|
Durinia posted:The GP100 die is aimed squarely at HPC/GPGPU, so that one will certainly have full DP support, and likely the improved scheduling. It's also using HBM. All I want is ISV-certified drivers for a Titan. Jen-Hsun? I'll loving suck your dick for some goddamn drivers that'll let me play games and do Solidworks on the same loving card!
|
#
?
Jan 5, 2016 01:06
|
|
|
|
| # ? Apr 29, 2024 12:09 |
|
|
Don Lapre posted:My uncle works at Nintendo  Someone get wccftech on the horn, pronto!
|
|
#
?
Jan 5, 2016 01:09
|
|
|
japtor posted:In the meantime Nvidia will be branding stuff as "VR Ready". (And apparently their CES conference is tonight) I think the configuration they outline will perform just fine for VR content.
|
|
#
?
Jan 5, 2016 01:43
|
|
|
Subjunctive posted:I think the configuration they outline will perform just fine for VR content. I disagree - the GTX 970 is listed as the "minimum recommended" spec and I think it's going to be very minimum. The Rift is basically a 1440p@90hz device, and since dropping frames is going to make you hurl cookies you really want to be running 90 fps minimum - not average. There are an awful lot of games where even a 980 Ti is not hitting 90 fps average on high settings at 1440p, let alone minimum 90 fps. From the FAQs, the 970 sounds like literally the cheapest thing you can get to run. They've stated that anything below that might be OK for virtual desktop and other things that don't involve a lot of motion, but that such configurations are absolutely and totally unsupported. To my mind that sounds like the 970 is truly the bare-minimum spec, where you can get a workable FPS that won't make you blow chunks, using the absolute lowest-quality settings. Perhaps even requiring pixel doubling. Even so there's no guarantee that every game will be able to target that framerate - there's always been some Crysis-type games that insist on pushing the envelope past any reasonable hardware. I think anyone who bought a 970 based on that spec is going to be very disappointed, or end up buying another for SLI (at which point you should have just bought a 980 Ti anyway). I guess we'll see, but I would absolutely not be racing out to buy hardware before we have benchmarks and reviews on real-world games instead of Job Simulator tech demos with toaster-quality graphics. Even so, you bet your rear end that a 980 Ti has been inside every demo rig. Guess we'll see before too long, here. Paul MaudDib fucked around with this message at 02:33 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 02:25
|
|
|
NVIDIA's CES 2016 event is starting now.
|
|
#
?
Jan 5, 2016 02:59
|
|
|
you are invited to the nvidia demoscene feat. wireframe logo edit: 
|
|
#
?
Jan 5, 2016 03:01
|
|
|
  Forget cars, imagine a datacenter full of these fuckers with a liquid cooling loop pulling the heat out. But what is a DL TOP? e: Grandpa has explained how difficult it is to drive for 15 minutes now. Finally moving on to deep learning. vvvv Ahhh, that makes sense, thanks. Paul MaudDib fucked around with this message at 03:16 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 03:05
|
|
|
Paul MaudDib posted:But what is a DL TOP?
|
|
#
?
Jan 5, 2016 03:14
|
|
|
They've got a neural net which is pretty good at recognizing cars but totally ignores pedestrians  e: Just switched to a new net that does recognize pedestrians. e2: Classifies cars, pedestrians, road signs, and street surface.  And it outperforms their actual drivers in adverse conditions:  Runs at 50fps on a Titan X, currently. Paul MaudDib fucked around with this message at 03:36 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 03:31
|
|
|
Specs on the Drive PX2: a pair of next-gen Tegras, a pair of Pascal GPUs, 6x the total performance of a Titan X. So something like a pair of GP100s and a pair of host Tegra SOCs with their own onboard GPUs. Paul MaudDib fucked around with this message at 04:07 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 04:05
|
|
|
Did Huang accidentally drop Pascal performance? They seem to be GDDR5 Pascal, so GP104, and they're MMX form factor.
|
|
#
?
Jan 5, 2016 04:07
|
|
|
FaustianQ posted:Did Huang accidentally drop Pascal performance? Yeah I think so. If a pair of Pascal GPUs and a pair of Tegra SOCs is 6x the Titan X, assuming that's GP100 we're talking 2x the performance vs GM200. Maybe 2.5x. FaustianQ posted:They seem to be GDDR5 Pascal, so GP104, and they're MMX form factor. I can't really see that being GP104 if it's really that powerful. I think what was on top might have been the Tegra and support chips, while the bottom is the Pascal GPU which was not shown? Paul MaudDib fucked around with this message at 04:14 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 04:09
|
|
|
FaustianQ posted:Did Huang accidentally drop Pascal performance?
|
|
#
?
Jan 5, 2016 04:10
|
|
|
FaustianQ posted:Did Huang accidentally drop Pascal performance? Paul MaudDib posted:Yeah I think so. If a pair of Pascal GPUs is 6x the Titan X, assuming that's GP100 we're talking 2x the performance vs GM200. Maybe 2.5x. I don't think it was an accident, "6x faster" fits with the AlexNet benchmarks they showed:  2800/450 = 6.2x faster, but that's highly specific to ML apps that go ham on FP16 instructions.
|
|
#
?
Jan 5, 2016 04:15
|
|
|
e:repiv posted:I don't think it was an accident, "6x faster" fits with the AlexNet benchmarks they showed: Nevermind the above, then. TFLOPS are what matters most to gaming performance and that's not GP100 level TFLOPS. Still, it's pretty decent that a pair of GDDR5 GP104s (maybe even GP106s?) are keeping pace with a big ol' GM200 (even after you factor in the CPU cores) - in compute performance no less. Needs a big ol' liquid cooler though 
Paul MaudDib fucked around with this message at 04:20 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 04:16
|
|
|
Just tell me when Tomb Raider is coming on PC guys, god drat.
|
|
#
?
Jan 5, 2016 04:18
|
|
|
Dogen posted:Just tell me when Tomb Raider is coming on PC guys, god drat. Nope, but they confirmed their collaboration with "Volvo".  Half-Life 3 Confirmed
|
|
#
?
Jan 5, 2016 04:21
|
|
|
Paul MaudDib posted:Yeah I think so. If a pair of Pascal GPUs is 6x the Titan X, assuming that's GP100 we're talking 2x the performance vs GM200. Maybe 2.5x. Nah, it seemed to be pretty clearly Pascal GPU mock ups and that those PCB were definitely arranged like an MMX card. Excuse the lovely screen caps. Tegra Side  Pascal Side  So maybe it's using Pascal mobile? If it's not a full mock up but functional Pascal is coming along impressively.
|
|
#
?
Jan 5, 2016 04:22
|
|
|
Yeah upon repiv posting the specs I take it back. 2 chips are putting out 8 TFLOPS with 2 Tegra SoCs helping them, vs 7 TFLOPS for the Titan X. It's either GP106 or GP104 with GDDR5 on an MXM. Since 2xGM204 already keep up with a Titan X it's hopefully GP106. GM206 to GM200 is a 2.66x bump in performance (2308 to 6144 GFLOPS), so if the GP106 to GP100 bump is the same and you give 3.5 TFLOPS per GP106 chip then you end up with GP100 around 9.31 TFLOPs, or a 50% bump in performance over GM200. Maybe a bit more depending on what those ARM cores actually can put out. 1 TFLOPS for 12x ARM cores is probably a bit generous but maybe they're factoring CUDA cores onboard the SoCs into the total. The fudge factor in the napkin math probably vastly outweighs such things though. e: What does that mockup say in the way of memory? 8x8Gbit chips or 8x4Gbit chips, do you think? Paul MaudDib fucked around with this message at 04:43 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 04:24
|
|
|
Paul MaudDib posted:Nope, but they confirmed their collaboration with "Volvo". 'Crash free future' Not with bethesda still in business.
|
|
#
?
Jan 5, 2016 07:03
|
|
|
Paul MaudDib posted:Specs on the Drive PX2: a pair of next-gen Tegras, a pair of Pascal GPUs, 6x the total performance of a Titan X. 
|
|
#
?
Jan 5, 2016 07:21
|
|
|
Do you guys think it is normal to be getting severe FPS drops (like to ~20 FPS in some places) running Fallout 4 with 970s in SLI at 3440x1440p? I get ~70-100 indoors but it is a nightmare sometimes when I'm roaming outside. Edit: I'm running the game on Ultra but moving to High doesn't seem to affect the FPS at all. beergod fucked around with this message at 07:33 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 07:30
|
|
|
Where are all the Thunderbolt 3 eGPU enclosures? 
|
|
#
?
Jan 5, 2016 07:33
|
|
|
beergod posted:Do you guys think it is normal to be getting severe FPS drops (like to ~20 FPS in some places) running Fallout 4 with 970s in SLI at 3440x1440p? I get ~70-100 indoors but it is a nightmare sometimes when I'm roaming outside. Are you running 359.06 or later drivers? FO4 has been problematic with SLI since launch, and even with the new drivers it's basically just an "official" hack (loading in some KillingFloor 2 bits to get SLI to work), so it's not exactly the most polished of things. Sad to say, but frankly you're lucky SLI works at all, Bethesda being what they are and all.
|
|
#
?
Jan 5, 2016 07:46
|
|
|
Paul MaudDib posted:Even so, you bet your rear end that a 980 Ti has been inside every demo rig. We didn't disclose the specific config of the demo rigs, but I don't think I'm out of line pointing out that we demoed Crescent Bay (which has CV1's resolution and frame rate) before the 980 Ti was on the market. (And before it was being produced, because we did have some Big Maxwell and Fiji prerelease cards)
|
|
#
?
Jan 5, 2016 08:59
|
|
|
Dogen posted:Just tell me when Tomb Raider is coming on PC guys, god drat. January 28th. Edit: One day off. SwissArmyDruid fucked around with this message at 16:09 on Jan 6, 2016 |
|
#
?
Jan 5, 2016 09:37
|
|
|
beergod posted:Do you guys think it is normal to be getting severe FPS drops (like to ~20 FPS in some places) running Fallout 4 with 970s in SLI at 3440x1440p? I get ~70-100 indoors but it is a nightmare sometimes when I'm roaming outside. Also everything I've heard about FO4 is that poo poo performance is normal. Are you running any mods? Some are supposed to improve performance.
|
|
#
?
Jan 5, 2016 09:46
|
|
|
Malloc Voidstar posted:You shouldn't be running FO4 above 60FPS, since its physics simulation is tied to framerate. It's not the just physics, pretty much the entire game is tied to the FPS. You'll shoot and reload faster, etc. I think it's rather pleasant at 75hz since that's what my freesync maxes at, but going really high can break/crash everything. Lighting in FO4 is brutal and multiplicatively shits things up outdoors. I had a 960 get it's rear end kicked until I pretty much turned off 99% of lightning and set other stuff to medium. I never went below 60fps with a 980ti but I could see SLI 970's having issues. Are you sure the SLI is working?
|
|
#
?
Jan 5, 2016 13:59
|
|
|
beergod posted:Do you guys think it is normal to be getting severe FPS drops (like to ~20 FPS in some places) running Fallout 4 with 970s in SLI at 3440x1440p? I get ~70-100 indoors but it is a nightmare sometimes when I'm roaming outside. This post in combination with the post two posts up from this go together well. As people have mentioned, the physics tie with the frame rate, so you're doing something unsupported to start with, and, well, this is Gamebryo, and a game Bethesda released, so it's no surprise it's buggy and has weird performance issues.
|
|
#
?
Jan 5, 2016 14:29
|
|
|
Also, Fallout 4 doesn't actually support SLI, let alone Crossfire (how they managed to bungle THAT up is a complete mystery as well)
|
|
#
?
Jan 5, 2016 15:05
|
|
|
As someone that didn't see the conference, was there any discussion about the new consumer cards or just this AI car thing?
|
|
#
?
Jan 5, 2016 15:21
|
|
|
Just the GPU powered car AI stuff, nothing about consumer cards.
|
|
#
?
Jan 5, 2016 15:24
|
|
|
Anime Schoolgirl posted:Also, Fallout 4 doesn't actually support SLI, let alone Crossfire (how they managed to bungle THAT up is a complete mystery as well) The same way UE4 does?
|
|
#
?
Jan 5, 2016 15:26
|
|
|
I'd hope they at least changed the renderer for Fallout 4 significantly from Skyrim to warrant that, but that's kind of expecting too much from that company
|
|
#
?
Jan 5, 2016 15:36
|
|
|
Paul MaudDib posted:
The liquid block is due to needing to seal the thing up for use in a car. You really don't want to use air cooling when you're driving around Houston in 100 degrees / 80% humidity. The whole thing (including the Tegras) is 250 W TDP, so still less than a 390.  And yes, pretty solid performance for a couple of smaller chips and GDDR5. I'm still waiting to see GP100, though. 
|
|
#
?
Jan 5, 2016 15:41
|
|
|
Paul MaudDib posted:Nope, but they confirmed their collaboration with "Volvo". The AMD collab didnt pan out like they'd hoped 
Panty Saluter fucked around with this message at 16:12 on Jan 5, 2016 |
|
#
?
Jan 5, 2016 15:43
|
|
|
Subjunctive posted:The same way UE4 does? Yep, engine developers are relying more and more on temporal effects which completely poo poo the bed when run under AFR SLi/CF. The benefits are too great to pass them up for the sake of the 1% who run multiple GPUs. https://forums.unrealengine.com/showthread.php?51916-Unreal-Engine-4-6-Preview!&p=180535&viewfull=1#post180535
|
|
#
?
Jan 5, 2016 15:55
|
|
|
Isn't that moot, seeing as how we're headed towards a future where we treat two video cards as one very long pipeline anyways? (I forgot the word for where Card A handles certain elements before handing off to Card B for finishing and output.)
|
|
#
?
Jan 5, 2016 16:39
|
|
|
SwissArmyDruid posted:Isn't that moot, seeing as how we're headed towards a future where we treat two video cards as one very long pipeline anyways? (I forgot the word for where Card A handles certain elements before handing off to Card B for finishing and output.) wouldnt it be better to have one card being a coprocessor for the other, compositing directly into its memory? just using it as a long rear end pipeline seems like it would introduce a bunch of latency
|
|
#
?
Jan 5, 2016 16:42
|
|
|
|
| # ? Apr 29, 2024 12:09 |
|
|
I still don't really get the requirements for the Rift, as it's mostly an I/O device and the requirements will depend almost entirely on the application - I can run Quake 3 at 1920x1200 at over 90fps on Intel® HD Graphics 3000 from 2011, if that's what I wanted to do. That said, the news about Pascal and Polaris seriously piss me off, as I was planning to build a new machine just when the Rift ships, but it seems like none of the next-gen cards will be out yet 
|
|
#
?
Jan 5, 2016 17:11
|
|