Originally posted by Templar82
View Post
Announcement
Collapse
No announcement yet.
NVIDIA 460.67 Linux Driver Brings A Few Fixes, 5.11 Kernel Compatibility
Collapse
X
-
Originally posted by jabl View Post
Because you'd rather have a piece of hardware even with a driver that gimps cryptomining, than having no hardware at all because the miners are hoovering up all the production?
(A) miners don't buy the card; Or
(B) miners want 2x the cards?
- Likes 1
Comment
-
Originally posted by mppix View PostSo what does "half the hash rate" mean?
(A) miners don't buy the card; Or
(B) miners want 2x the cards?
Since greater density -> lower overheads, I think it's generally the case that having more GPUs that are slower isn't ideal.
Comment
-
Originally posted by coder View PostCute, but if they can make money off a card, then I think the reality is that they'll take every card they can get, house, power, and cool.
Since greater density -> lower overheads, I think it's generally the case that having more GPUs that are slower isn't ideal.
Also we are at the point, where miners build farms with notebooks.
So yeah, they will buy any card from the last generation... and that includes increasingly mobile where a cpu, hdd, screen, etc. happens to be attached.
Comment
-
Originally posted by mppix View PostActually profitability is about energy efficiency and longevity; any miner worth something operates cards at reduced hash rates.
Originally posted by mppix View PostAlso we are at the point, where miners build farms with notebooks.
Comment
-
Originally posted by coder View PostWow. I hadn't realized it was that bad. This crypto craze is really a scourge for the planet.
anyhow, the nvidia hash limit is not to help gamers; afaik, they try to (a) sell cheaper (no display output etc) cards at the same/higher price to miners and (b) reduce the number of mining cards that show up on ebay in the future (and reduce their new gaming card sales)...
Anyhow, Nvidia always used to be 'greedy' and they could because they had limited competition. Some of their latest actions suggests that they may actually be 'evil'. Just my 2c.
- Likes 1
Comment
-
Originally posted by coder View PostSuch as? Just curious.
Example 1: reduced hashrates on some gaming cards; this could open the doors for reduced Cuda capabilities (and indirectly ensuring that mining cards are a separate tier that is worthless after their primary use through driver and lack of display output)
Example 2: vm detection that essentially disables nvidia gaming gpu if passed through to vm (workaround is masking the vm.. but why?). Also, we desperately need an approach to high performance gpu virtualization with 2d/3d acceleration (we can paravirtualize everything! else) ... but nvidia is making that a feature that requires high dollar licenses on high dollar data center cards (that is even locked to specific use cases)
Example 3 (the old story): signing drivers and/or not releasing necessary binaries and/or not supporting some apis: in short they don't want to support som apis and do everything to prevent anybody else from doing so..
Note that I'm typing this on a computer filled up with titan cards for Cuda compute. I'm at the stage where I run the DE in the integrated Intel gpu. This enables me to update the kernel and/or de without breaking my system every few months. Also, Wayland, va-api (for hw Firefox, vlc, etc accel.) just work now. Also, I sometimes pass through a card to a windows vm.. but I don't recommend that to beginners (there is a reason why every tutorial says 'if you have nvidia, you also need....')
Personally, I would not consider future Nvidia purchases if amd introduces rdna2 rocm support or Intel has a solution. i would even trade lower performance for stability (and I look forward to the day where my gpu is as boring as my cpu and 'just works')Last edited by mppix; 21 March 2021, 12:34 PM.
- Likes 1
Comment
-
Originally posted by mppix View Post... Personally, I would not consider future Nvidia purchases if amd introduces rdna2 rocm support or Intel has a solution. i would even trade lower performance for stability (and I look forward to the day where my gpu is as boring as my cpu and 'just works')
The trouble with that view is that GPU designers are always dreaming up new bits of specialized hardware to differentiate their offerings, which complicates matters involving APIs and drivers. And even CPU vendors are in the game of holding back certain capabilities from mainstream CPU.
My wish is that ROCm-support can be as drama-free as graphics acceleration. It used to be that achieving proper OpenGL acceleration, on a particular machine, was a project unto itself. If all the default AMD kernel drivers just worked for graphics and compute out-of-the-box, then they'd have an advantage over Nvidia, because nouveau sometimes causes issues for them (in my experience).
Comment
-
Comment