|
quite the gap https://twitter.com/XpeaGPU/status/1590221795462107136 on a similar page, tpu's gpu database would probably be the easiest thing to scrape if one wanted to do some napkin comparisons between cards over time, right?
|
#
?
Nov 9, 2022 17:12
|
|
|
|
| # ? May 30, 2024 09:41 |
|
|
Is the 4090Ti happening for sure? If we assume Ada tops out at the 4090 then everything kind of falls into the right place, historically. Only the last gen was weird really. So it's really only the 4080 12GB scam and the prices overall that are out of whack.
|
|
#
?
Nov 9, 2022 17:28
|
|
|
Who gives a gently caress about model names, the problem is the pricing, just like with the 2xxx launch before the Super refreshes. We're going to get a 4070 Super at some point that fixes this whole mess and gives people on 1080 Tis or 2070S something that's a clear upgrade without being an awful value.
|
|
#
?
Nov 9, 2022 17:30
|
|
|
we haven't heard anything definitive about a 4090ti but nvidia could certainly do it if they wanted to, the 4090 is enough of a cutdown that a 4090ti could be a good ~12% faster on paper
|
|
#
?
Nov 9, 2022 17:30
|
|
|
Twerk from Home posted:Who gives a gently caress about model names, the problem is the pricing, just like with the 2xxx launch before the Super refreshes. Otherwise of course it's the pricing, $1000 is ridiculous whether they call it a 4080 or 4070 or whatever.
|
|
#
?
Nov 9, 2022 20:03
|
|
|
mobby_6kl posted:Is the 4090Ti happening for sure? If we assume Ada tops out at the 4090 then everything kind of falls into the right place, historically. Only the last gen was weird really. The 4090 die is about 200mm^2 less than the reticle limit of TSMC's 4nm process. That is to say, they can make a bigger chip, but it would be a true monster to power, cool and pay for. So perhaps yes, they can make a 4090ti, but from what we have seen from AMD, Nvidia would be competing with themselves.
|
|
#
?
Nov 9, 2022 20:31
|
|
|
The 4090 does not use the whole die. The full AD102 has 18432 cuda cores, while the 4090 uses only 16384. So the 4090 Ti wouldn't be based on a new die closer to the reticle limit--it would just actually use the entire AD102 die. There were some reports that Nvidia has already started binning chips for this, but then there were reports that Nvidia canceled plans for a 4090 Ti, so who knows. Nvidia has a workstation GPU that uses more of the die, so the best samples are probably going there.
|
|
#
?
Nov 9, 2022 22:56
|
|
|
they didn't announce a price for the nearly fully unlocked AD102 workstation card, who's going to do the needful and buy one to benchmark cyberpunk speaking of workstation cards gamersnexus did an expensive teardown https://www.youtube.com/watch?v=4r_SwjYogQE repiv fucked around with this message at 23:09 on Nov 9, 2022 |
|
#
?
Nov 9, 2022 23:05
|
|
|
Cheap Trick posted:I'm looking to build a new PC next week and will probably pick up a 3080 10GB for $1200-ish. gently caress paying 3 grand for a video card last month 3080s were $200 AUD less than that fwiw Dr. Video Games 0031 posted:The 4090 does not use the whole die. The full AD102 has 18432 cuda cores, while the 4090 uses only 16384. So the 4090 Ti wouldn't be based on a new die closer to the reticle limit--it would just actually use the entire AD102 die. There were some reports that Nvidia has already started binning chips for this, but then there were reports that Nvidia canceled plans for a 4090 Ti, so who knows. Nvidia has a workstation GPU that uses more of the die, so the best samples are probably going there. weren't the cancellation reports about the crazy 48GB/800W Titan that had been rumoured, I think the 4090 Ti is still a possibility lih fucked around with this message at 23:29 on Nov 9, 2022 |
|
#
?
Nov 9, 2022 23:27
|
|
|
My Ampere A6000 performed between a 3080 and 3090, I'd expect the Ada 6000 to slot in similarly partially due to not having G6X VRAM
|
|
#
?
Nov 9, 2022 23:33
|
|
|
Dr. Video Games 0031 posted:The 4090 does not use the whole die. The full AD102 has 18432 cuda cores, while the 4090 uses only 16384. So the 4090 Ti wouldn't be based on a new die closer to the reticle limit--it would just actually use the entire AD102 die. There were some reports that Nvidia has already started binning chips for this, but then there were reports that Nvidia canceled plans for a 4090 Ti, so who knows. Nvidia has a workstation GPU that uses more of the die, so the best samples are probably going there. I didn't know that they had comparatively easy room to "scale" ad102. Thanks for explaining.
|
|
#
?
Nov 10, 2022 00:23
|
|
|
Thanks to the person who posted about the promotion that's happening with some RX 6000 cards. I didn't go for that, but I did find myself a 6600 XT which was nicely on sale. Now I can quit fretting about how long it's gonna take AMD and/or Nvidia to release midrange cards. I'm still gonna be reading about them when they do happen, tho.
|
|
#
?
Nov 10, 2022 01:13
|
|
|
GPU game bundles are a scam, I bought an R9 290 with free Star Citizen and the GPU is gone now but the game never released.
|
|
#
?
Nov 10, 2022 01:22
|
|
|
I think it's just Star Citizen that's a scam there
|
|
#
?
Nov 10, 2022 01:27
|
|
|
Nvidia/AMD got mad that people were selling the keys so now all you get is a locked-down copy of stuff like DLC for Doom: Eternal with your $1200 card I'm sure it was the publishers that were complaining but I never got why anyone gave a crap since they all still got paid.
|
|
#
?
Nov 10, 2022 01:29
|
|
|
Twerk from Home posted:GPU game bundles are a scam, I bought an R9 290 with free Star Citizen and the GPU is gone now but the game never released. Look, if you got a game 8 years ago, you would have likely got a few hours of enjoyment out of it. Star Citizen has given you over seventy thousand hours of anticipation. That kind of value has never existed in the gaming space.
|
|
#
?
Nov 10, 2022 01:30
|
|
|
https://twitter.com/VideoCardz/status/1590494651579912192
|
|
#
?
Nov 10, 2022 01:51
|
|
|
Cheap Trick posted:Up until mid-March, it would have been more like $2200 AUD I'm sure that's a good deal. Sorry, wasn't clear - I read your post as USD - "3080 10GB sweet deal at $1200 USD" was so true last year that I had some time dissonance for a sec.
|
|
#
?
Nov 10, 2022 02:16
|
|
|
$1200 AUD for a 3080 10GB is just a little over MSRP, it's not bad but there were much better deals a month ago https://www.pcbyte.com.au/p/pny-nvidia-geforce-rtx-3080-xlr8-revel-epic-x-rgb-10gb-lhr-video-card-vcg308010ltfxppb-325664 there's one for $1100 AUD here, that's not the lowest it's been but it's less than MSRP so still quite a good deal lih fucked around with this message at 02:23 on Nov 10, 2022 |
|
#
?
Nov 10, 2022 02:20
|
|
|
Tiny Timbs posted:Nvidia/AMD got mad that people were selling the keys so now all you get is a locked-down copy of stuff like DLC for Doom: Eternal with your $1200 card It was probably AMD/Nvidia pushing that methodology. By locking down the copies they can negotiate for a reduced tie-in valuation for a bunch of reasons. In other words it's cheaper to bundle a game when the utilization rate is lower through consumer unfriendly lockdown mechanisms. Doing it this way is likely only fractionally as expensive as offering a completely agnostic key like they did during the good ol' days. Taima fucked around with this message at 06:25 on Nov 10, 2022 |
|
#
?
Nov 10, 2022 06:22
|
|
|
lih posted:last month 3080s were $200 AUD less than that fwiw drat. Too bad I wasn't in a position to pick up the parts back then. Still have my eyes on the Asus TUF but maybe I'll get that PNY model instead. Thanks for the tip v1ld posted:I'm sure that's a good deal. Sorry, wasn't clear - I read your post as USD - "3080 10GB sweet deal at $1200 USD" was so true last year that I had some time dissonance for a sec. No worries, you're cool 
Cheap Trick fucked around with this message at 06:59 on Nov 10, 2022 |
|
#
?
Nov 10, 2022 06:54
|
|
|
Is there any go-to 10 ft HDMi 2.1 cable that won't break the bank?
|
|
#
?
Nov 10, 2022 10:07
|
|
|
My little system builder and Pc part shop here in Germany leaked their internal prices. The 7900xtx top customs will listed at 1500 Euro and go up to 1600 Euro lol The 4080 top customs like the Strix will be 1990 Euro. So no, RDNA3 is cheaper but it isn’t saving high end pc gaming gpu prices over here.
|
|
#
?
Nov 10, 2022 13:25
|
|
|
That’s the other issue, it’s all well and good having MSRP lower but stores still have dollar signs in their eyes and will continue to jack up prices and hand wave it away because something something inflation.
|
|
#
?
Nov 10, 2022 13:36
|
|
|
ChazTurbo posted:Is there any go-to 10 ft HDMi 2.1 cable that won't break the bank? I order kabeldirekt cables and two sided velcro straps from amazon.de. I live in EU though.
|
|
#
?
Nov 10, 2022 13:56
|
|
|
Mr.PayDay posted:My little system builder and Pc part shop here in Germany leaked their internal prices. one of the things i like about amd is that they have their own shop, just in case the stores here decide to get too stupid with their price gouging
|
|
#
?
Nov 10, 2022 13:58
|
|
|
kliras posted:damning with faint praise here, but it is genuinely nice that the first-party gpu's are available to a european customer base, unlike fe which seem to basically be na-only at this point? Nvidia do as well. And in the UK you can buy FE cards from there. They don’t have a retailer exclusivity agreement like they do in the US with Best Buy.
|
|
#
?
Nov 10, 2022 14:01
|
|
|
Got my 2060S in, and 3DVision is still cool dammit. Scratches an itch that VR doesn't, and it's much less hassle throwing on a pair of shutter glasses than putting on a headset. Wish it had taken off, but there are a lot of good games on the supported list. Had to roll back to Win10 1909 and run a hacked 452.06 driver, but community tools made that pretty easy.
|
|
#
?
Nov 10, 2022 14:56
|
|
|
whoops https://www.youtube.com/watch?v=CXBlkt7oZuw full video which looks quite good overall though: https://www.youtube.com/watch?v=FOP_p_7EDX8 background of the car sequence are a bit uneven, but it's still pretty amazing to have this versus the massive trade-offs of taa/fxaa/smaa. we'd be pretty spoiled if more games supported it
|
|
#
?
Nov 10, 2022 15:47
|
|
|
the one comparison of it on and off on the Series S I saw was impressive. 1.61 with FSR2 has no performance drops where 1.6 without FSR2 had drips into the mid to low 50s.
|
|
#
?
Nov 10, 2022 16:17
|
|
|
https://store.steampowered.com/news/app/1599660/view/3430081245013487928quote:Fixed Stutters Relating to Shader Compilation 
|
|
#
?
Nov 10, 2022 16:33
|
|
|
kliras posted:full video which looks quite good overall though:
|
|
#
?
Nov 10, 2022 17:41
|
|
|
I wonder what the hold up with DLSS3 is.
|
|
#
?
Nov 10, 2022 17:47
|
|
|
Combat Pretzel posted:The NPC bobbing her head around at 0:55, DLSS creates way less artifacting. Check out the trees around 0:42, yikes.
|
|
#
?
Nov 10, 2022 18:21
|
|
|
A friend of mine (he worked on Space/TimeWarp and general GPU insanity at Oculus when I was there) brought his company out of stealth today: Juice, which does IP-based transparent remoting of GPU resources at high speed. https://www.juicelabs.co/ Binaries available for Windows, Linux (Ubuntu), and Mac; works inside VMs; no client program modifications required. I’ve seen some demo videos of it before and it is pretty friggin’ nuts.
|
|
#
?
Nov 10, 2022 18:29
|
|
|
Not surprising. Cyberpunk is one of the visually noisiest games ever made. If you remember how it looks with TAA forced off, it's a horror show of single-pixel flickering. DLSS does incredible work taming that. I don't know that FSR will ever be that great at dealing with extreme levels of noise.
|
|
#
?
Nov 10, 2022 18:29
|
|
|
even DLSS isn't really designed to handle extreme noise on the input, they recommend running stochastic effects through some kind of denoiser before the DLSS pass cyberpunk just YOLOs it and leaves the TAA/TAAU to figure it out
|
|
#
?
Nov 10, 2022 18:35
|
|
|
AI-generation folks: Is there a Stable Diffusion but for speech synthesis? Asking for a friend that is mute and just dropped making youtube videos because of the cost to have other people make TTS on THEIR servers instead of using the perfectly-good hardware they have in their own computer.
|
|
#
?
Nov 10, 2022 23:19
|
|
|
SwissArmyDruid posted:AI-generation folks: Is there a Stable Diffusion but for speech synthesis? Asking for a friend that is mute and just dropped making youtube videos because of the cost to have other people make TTS on THEIR servers instead of using the perfectly-good hardware they have in their own computer. Also maybe: https://github.com/coqui-ai/TTS Coqui took a bit too much memory and was a bit too slow for me to like it when I was doing some TTS stuff. e: also it is not at all user friendly, it's dev-focused Malloc Voidstar fucked around with this message at 23:32 on Nov 10, 2022 |
|
#
?
Nov 10, 2022 23:27
|
|
|
|
| # ? May 30, 2024 09:41 |
|
|
Subjunctive posted:A friend of mine (he worked on Space/TimeWarp and general GPU insanity at Oculus when I was there) brought his company out of stealth today: Juice, which does IP-based transparent remoting of GPU resources at high speed. This looks incredibly promising, thanks for sharing! Would be a great way to have GPU sharing in a homelab situation without needing to either buy in to NVIDIA's vGpu ecosystem or wait for Intel to provide SRIOV support for Arc-based GPUs.
|
|
#
?
Nov 10, 2022 23:31
|
|
