce GTX 570 Rev. 2] (rev a1)
01:00.1 Audio device: NVIDIA Corporation GF110 High Definition Audio Controller (rev a1)

nvidia-smi
Tue Apr 10 02:11:31 2018       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 390.48                 Driver Version: 390.48                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 570     Off  | 00000000:01:00.0 N/A |                  N/A |
| 16%   38C   P12    N/A /  N/A |    349MiB /  1218MiB |     N/A      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0                    Not Supported                                       |
+-----------------------------------------------------------------------------+

nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2016 NVIDIA Corporation
Built on Tue_Jan_10_13:22:03_CST_2017
Cuda compilation tools, release 8.0, V8.0.61
nvcc warning : The 'compute_20', 'sm_20', and 'sm_21' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).

[kriesel]

Stumbled on this method of preventing NVIDIA driver upgrade to a version incompatible with the old gpu cards:

An obsolete OS version on an old computer ensures drivers won't update to a version incompatible with the old gpu.

Bought a pci/pcie adapter to try (experiment) with a pcie1x/16x data and power extender for low power cards, had an old slow Quadro 2000 card (CC2.1), tried the combination in a system that had never had a gpu in it since it was built, and it gave the appearance of that gpu & extender & adapter kludge not working. Except that it appeared to begin to work once.
Dove into investigating it last evening after having closed almost all my Windows Remote Desktop sessions before shutdown/restart of my usual remote management station, to clean up an unrelated issue (hung web browser session there).

Box with the adapter/extender /gpu experiment was XP Pro. That guarantees NVIDIA driver will not be updated to a version incompatible with the gpu, because XP is a no longer supported OS & will not update further than its current driver version.

It turns out that, it's not only RX550 cards' drivers that are problematic, on Windows 7 with GPU-Z sensor data collection (losing display of clock rates and temperature and with the Feb 2018 version, also memory usage) while Windows Remote Desktop is in use, although gpu computations continue, requiring a reboot to recover sensor readings for console display, and the alternative TightVNC has larger issues there, with GPU-Z working but gpu computation not. The necessarily old NVIDIA drivers on XP and a Quadro 2000 goes one or two better, with Remote Desktop causing all sensor readings to stop, and also CUDA to go off the rails, ending gpu computing capability, until the remote desktop session is terminated and the system is rebooted. I've not seen such behavior on Windows 7, which hosts most of my NVIDIA cards. The available Vista and Windows 10 systems with gpus can't be tested with remote desktop because they don't include remote desktop host capability. There's no v8.x present currently so no data there.

[Mark Rose]

I'll just run an LTS version of Ubuntu that includes the old drivers.