|
If the microthrottling thing holds true and is the reason for the apparent IPC regression, we might end up seeing an AMD GPU that performs better when underclocked 
|
#
?
Jul 5, 2017 18:39
|
|
|
|
| # ? May 31, 2024 18:44 |
|
|
I have a Samsung 4K TV that I'm using as a monitor. The TV will do 4k 60 Hz at 4:4:4 color using either YCbCr or RGB Full through the HDMI 2.0 output on a Radeon RX 480. Does it make a difference which one I choose? sincx fucked around with this message at 19:01 on Jul 5, 2017 |
|
#
?
Jul 5, 2017 18:58
|
|
|
SlayVus posted:So if we underclock Vega for reduced power consumption, we should we engagement of all cores all the time. Which should in theory provide a more stable environment for gaming and other junk. One of the reviews mentioned that if you touch any voltage settings at all, the HBM modules will go to half speed permanently until you reset everything and reboot.
|
|
#
?
Jul 5, 2017 19:01
|
|
|
sauer kraut posted:One of the reviews mentioned that if you touch any voltage settings at all, the HBM modules will go to half speed permanently until you reset everything and reboot. An overclockers dream
|
|
#
?
Jul 5, 2017 19:07
|
|
|
Jim Keller taught the Zen team how to suck the powers out of the Radeon team like some kind of engineering vampire
|
|
#
?
Jul 5, 2017 21:07
|
|
|
Kazinsal posted:If the microthrottling thing holds true and is the reason for the apparent IPC regression, we might end up seeing an AMD GPU that performs better when underclocked my old reference r9 290x had such an inadequate cooler that running it at the stock speed with some games would cause it to periodically overheat and throttle, turning them into a stuttery mess. performing better when underclocked is practically an amd tradition at this point; it's almost endearing
|
|
#
?
Jul 5, 2017 21:20
|
|
|
Kazinsal posted:Jim Keller taught the Zen team how to suck the powers out of the Radeon team like some kind of engineering vampire One of my portfolio companies has a couple of ex-AMD engineers (and Jim Keller as an investor/advisor). They're always very quick to point out that they're from the CPU side of the house, and of the Zen era.
|
|
#
?
Jul 5, 2017 21:27
|
|
|
I wonder if a GTX 1060 with no display outputs could be used for Steam streaming.
|
|
#
?
Jul 5, 2017 21:43
|
|
|
Knowing Nvidia's iron-fisted grip against blurring market segmentation lines? Probably Error 43.
|
|
#
?
Jul 5, 2017 21:45
|
|
|
PBCrunch posted:I wonder if a GTX 1060 with no display outputs could be used for Steam streaming. Many games don't stream unless you plug in a monitor or dummy-dongle so I doubt it.
|
|
#
?
Jul 5, 2017 22:10
|
|
|
eGPU that's always routed back to the laptop screen.
|
|
#
?
Jul 5, 2017 22:12
|
|
|
eames posted:Many games don't stream unless you plug in a monitor or dummy-dongle so I doubt it. This is correct. There is no "default display" in Windows, if you don't have a monitor or a dummy-dongle then the display is actually not rendered at all. Steam would have to implement a "virtual monitor" that captures the output and streams it. What they actually have uses basically the same mechanism as Shadowplay/OBS - it framegrabs the output being rendered to another display. Remote Desktop does provide some of this functionality though (you can RD into a headless machine, you cannot VNC into a headless machine). Maybe there would be a way to piggyback on that, but I doubt RD is optimized around gaming (it may actually even be a software renderer, I'm not sure). NewFatMike posted:eGPU that's always routed back to the laptop screen. Nope, the laptop screen counts as an output display here, even if magic is happening in the background to get the frames to the screen. Paul MaudDib fucked around with this message at 22:39 on Jul 5, 2017 |
|
#
?
Jul 5, 2017 22:28
|
|
|
I'm sure people would come up with some cool software solutions if it weren't for the fact that these are only going to be like 20bux cheaper. Maybe when the secondhand ones hit ebay for like $30.
|
|
#
?
Jul 5, 2017 22:33
|
|
|
Paul MaudDib posted:This is correct. There is no "default display" in Windows, if you don't have a monitor or a dummy-dongle then the display is actually not rendered at all. Ah I was just kind of thinking out loud about other uses for the no output cards.
|
|
#
?
Jul 5, 2017 22:37
|
|
|
NewFatMike posted:Ah I was just kind of thinking out loud about other uses for the no output cards. Oh, I see what you're saying. That's a good point, you actually could use an eGPU+miner card with a laptop. The downside is that streaming the display back to the laptop kills performance - PCIe 3.0x4 is already bottlenecking just for the drawcalls, streaming uncompressed bitmaps back too is far too much bandwidth. Maybe when Display Stream Compression comes out, that might be enough to squish it down to a reasonable extent.
|
|
#
?
Jul 5, 2017 22:42
|
|
|
Paul MaudDib posted:Oh, I see what you're saying. That's a good point, you actually could use an eGPU+miner card with a laptop. The downside is that streaming the display back to the laptop kills performance - PCIe 3.0x4 is already bottlenecking just for the drawcalls, streaming uncompressed bitmaps back too is far too much bandwidth. A 1060 or 570 or whatever, even with the link speed gimped and streaming output back (on TB3 at least), is still a significant boost for something like an ultrabook with no dedicated GPU. If we actually see TB3 eGPU prices come down to earth in the near future it could be a cool combo.
|
|
#
?
Jul 5, 2017 22:44
|
|
|
If they could ever do like an 8x connection, you do a 4x directional which shouldn't cause any problems. But that would be up to Intel making a new thunderbolt connection that could sustain that kind of throughput, making it come off the CPU directly, and manufacturers adopting it.
|
|
#
?
Jul 5, 2017 23:10
|
|
|
Dell was briefly selling the PNY 3-fan 1080 Ti for $634 today, which I believe would be a new low for a 1080 Ti, but its back up to $699 on their page now.
|
|
#
?
Jul 5, 2017 23:28
|
|
|
It's very disappointing to see AMD's GPU division basically collapse. I hope Nvidia doesn't price gouge to hard with Volta. I don't think I can do another $80 hike like the 1070 was over the 970. edit: Oh and of course the miners jacking prices up on nearly everything as well.
|
|
#
?
Jul 6, 2017 03:01
|
|
|
OhFunny posted:It's very disappointing to see AMD's GPU division basically collapse. Remember though the 970 was a $70 price decrease (for no real reason) so really we're in for a $160 increase now
|
|
#
?
Jul 6, 2017 04:24
|
|
|
1gnoirents posted:Remember though the 970 was a $70 price decrease (for no real reason) so really we're in for a $160 increase now
|
|
#
?
Jul 6, 2017 04:34
|
|
|
I'll pay it gladly if Nvidia keeps doing these great energy efficiency gains every generation. They're doing a better job than Intel at this point.
|
|
#
?
Jul 6, 2017 06:27
|
|
|
repiv posted:That's more evidence for the micro-throttling theory then. If Vega is disabling shader cores to moderate its power consumption without affecting geometry throughput then you'd expect a drop in IPC as the target clock increases. https://twitter.com/GamersNexus/status/882769416659881984
|
|
#
?
Jul 6, 2017 06:54
|
|
|
One way or another, this is going to be good.
|
|
#
?
Jul 6, 2017 11:43
|
|
|
I saw a thread on overclockers.com forums where somebody bought a Vega FE to use as a VM passthrough card. On a Promox linux home server... passing through to Ubuntu  https://www.overclockers.com/forums/showthread.php/783058-Vega-Frontier-in-my-server
|
|
#
?
Jul 6, 2017 13:32
|
|
|
The Lord Bude posted:One way or another, this is going to be good. "We put it on a X299 board and then all the VRMs exploded into flames at once. This combo is a very hot commodity right now."
|
|
#
?
Jul 6, 2017 13:46
|
|
|
I'm just glad amd gpu's suck at the moment so my conscience can keep telling itself my overpriced gsync monitor was justified.
|
|
#
?
Jul 6, 2017 15:19
|
|
|
Otakufag posted:I'm just glad amd gpu's suck at the moment so my conscience can keep telling itself my overpriced gsync monitor was justified. Same except I paid $270 for my XB270HU.  Really though, the "lock-in" on GSync is no big deal right now given that NVIDIA has total dominance of the market. AMD hasn't managed to compete with anything faster than the 1060 and Vega probably won't compete with anything faster than an 1160. Right now NVIDIA lock-in is solid gold handcuffs.
|
|
#
?
Jul 6, 2017 16:03
|
|
|
When new cards are put onto Newegg, what time of day is that generally? I'm specifically looking at getting the Aorus Gaming Box if that helps.
|
|
#
?
Jul 6, 2017 16:51
|
|
|
AT has some info on the mining graphics card craze, and why the board partners are doing it http://www.anandtech.com/show/11607/cryptocurrency-mining-cards-update-zotac-manli-biostar-products-formally-confirmed
|
|
#
?
Jul 6, 2017 17:24
|
|
|
AMD putting the brakes on the hype train (Rys from Beyond3D, he's their EU marketing manager or something similar):quote:The "Fiji fallback driver" or "Fiji drivers" meme needs to stop. That's not how it works or should be described. Otherwise you should start calling Volta drivers Pascal drivers Maxwell drivers. There is obviously commonality -- that's just how software engineering needs to work for a GPU -- but calling it a Fiji driver or Fiji fallback is wrong. quote:About the driver comment: it's normal and completely expected for there to be common code in a GPU driver that applies to some or all of the GPUs a driver supports, alongside the specifics for the GPU being driven. That's hopefully just a given. So it was just a guiding hand to not conflate any commonality with it running the driver for a different ASIC, and then reading things into that. 
Paul MaudDib fucked around with this message at 17:31 on Jul 6, 2017 |
|
#
?
Jul 6, 2017 17:27
|
|
|
The Vega FE driver is also reporting support for numerous DX12 features that weren't supported by Fiji or Polaris, which just debunks the "Fiji mode" theory even more.
|
|
#
?
Jul 6, 2017 17:50
|
|
|
i guess everything is fine then and its a steal at $1000
|
|
#
?
Jul 6, 2017 17:57
|
|
|
repiv posted:The Vega FE driver is also reporting support for numerous DX12 features that weren't supported by Fiji or Polaris, which just debunks the "Fiji mode" theory even more. This doesn't mean anything if it performs like a dumpster fire
|
|
#
?
Jul 6, 2017 18:46
|
|
|
Fauxtool posted:i guess everything is fine then and its a steal at $1000 In the grim, dark future of next month, 1070s, 1060s, 580s will also cost $1000, so Vega's price / perf will be fine. 1080s will remain a value option at around $600 because it doesn't mine well.
|
|
#
?
Jul 6, 2017 18:48
|
|
|
I prefer the fiji mode theory because that means its okay hardware being held back and not crap hardware that wont get better
|
|
#
?
Jul 6, 2017 18:48
|
|
|
Twerk from Home posted:In the grim, dark future of next month, 1070s, 1060s, 580s will also cost $1000, so Vega's price / perf will be fine. Think I can sell my RX480 for enough to buy a 1080?
|
|
#
?
Jul 6, 2017 18:51
|
|
|
Risky Bisquick posted:This doesn't mean anything if it performs like a dumpster fire The implication was that these can't be drivers from like 8 months ago if they include support for features that were only released like 3 months ago. To me, "old drivers" has always meant "drivers released today were feature-frozen like a month ago for stability testing", not "you could have had a baby in the timeframe since we updated the drivers for the launch of our flagship product". It doesn't make any sense at all that a product like Vega would release with ancient drivers, but the performance of Vega doesn't make sense either and since AMD's not talking, people are doing code archaeology to try and figure out what could possibly be wrong. (the "microthrottle" theory seems to make sense to me both in terms of the behavior and the engineering logic behind it, but I certainly don't regard it as proven by any means and would love if AMD would explain what the gently caress we're looking at here)
|
|
#
?
Jul 6, 2017 19:03
|
|
|
if they were doing anything they could be proud of you can be sure they would be telling us.
|
|
#
?
Jul 6, 2017 19:04
|
|
|
|
| # ? May 31, 2024 18:44 |
|
|
Risky Bisquick posted:This doesn't mean anything if it performs like a dumpster fire That's not the point, only that the drivers are different. Of course the whole thing is a mess.
|
|
#
?
Jul 6, 2017 19:05
|
|
