|
Josh Lyman posted:This is the size of a small car 
|
#
?
Oct 6, 2022 07:55
|
|
|
|
| # ? May 29, 2024 07:15 |
|
|
Josh Lyman posted:This is the size of a small cat Where the heck are you getting your cats from, the Kruger national park?
|
|
#
?
Oct 6, 2022 07:59
|
|
|
Josh Lyman posted:This is the size of a small cat 
|
|
#
?
Oct 6, 2022 08:04
|
|
|
OEM video cards are usually ok, they're usually the actual Nvidia reference design (the FE is not the reference card, at least not since the 20-series). Here's Steve's review of a Dell 3090, which was shoved into an otherwise awful prebuilt PC: https://www.youtube.com/watch?v=_8k5PCmFm7c
|
|
#
?
Oct 6, 2022 08:13
|
|
|
repiv posted:incredible how hard intel fumbled the bag by not getting these out the door during the GPU shortage By all accounts the drivers were basically unusable back then. Even booting into windows was a crapshoot. If Intel puts in the time and money to fix them, after like 6 months at least (maybe a year), they'll be OK but as is its outright stupid to buy one. I guess if you're a GPU collector is the only 'use case' I can see for them right now. But those guys usually wait until it shows up on ebay for cheap if they can rather than pay full retail brand new.
|
|
#
?
Oct 6, 2022 08:49
|
|
|
I think one of those support posts that goes directly under the GPU isn't such a dumb idea these days.
|
|
#
?
Oct 6, 2022 09:22
|
|
|
infraboy posted:I think one of those support posts that goes directly under the GPU isn't such a dumb idea these days. Which is why the manufacturers are making RGB ones.
|
|
#
?
Oct 6, 2022 09:28
|
|
|
Somebody on the chinese forum Chiphell supposedly has a 4080 16GB: https://videocardz.com/newz/alleged-nvidia-geforce-rtx-4080-16gb-3dmark-benchmarks-have-been-leaked Take these numbers with a grain of salt. If this is real, then my main takeaway is that ray tracing performance has not improved if developers do not use those special features Nvidia promoted during the announcement (Shader Execution Reordering and the geometry/mesh stuff with fancy names). So if you care about RT performance, you are really betting on the promise of future performance, one that can't be tested at launch unless reviewers have beta access to the new cyberpunk RT mode.
|
|
#
?
Oct 6, 2022 10:44
|
|
|
infraboy posted:I think one of those support posts that goes directly under the GPU isn't such a dumb idea these days. E-GTX form factor soon
|
|
#
?
Oct 6, 2022 12:05
|
|
|
Funnily enough, the 4090 FE will still fit in some ITX cases like the Lian Li Q58 and NR200, so I look forward to seeing some horrible monstrosities soon
|
|
#
?
Oct 6, 2022 13:35
|
|
|
Palladium posted:E-GTX form factor soon Really does seem like if it's gonna be that big it should just go in its own box.
|
|
#
?
Oct 6, 2022 13:47
|
|
|
TheScott2K posted:Really does seem like if it's gonna be that big it should just go in its own box. I think a better solution would be standardized screw mounting points and the ability to screw it into your chassis, with the assumption that it needs clearance for a big tower air cooler on it. Or bring back reasonably priced factory liquid cooled GPUs like the Fury X.
|
|
#
?
Oct 6, 2022 13:49
|
|
|
I mean those aren't exactly small for SFF and the 3090 fit, the two aren't that different so it's not that surprising.
|
|
#
?
Oct 6, 2022 13:51
|
|
|
change my name posted:Funnily enough, the 4090 FE will still fit in some ITX cases like the Lian Li Q58 and NR200, so I look forward to seeing some horrible monstrosities soon It seems like the 4090 FE is 6mm too long according to the official clearances. Maybe you can massage the extra 6mm and squeeze it in there anyway? You may also be able modify the case by cutting out an opening in the SSD mounting locations.
|
|
#
?
Oct 6, 2022 14:05
|
|
|
Waiting to see the unboxing of the AIOs - the contrast between them and the fan based cards is going to be nuts
|
|
#
?
Oct 6, 2022 14:11
|
|
|
I know for a fact that none of these 40 series cards will fit in my current Phanteks case I’m clearly going to need to buy a whole new case for when the 50 series eventually comes out
|
|
#
?
Oct 6, 2022 14:38
|
|
|
Dr. Video Games 0031 posted:It seems like the 4090 FE is 6mm too long according to the official clearances. Maybe you can massage the extra 6mm and squeeze it in there anyway? You may also be able modify the case by cutting out an opening in the SSD mounting locations. For which? It's actually shorter than the 3090 at 304 mm, just thicker. Max GPU length for the NR200 is 336mm, max for the Q58 is 320mm. The heat and finding a 1000-watt SFX (or SFX-L) PSU is the bigger problem change my name fucked around with this message at 14:41 on Oct 6, 2022 |
|
#
?
Oct 6, 2022 14:38
|
|
|
change my name posted:For which? It's actually shorter than the 3090 at 304 mm, just thicker. Max GPU length for the NR200 is 336mm, max for the Q58 is 320mm. The heat and finding a 1000-watt SFX (or SFX-L) PSU is the bigger problem Oh, TechPowerUp lied to me by saying it was 336mm. Never mind, then.
|
|
#
?
Oct 6, 2022 14:46
|
|
|
change my name posted:For which? It's actually shorter than the 3090 at 304 mm, just thicker. Max GPU length for the NR200 is 336mm, max for the Q58 is 320mm. The heat and finding a 1000-watt SFX (or SFX-L) PSU is the bigger problem coolermaster announced a 1300w SFX supply https://www.tomshardware.com/news/cooler-master-launches-1300w-sfx-psu-with-pcie5-power-connector probably sounds like a jet engine at full load though
|
|
#
?
Oct 6, 2022 14:52
|
|
|
An 850W PSU will probably be fine if you have a ~120W CPU, to be honest. A higher-power CPU might get you into trouble, depending on how much other random poo poo you have in your PC. Nvidia claims they've solved the transient problems on the GPU side. This apparently happened with the 3090 Ti and a report from Igor's Lab confirmed this is the case. If this holds true, then you won't need to over-provision on the PSU and all you'll really need to worry about is max sustained load. Dr. Video Games 0031 fucked around with this message at 15:01 on Oct 6, 2022 |
|
#
?
Oct 6, 2022 14:56
|
|
|
Palladium posted:E-GTX form factor soon Uh, E-RTX? Prebuilt vendors are going to have a heck of a time shipping systems with these new cards, huh? How many motherboards are going to show up with the metal-reinforced PCIe slot ripped clean out of the motherboard?
|
|
#
?
Oct 6, 2022 15:16
|
|
|
GPUs will have to start coming in separate packaging with a "Some Assembly Required" disclaimer. e: also one of these stickers on the box 
Enos Cabell fucked around with this message at 15:20 on Oct 6, 2022 |
|
#
?
Oct 6, 2022 15:18
|
|
|
PBCrunch posted:Uh, E-RTX? I am pretty sure they are going to have to ship them in another box with instructions. There will be too many RMAs if they are shipped with the 4000 series cards put into the system and ship it out.
|
|
#
?
Oct 6, 2022 16:03
|
|
|
Probably also going to be a lot of RMA’s when someone who knows dick all about putting a PC together manages to gently caress up bad enough that they damage the card or the board etc. It’s not rocket science but any time you put any kind of assembly in customer hands you’re going to see a certain percentage that do some wild poo poo.
|
|
#
?
Oct 6, 2022 16:06
|
|
|
I'm pretty sure I've seen prebuilts where they use those inflated airbag things to fill the space inside the case so that there's no chance of things rattling around or breaking-off against their tension points while in handling and transit part of the raison d'etre for AIO CPU cooling is also that a pump-and-radiator set-up is less fragile in transit than having a big tower cooler hanging off the backplate potentially exerting stress on the CPU/motherboard. Perhaps more OEMs will switch to water-cooled GPUs for that reason I think there have also been untapped cooler designs that are designed to screw-in to the front side of the case it's not impossible for OEMs/builders to make these things shippable, if they really try, but they do have to be willing to put in the effort, and I doubt they're going to take the path of asking the user to assemble it themselves, because that undermines the whole concept
|
|
#
?
Oct 6, 2022 16:14
|
|
|
Shumagorath posted:Also a fully modular PSU could conceivably use a single new cable, right? No, the ATX 3.0 thing includes some USB-style negotiation bullshit to set OCP limits, so the PSU can understand whether that’s a 1030 that shorted out or a 4090 in normal operation. You could conceivably have a custom cable with a controller that talks to the card and just makes some bullshit and lets the PSU have its normal hardware limits I guess, but, functionally that’s the same as a passive adapter. I guess I’m kinda answering my own question here, it kinda seems like this will be a “nice to have” thing that lets the PSU be smarter about failure cases but isn’t something that GPU companies will probably be willing to cut into their sales over if you need a passive adapter.
|
|
#
?
Oct 6, 2022 17:26
|
|
|
Dr. Video Games 0031 posted:That's not to say that AMD's approach has no downsides. Spider-Man Remastered is a heavily bandwidth-sensitive game, and RDNA2 cards perform quite a bit worse than the 40-series cards. Maybe there's some other architectural reason for that, but it sure looks like a bandwidth issue. RDNA2 just has bad RT perf in general, I think it’s just a lower priority for AMD at least as of last gen, especially with consoles being in the drivers seat, but AMD could have designed it more scalable if they wanted. Like they could have had consoles use a have double-pumped or quad-pumped version that executes more slowly, or design faster units but only put them in half the TMUs on the consoles. AMD doesn’t seem to prioritize it right now, kinda like tessellation back in the GCN1 days. (RT is around 4% of the NVIDIA die as of Turing, and rdna2 is supposedly around 2%, and AMD’s synthetic/path-traced RT performance is around 50% of Turing relative to raster, so the numbers kinda line up there.) But anyway RT is kinda hard on cache I think, there’s not really a way to predict what material/texture you’re going to hit after a bounce, and random access is tough on caches, it’s the same reason why ethereum mining was uncacheable…
|
|
#
?
Oct 6, 2022 17:33
|
|
|
one thing that occurred to me w.r.t. RDNA2 raytacing is that since the traversal is done in software, they could potentially backport the new micromesh and opacity map features that nvidia introduced with ada the double edged sword of nvidia pushing traversal into dedicated hardware is they can't add any new functionality there without new hardware
|
|
#
?
Oct 6, 2022 17:43
|
|
|
repiv posted:the double edged sword of nvidia pushing traversal into dedicated hardware is they can't add any new functionality there without new hardware I think that's only upside as far as nvidia cares
|
|
#
?
Oct 6, 2022 17:52
|
|
|
upside for nvidia: people have to buy new cards to get the new features downside for nvidia: engines aren't going to actually adopt these features that don't work on 99% of RTX GPUs in the wild
|
|
#
?
Oct 6, 2022 17:55
|
|
|
That makes sense, the 600w rumor hung around like a bad smell for like a year. It was only recently they apparently went back on it
|
|
#
?
Oct 6, 2022 18:05
|
|
|
Paul MaudDib posted:No, the ATX 3.0 thing includes some USB-style negotiation bullshit to set OCP limits, so the PSU can understand whether that’s a 1030 that shorted out or a 4090 in normal operation. It's a lot more simple than the USB power delivery thing; two pins are used as open/ground sense lines by the PSU to determine whether the card can use up to 150, 300, 450, or 600 watts. Naturally I suspect cheater cables will just short both sense pins to ground since that means "yep go ham 600 watts". https://edc.intel.com/content/www/u...sign-guide/2.0/
|
|
#
?
Oct 6, 2022 18:11
|
|
|
MarcusSA posted:My gaming rig is on the second floor what can I do to reinforce the floor so it doesn’t collapse if I get one of these. Point airflow down, it should generate the same lift as hovercraft. New nvidia-branded mini-O11D case from Derbauer? Nice.
|
|
#
?
Oct 6, 2022 18:20
|
|
|
power crystals posted:Igors Lab thinks the ludicrous coolers are because the 4000 series cards were meant to be on an older node originally, resulting in the 4090 coolers being built for a 600W version of the card that isn't the actual shipping version. There is no way that is true. Silicon design and fabrication technology are too closely linked for a "last minute switcheroo" on process tech.
|
|
#
?
Oct 6, 2022 18:25
|
|
|
But it's already been shown that nvidia don't really respect third party vendors
|
|
#
?
Oct 6, 2022 18:28
|
|
|
designing my next house to incorporate video cards for heating in the winter and appropriate measures for cooling them in spring/summer
|
|
#
?
Oct 6, 2022 18:32
|
|
|
PBCrunch posted:There is no way that is true. Silicon design and fabrication technology are too closely linked for a "last minute switcheroo" on process tech. It is possible and entirely on-brand for nvidia that they realized the 600W/8nm version was going to be a problem early in the dev cycle and pivoted to 4nm to try to compensate, and then just... didn't tell their partners, at least for too long for them to swap coolers. Not that that's logical at all, but when you're riding the high of infinite crypto mining demand as they would have been at the time, why not blow a stupid amount of money on double-R&D?
|
|
#
?
Oct 6, 2022 18:39
|
|
|
Lol if evga jumping off this gen turns out to have been a great idea
|
|
#
?
Oct 6, 2022 18:41
|
|
|
|
| # ? May 29, 2024 07:15 |
|
|
power crystals posted:It is possible and entirely on-brand for nvidia that they realized the 600W/8nm version was going to be a problem early in the dev cycle and pivoted to 4nm to try to compensate, and then just... didn't tell their partners, at least for too long for them to swap coolers. Not that that's logical at all, but when you're riding the high of infinite crypto mining demand as they would have been at the time, why not blow a stupid amount of money on double-R&D? making a 4090 on Samsung 8nm transistors would be around 2x the Samsung 8nm reticle limit, and that may be implicitly assuming some improvements in cache density too (cache scales differently from logic). There was never any possibility that Ada was on 8nm, that is STDH.txt let alone re-designing and re-validating a whole product in the last 3 months. Maybe Samsung 5nm would have been a possibility, that’s around tsmc 7nm density, but Igor specifically says Samsung 8nm and lol no I’ve previously said “maybe they went from N5P to 4N at some point” but that’s not gonna get you a 33% reduction in power most likely. Remember, those early rumors were TGP, not TBP, so 600W TGP -> 650-675W TBP and 900W TGP->1kw TBP ish. You have like a 33% reduction in power or more to account for. also apparently TSMC 5nm clocks like 50% higher as well lol. I still think that’s the real angle here, someone saw a thermal sample cranked to max and thought those were going to be the reference TGP. Or maybe that’s really what NVIDIA told partners to design their coolers for? Seems weird and I doubt that could have been in any context where it wasn’t made clear that was a max OC and not typical power. It’s hard to say whether that was really something told to partners or not, given how much of this leaking business has turned into a giant ourobos of Twitter circlejerking… Igor may not actually know that was told to partners and may just be taking Kopite on faith that’s correct. Paul MaudDib fucked around with this message at 19:04 on Oct 6, 2022 |
|
#
?
Oct 6, 2022 18:50
|
|