|
unixbeard posted:I'm about to get a Nvidia 970. I want to play games (windows) and do machine learning (linux). I thought I would be able to spin up vm's depending on what I was doing, but apparently the GPU device does not pass through with VMWare, even using ESXi. Is this correct? To be able to do both is my only option to dual boot? Only way to do gpu passthrough is a Linux kvm-qemu setup.
|
#
?
Apr 23, 2016 08:47
|
|
|
|
| # ? May 31, 2024 02:26 |
|
|
More Pascal leaks: http://videocardz.com/59288/nvidia-geforce-gtx-1080-gp104-400-gpu-pictured-paired-with-gddr5x-memory http://videocardz.com/59310/leaked-gp104-400-is-from-msi-geforce-gtx-1080-gaming-8g Now they think GP104 will be sold fully unlocked  Going on that assumption and the more accurate measurement, the GTX1080 should be around 5900 GFLOPS (980ti is 5632). repiv fucked around with this message at 14:41 on Apr 23, 2016 |
|
#
?
Apr 23, 2016 14:33
|
|
|
A website called videocardz citing wccft, wow, must be accurate!
|
|
#
?
Apr 23, 2016 15:10
|
|
|
unixbeard posted:I'm about to get a Nvidia 970. I want to play games (windows) and do machine learning (linux). I thought I would be able to spin up vm's depending on what I was doing, but apparently the GPU device does not pass through with VMWare, even using ESXi. Is this correct? To be able to do both is my only option to dual boot? What are you using for machine learning now? As you hinted at, a lot of tools only work on Linux, so you will probably just want to dual boot. If you're just starting out, this is probably a more reasonable time to take the common advice of 'wait for Pascal' or to buy a better card like a 980ti, because the Edit: Apparently the 970 is 5.2 and not 3.5 like i had thought. See this site for the compute capabilities, assuming you want to use cuda. ItBurns fucked around with this message at 15:30 on Apr 23, 2016 |
|
#
?
Apr 23, 2016 15:13
|
|
|
Dead Goon posted:A website called videocardz citing wccft, wow, must be accurate! Eh, videocardz actually seems to put effort into what they do and not stack an article with repeat filler. Videocardz is at least a 3 on a scale of 10 to WCCFTech.
|
|
#
?
Apr 23, 2016 15:39
|
|
|
Dead Goon posted:A website called videocardz citing wccft, wow, must be accurate! Hey, wccft is nearly always right! They're also always wrong on the same topic at least three different ways.
|
|
#
?
Apr 23, 2016 17:14
|
|
|
So remember how AMD seems to be targeting a really good price/perf curve? Samsung just announced 14nm LPC, which reduces cost (significantly?). That and Samsung seems ready to deploy 10nm for 2017, about on time for Zen+ and Navi.
|
|
#
?
Apr 23, 2016 17:26
|
|
|
i think the 1080 will be ~$699+. the 980/970 were based on a extremely mature 28nm 5 year old process and priced accordingly
|
|
#
?
Apr 23, 2016 17:40
|
|
|
calusari posted:i think the 1080 will be ~$699+. the 980/970 were based on a extremely mature 28nm 5 year old process and priced accordingly I think it will be free and nvidia will move to serving in-game ads as a source of revenue.
|
|
#
?
Apr 23, 2016 17:46
|
|
|
ItBurns posted:I think it will be free and nvidia will move to serving in-game ads as a source of revenue. Dlc gpus is the next big thing.
|
|
#
?
Apr 23, 2016 19:30
|
|
|
calusari posted:i think the 1080 will be ~$699+. the 980/970 were based on a extremely mature 28nm 5 year old process and priced accordingly I'd think a $600-$650 target would be a better plan, especially if they're expecting AMD to be competitive this time around.
|
|
#
?
Apr 23, 2016 19:37
|
|
|
DrDork posted:If the 1080 is <10% faster than the 980Ti, $700 would be a dead-in-the-water price point unless availability was really low. By today's prices, that'd be a 25% premium for minimal performance gain today, and vague promises of better performance "in the future" with VR and DX12. Intel has made 3-5% gains the new normal.
|
|
#
?
Apr 23, 2016 19:44
|
|
|
And is laying off 10% of their workforce.
|
|
#
?
Apr 23, 2016 20:04
|
|
|
I thought about using the EVGA step-up program and get a new GPU for Total Warhammer, but then I realized EVGA only makes nvidia cards and holy poo poo nvidia is getting hosed next gen.
|
|
#
?
Apr 23, 2016 20:08
|
|
|
We'll see, but I don't expect them to get hosed. Rather, I expect AMD to stop face-loving themselves and actually be competitive again. The real thing I want to see addressed is how much do we expect to be able to squeeze out of Pascal via overclocking. I mean, it's all well and good to say that the eventual 1080Ti should be 30-50% faster than the 980Ti, but if that's stock vs stock and the 1080Ti only can push another 5% in overclocking (see: current AMD offerings), then suddenly that performance edge shrinks to more like 5-20% for everyone who's not completely retarded. Of course the whole DX12 issue might make the practical performance gains massive anyhow, but it'll be a hot minute before most of the games coming out get the DX12 treatment.
|
|
#
?
Apr 23, 2016 20:24
|
|
|
DrDork posted:We'll see, but I don't expect them to get hosed. Rather, I expect AMD to stop face-loving themselves and actually be competitive again. The GP100 whitepaper was already showing a normal clock speed of like 1200 with a boost clock of 1400. The TDP is 300w, I don't you'll very much room for overclocking.
|
|
#
?
Apr 23, 2016 21:04
|
|
|
Riso posted:Only way to do gpu passthrough is a Linux kvm-qemu setup. And even then, you need dirty hacks to get it working with NVidia consumer cards, because NVidia wants you to buy GRID cards for that purpose, and disables it for other cards in their drivers when they detect a passthrough setup.
|
|
#
?
Apr 23, 2016 21:51
|
|
|
Duuuuuaallboot
|
|
#
?
Apr 23, 2016 22:28
|
|
|
wargames posted:Intel has made 3-5% gains the new normal. graphics is pretty close to being embarrisingly parallel and we have not one but 2 die shrinks so just slapping more shader cores and ROPs would increase perf in proportion
|
|
#
?
Apr 24, 2016 01:16
|
|
|
I'm getting a new machine primarily for gaming, but I would like to experiment with deep learning, just running examples, playing with tensorflow, stuff like that. If I ever do any heavy stuff I'll probably just use AWS. As such, I don't really mind if I'm not getting 100% from the GPU. Basically, I'd be willing to compromise on performance for convenience at this stage, it'd be cool if I could just boot a linux vm and have some sort of GPU available, and it's not clear to me how that might work with a 970 or any consumer/gamer level device. Waiting for Pascal is sensible advice but without a clear release timeline I'm not too fussed. I've never had a gaming machine so a 970 will be a big step up from what I currently have, I figure I can just upgrade the card later down the line.
|
|
#
?
Apr 24, 2016 08:20
|
|
|
If that's the case then you don't necessarily need to use the GPU and could use a VM. There are also other frameworks that support windows.
|
|
#
?
Apr 24, 2016 15:07
|
|
|
Anybody wanting a rift. Bestbuys bundles are going in and out of stock. Order a bundle, cancel the pc after the order and the rift will ship on its own. Ordered saturday afternoon and just got tracking.
|
|
#
?
Apr 24, 2016 17:58
|
|
|
Don Lapre posted:Anybody wanting a rift. Bestbuys bundles are going in and out of stock. Order a bundle, cancel the pc after the order and the rift will ship on its own. Wow. People are gonna lose their poo poo over that lol
|
|
#
?
Apr 24, 2016 19:11
|
|
|
THE DOG HOUSE posted:Wow. People are gonna lose their poo poo over that lol https://www.reddit.com/r/oculus/comments/4g4m4x/people_are_ordering_riftpc_bundles_on_bestbuy_and/ Doctor's warning: not suitable for those on a low-sodium diet
|
|
#
?
Apr 24, 2016 21:11
|
|
|
Don Lapre posted:Anybody wanting a rift. Bestbuys bundles are going in and out of stock. Order a bundle, cancel the pc after the order and the rift will ship on its own. Yeah, they absolutely screwed people who actually pre-ordered the Rift. Even the launch of the Vive, where HTC used Digital River as their card processing agent, who proceeded to screw up charging credit cards; managed to actually get the Vives out of the door within the first couple of weeks. It actually looks incredibly smooth in comparison. vv It makes sense that they'd want a great retail presence, but drat if it doesn't look like the middle finger to all the people that wanted to support Oculus. HalloKitty fucked around with this message at 21:18 on Apr 24, 2016 |
|
#
?
Apr 24, 2016 21:13
|
|
|
HalloKitty posted:Yeah, they absolutely screwed people who actually pre-ordered the Rift. TBF Amazon/MS/Bestbuy would have prepaid a long time ago for their allocation and had contracts, not to mention marketing materials and poo poo already in the works. Im gonna monitor ebay prices and if they stay good i might just keep my oculus.com preorder as well.
|
|
#
?
Apr 24, 2016 21:15
|
|
|
I'm surprised there is a wait on the rift with such a lukewarm reception on a very niche market.
|
|
#
?
Apr 24, 2016 21:24
|
|
|
When your production capacity is equally low, it's easy to run out. Alternately, "intentional scarcity" to drive up interest and demand.
|
|
#
?
Apr 24, 2016 21:33
|
|
|
It's a supply-chain issue.
|
|
#
?
Apr 24, 2016 21:41
|
|
|
Yea lucky confirmed there was an issue getting one of the components but didn't specify which for obvious reasons. Wow, my BB order is showing delivery Tuesday with the free shipping. Even got an eve code. EDIT: ups has mine in louisville KY where its coming out of. Same place preorders are doing. So looks like these are being drop shipped from oculus own warehouse lol. Don Lapre fucked around with this message at 00:38 on Apr 25, 2016 |
|
#
?
Apr 24, 2016 21:44
|
|
|
DrDork posted:When your production capacity is equally low, it's easy to run out. Alternately, "intentional scarcity" to drive up interest and demand. I dunno. I'm pretty quick to jump on manufactured scarcity but I dont think that'd be beneficial for such a huge investment on a gamble of technology. And its clearly hurting their image overall right now. People are pissed. I'd imagine they'd want to sell as many as possible as perfectly as possible to a first generation crowd to increase the odds of there being a successful second generation. But I mean I guess if anything its not a huge stretch to imagine a company like Facebook having trouble delivering their first manufactured product that also happens to be the first of its kind (basically).
|
|
#
?
Apr 25, 2016 00:33
|
|
|
THE DOG HOUSE posted:I dunno. I'm pretty quick to jump on manufactured scarcity but I dont think that'd be beneficial for such a huge investment on a gamble of technology. And its clearly hurting their image overall right now. People are pissed. I'd imagine they'd want to sell as many as possible as perfectly as possible to a first generation crowd to increase the odds of there being a successful second generation. But I mean I guess if anything its not a huge stretch to imagine a company like Facebook having trouble delivering their first manufactured product that also happens to be the first of its kind (basically). The issue is it only takes one company to gently caress up the launch who cant deliver their widget.
|
|
#
?
Apr 25, 2016 00:39
|
|
|
Don Lapre posted:The issue is it only takes one company to gently caress up the launch who cant deliver their widget. Wise.
|
|
#
?
Apr 25, 2016 01:01
|
|
|
unixbeard posted:I'm getting a new machine primarily for gaming, but I would like to experiment with deep learning, just running examples, playing with tensorflow, stuff like that. If I ever do any heavy stuff I'll probably just use AWS. This guy's guide to GPUs for deep learning development is pretty spot on. I expect it to be updated for Pascal once it launches. I have trouble justifying a Pascal GPU for my deep learning hobby but if I could get a pretty cheap (< $400) 980 Ti I may be interested. I'm mostly interested in the high memory models unfortunately for a game-change in how I work on my pipelines. For myself, if I somehow manage to get the time to both play games and work on deep learning / belief networks aside from my rather involving work, sure a Pascal GPU is probably appropriate and hardly an expense but a necessity, but until then I'm going to slum it on a GTX 970. PS. Resizing all images down to an optimal size for your deep learning problem is probably a Good Idea if you're doing computer vision down to even the tiny images in CIFAR-10/100. I'm experimenting with images that are way larger than they should be probably to measure and better understand precisely how the net works for different image resolutions. That and I'm working on perceptual hashes via GPGPU methods.
|
|
#
?
Apr 25, 2016 01:29
|
|
|
So, I left my window open above my PC and it rained. Woke up in the morning and the PC wasn't posting so after a bunch of unplugging everything I have narrowed it down to my 290x being hosed. I can now boot the machine using the iGPU but it's killing me that the most expensive part of the box is now ruined. Am wondering if there is any way to fix this card? As far as I can tell there's nothing 'physically' wrong with the part, it just fails to boot. If I boot with it as the main GPU, the PC fails to POST, however if I boot it as the secondary GPU (iGPU as primary), Windows detects the Radeon 200x card as 'not working' - but it's at least showing up. The only other option I can think of may be that the PCI-e extra power rails in the PSU have given up and it's not giving the GPU enough power to fully initialize. I know this is a long shot but I'd rather not have to go drop another $600 on a Fury X to replace it (I just went and bought a 40" 4K monitor and can't even see it in 60hz/4K because the iGPU is so underpowered). Any ideas? I've read people doing the 'oven trick' but that seems to be the last possible resort for this kind of thing.
|
|
#
?
Apr 25, 2016 02:18
|
|
|
jt posted:So, I left my window open above my PC and it rained. Woke up in the morning and the PC wasn't posting ya dun goofed
|
|
#
?
Apr 25, 2016 02:22
|
|
|
Panty Saluter posted:ya dun goofed Not even kidding. My setup was practically perfect waiting for HTC Vive to arrive. The suffering is real
|
|
#
?
Apr 25, 2016 02:26
|
|
jt posted:So, I left my window open above my PC and it rained. Woke up in the morning and the PC wasn't posting so after a bunch of unplugging everything I have narrowed it down to my 290x being hosed. Put it in a big ziplock bag with a cup or two of uncooked rice and wait a week, the rice will pull any moisture out of the card and that might save it.
|
|
|
#
?
Apr 25, 2016 02:40
|
|
|
AVeryLargeRadish posted:Put it in a big ziplock bag with a cup or two of uncooked rice and wait a week, the rice will pull any moisture out of the card and that might save it. I've saved an mp3 player (creative zen with flash memory) that went through a complete wash cycle this way, and also an iPhone that my wife dropped in a bathtub. The other thing is if you happen to have any of those desiccant packs those would work better than the rice I'd wager.
|
|
#
?
Apr 25, 2016 02:48
|
|
|
|
| # ? May 31, 2024 02:26 |
|
|
jt posted:Any ideas? I've read people doing the 'oven trick' but that seems to be the last possible resort for this kind of thing.  If you can't get it working again, do not buy a Fury right now--hell, if you MUST get something, get a used 380 for <$200 and slum it for a bit. Both Polaris and Pascal should be hitting the streets in the next few months, and if nothing else that should tank the prices of current cards, while offering shiny new options, as well.
|
|
#
?
Apr 25, 2016 02:51
|
|