|
repiv posted:yeah they dynamically clock the bus down on idle to save power and GPU-Z reflects the current state Here it is with the stress test running.  Since I connected the extra power plug, the card beats the 3DMark TimeSpy Extreme database average for my hardware, runs great. But the problem persists under lower loads at 1440p like TimeSpy. Ugh I hope I don't have to end up doing a clean Windows install!
|
#
?
May 1, 2023 14:41
|
|
|
|
| # ? May 30, 2024 12:20 |
|
|
is this on windows 10 or 11? 11 has some scheduling improvements for big.little CPUs like yours, it usually doesn't make a huge difference but who knows e: nevermind it shows win11 in your image
|
|
#
?
May 1, 2023 14:48
|
|
|
repiv posted:is this on windows 10 or 11? And all CPU benchmarks show its performing as expected. Help me repiv! I spent $1700 on this GPU!!!
|
|
#
?
May 1, 2023 14:51
|
|
|
Well aside from the bus speed thing my 4090 supposedly is worse than yours (no factory OC) and I didn't have any of these issues, I just got capped by my 5800X3D being uncoolable and eventually got slightly above average by undervolting my CPU. Have you tried any of the non-timespy tests? Do they all have this issue?
|
|
#
?
May 1, 2023 14:58
|
|
|
What is your PSU? Anything below 850w could result in unexpected/inconsistent performance
|
|
#
?
May 1, 2023 15:01
|
|
|
From looking around, this overclock.net forum thread has some people who improved their 3dmark performance on intel platforms by updating some intel chipset firmware thing: https://www.overclock.net/threads/3d-mark-terrible-4090-scores-after-upgrade.1801819/page-2 Not sure if it'll apply here, but maybe it's worth a shot?
|
|
#
?
May 1, 2023 15:10
|
|
|
power crystals posted:Well aside from the bus speed thing my 4090 supposedly is worse than yours (no factory OC) and I didn't have any of these issues, I just got capped by my 5800X3D being uncoolable and eventually got slightly above average by undervolting my CPU. Have you tried any of the non-timespy tests? Do they all have this issue? Yeah the other high end 4k DX12 tests run great and I get above average scores. But regular Firestrike and Timespy get low results. And some games that are in this same performance envelope also get lower than expected FPS. Zedsdeadbaby posted:What is your PSU? Anything below 850w could result in unexpected/inconsistent performance Corsair SF750. Other people online use it just fine. And it's doing great on the more intensive benchmarks and Cyberpunk path traced
|
|
#
?
May 1, 2023 15:13
|
|
|
Dr. Video Games 0031 posted:From looking around, this overclock.net forum thread has some people who improved their 3dmark performance on intel platforms by updating some intel chipset firmware thing: https://www.overclock.net/threads/3d-mark-terrible-4090-scores-after-upgrade.1801819/page-2 Just updated it. No change. 
|
|
#
?
May 1, 2023 15:26
|
|
|
Is the card's power consumption materially higher in timespy non-extreme? If so it might still be the PSU. Likewise, does hwinfo show the voltages dropping at all under load? None of this should matter but this is really weird.
|
|
#
?
May 1, 2023 15:38
|
|
|
power crystals posted:Is the card's power consumption materially higher in timespy non-extreme? If so it might still be the PSU. Likewise, does hwinfo show the voltages dropping at all under load? No, it comsumes around 250w on regular timespy, and over 450w with extreme This is what it looks like in Cyberpunk, compare to the screenshot I shared above 
Animal fucked around with this message at 15:50 on May 1, 2023 |
|
#
?
May 1, 2023 15:39
|
|
|
Looks like the $100 Steam gift card with the 4000 series purchase deal at Micro Center is gone now. Too bad, that was one of the best GPU bundles I've seen.
|
|
#
?
May 1, 2023 19:42
|
|
|
Animal posted:No, it comsumes around 250w on regular timespy, and over 450w with extreme At this point all I've got to suggest is do you have a friend nearby with a system you can try the card in?
|
|
#
?
May 1, 2023 19:49
|
|
|
power crystals posted:At this point all I've got to suggest is do you have a friend nearby with a system you can try the card in? I'd wine and dine them first, imo.
|
|
#
?
May 1, 2023 19:52
|
|
|
Nfcknblvbl posted:I'd wine and dine them first, imo. wrong brand for the fine wine
|
|
#
?
May 1, 2023 20:53
|
|
|
power crystals posted:At this point all I've got to suggest is do you have a friend nearby with a system you can try the card in? Yeah right now I’m DDU’ing the driver again and installing an old one from February. If that doesn’t work I’ll just  and hope the next driver update fixes it.
|
|
#
?
May 1, 2023 21:08
|
|
|
Is there anywhere that a framerate cap could be hidden, other than NVIDIA control panel and Rivatuner server? Does Windows 11 implement one? V-Sync is off.
|
|
#
?
May 1, 2023 21:19
|
|
|
I'd test that card in someone elses system, or even put a different card in yours and see if there still a framerate issue. While this seems like a Windows/Driver issue you really, REALLY want to make sure it's not a problem with the card.
|
|
#
?
May 1, 2023 22:10
|
|
|
Lockback posted:I'd test that card in someone elses system, or even put a different card in yours and see if there still a framerate issue. While this seems like a Windows/Driver issue you really, REALLY want to make sure it's not a problem with the card. I’ll try to find somewhere to take it. My old 3080 worked perfectly fine
|
|
#
?
May 1, 2023 22:13
|
|
|
Dr. Video Games 0031 posted:Just hook up one of the daisy chained power connectors to the adapter. That's perfectly fine to do. Daisy-chaining the power adapters is bad actually, you can overload the connectors and make the magic smoke come out, especially when you're dealing with a 4090. The 4090 draws 450W, which means you need 3x 150w connectors to guarantee you can drive it, and the 4th is for transients. Don't daisy chain your 4090. Each 8 pin connector is rated to pass 150w, and while the wires can handle more, even up to 300-400w, do you want that power going through the connector that is rated to handle 150w? Get a slight power hiccup and suddenly your GPU is smoking from the power pins.
|
|
#
?
May 1, 2023 23:58
|
|
|
Branch Nvidian posted:Reporting back in on my FFXIV GPU issue. Turns out the problem was being caused by LG's on-screen control application, which is used to update monitor firmware and such.... I cannot believe that was the issue, but it figures it would be some bizarre thing causing it. GPU utilization is still reported jumping around from the low 50% range up to 100% at times, but I kinda think that might be due to the chiplet design of RDNA3 and stuff not knowing how to accurately report it or something? I'm back to getting super high frame rates with middling GPU utilization, but as long as the game isn't stuttering constantly I don't care anymore. LMAO. RTX Off and RTX Really Off.
|
|
#
?
May 2, 2023 00:04
|
|
|
orange juche posted:Daisy-chaining the power adapters is bad actually, you can overload the connectors and make the magic smoke come out, especially when you're dealing with a 4090. The 4090 draws 450W, which means you need 3x 150w connectors to guarantee you can drive it, and the 4th is for transients. Don't daisy chain your 4090. The paranoia about daisy chaining is silly and unwarranted. The cables every reputable power supply company includes in the box are more than good enough to handle this. if you don't trust a power supply company to produce safe cables, then you shouldn't trust that power supply company period. Dr. Video Games 0031 fucked around with this message at 00:23 on May 2, 2023 |
|
#
?
May 2, 2023 00:16
|
|
|
Picked up the 7900xt. It works, but has fairly bad coil whine. I returned a 6950xt from the same aib which developed more of a scream than a whine--the 7900xt I could live with, but I likely will return it. I feel like it wouldn't be too hard to do some QC and prevent this type of thing, but instead board makers spend their energy trying to normalize it.
|
|
#
?
May 2, 2023 03:01
|
|
|
Dr. Video Games 0031 posted:The paranoia about daisy chaining is silly and unwarranted. The cables every reputable power supply company includes in the box are more than good enough to handle this. Paranoia? Not so much. But more copper is going to be better. Lower resistance. There are those that have eliminated coil whing by not daisy chaining. I don't see why you wouldn't use as many individual pcie power connections as necessary. You paid for the extra conductors after all, so why leave them in a box?
|
|
#
?
May 2, 2023 07:25
|
|
|
HalloKitty posted:Paranoia? Not so much. But more copper is going to be better. Lower resistance. There are those that have eliminated coil whing by not daisy chaining. The OP didn't have the extra cable and we were testing if having a fourth connector plugged into the squid adapter would help. It is best practice to use separate cables for each connector, certainly, but some people seem to confuse "best practice" with "doing literally anything else will kill your gpu" which is overly paranoid thinking in my opinion.
|
|
#
?
May 2, 2023 10:16
|
|
|
K8.0 posted:LMAO. RTX Off and RTX Really Off. and the game still looks good mostly from the art direction
|
|
#
?
May 2, 2023 11:35
|
|
|
I hadn’t looked up anything with GPUs in awhile but I was very surprised to see that Amazon has a bunch of 4070s in stock at MSRP now. Clearly the market has stabilized as that card just came out, right?
|
|
#
?
May 2, 2023 13:44
|
|
|
If by stabilised you mean priced a lot of people out of it and dissuaded people to upgrade then yes.
|
|
#
?
May 2, 2023 13:49
|
|
|
The pricing on the 4090/4080 is pretty lovely for sure, but the 4070 seems more reasonable
|
|
#
?
May 2, 2023 13:53
|
|
|
FlamingLiberal posted:The pricing on the 4090/4080 is pretty lovely for sure, but the 4070 seems more reasonable 20% more than the previous gen may seem more reasonable than the rest of the lineup, but the catch is that it's more of a 4060 in disguise. It only offers a very minor price-to-performance improvement over previous generations as a result. GPU market "stabilization" happened sometime around last summer, and cards have been in stock on store shelves ever since. Pricing remains the biggest issue. Dr. Video Games 0031 fucked around with this message at 14:07 on May 2, 2023 |
|
#
?
May 2, 2023 14:02
|
|
|
And the 4090, despite the ludicrous price tag, is actually priced reasonably. Just think of it as a Titan. The 4080 is the real tragedy: Great Card Horrific Price
|
|
#
?
May 2, 2023 16:16
|
|
|
Lockback posted:And the 4090, despite the ludicrous price tag, is actually priced reasonably. Just think of it as a Titan.
|
|
#
?
May 2, 2023 16:19
|
|
|
FlamingLiberal posted:The pricing on the 4090/4080 is pretty lovely for sure, but the 4070 seems more reasonable Not really. The 4090 is the only reasonably priced current nvidia card in a sense, but is what it is. AMD needs to get some more cards out
|
|
#
?
May 2, 2023 16:22
|
|
|
Josh Lyman posted:I keep seeing that the 4090 shouldn’t be considered a Titan since it’s not the full die. the titan x (pascal) wasn't the full die either, they double-dipped the whales with the titan xp later which was the full die likewise with the original kepler titan (cut-down) and titan black (full die)
|
|
#
?
May 2, 2023 16:23
|
|
|
Josh Lyman posted:I keep seeing that the 4090 shouldn’t be considered a Titan since it’s not the full die. Sure, but I'm speaking more from a consumer point of view, not a silicon point of view. Where it sits on the stack and the performance it gives for the price is more like a Titan (as compared to how the 3090 sat much more bunched up with the cheaper 3080. It's your Halo performance at Halo price. That is exactly what the Titan used to be for a consumer (again, the silicon is irrelevant. Price, Performance and Power are the outcomes). repiv posted:the titan x (pascal) wasn't the full die either, they double-dipped the whales with the titan xp later which was the full die Right, NVidia places things on the performance arc, not the silicon arc.
|
|
#
?
May 2, 2023 16:25
|
|
|
Josh Lyman posted:I keep seeing that the 4090 shouldn’t be considered a Titan since it’s not the full die. Even the original titan wasn’t the full die - it was the same cutdown as the (later) 780 non-Ti. Titan Z used two of them. Titan Black was when you got the full die and that had an even more astronomical price… it was as much or more than a titan Z iirc, for a single-chip card. Offhand I think it was $2500 in 2013 dollars or something silly like that. (Silly expensive halo cards have existed for a while, like the 8800 Ultra and Titan Black, and they rarely get you that much more than the sensible model. If nvidia drops a 4090 ti at some point (which is very possible imo) I fully expect that to be like $2500-3000. If they gave you Quadro drivers maybe it’d be worth it but the 3090 and 4090 are segmented to be halo gaming cards not prosumer.) I’d love to see AMD put out a double-memory 7900XTX 32gb frontier edition too… $749 is the historical price point if we are holding vendors to 2017 pricing, right? /s But seriously a 32gb 7900XTX at say $1299 or $1499 would be a pretty nice offering right now imo, and would let people get a foothold into ROCm with a quasi-supported config with big VRAM etc. Unfortunately they are more interested in talking trash than doing something about it, nobody is going to cut into their own segmentation profit margins. They just announced their new workstation card at a cool $4k instead, but reportedly you can get an RTX 6000 Ada with 48gb and CUDA for about the same price if you buy a pair… Paul MaudDib fucked around with this message at 17:08 on May 2, 2023 |
|
#
?
May 2, 2023 16:51
|
|
|
back then the gimmick was that the titans had unlocked FP64 performance (and ECC IIRC?) but they can't do that anymore since the bifurcated their architectures between gaming chips and datacenter chips now titans are just big gaming cards with more VRAM and maybe less crippled CAD performance if you're lucky
|
|
#
?
May 2, 2023 16:56
|
|
|
I've got another 12 or so hours of data collection to go before I can post a nice chart with 3 day averages for both scenarios, but I'm prepared to say this: If you're into hobbyist GPGPU stuff (where you're paying the power bill so efficiency matters), I think the only thing stopping the 4070 from being a world-beater is that it costs $599 instead of $449. Ada looks to be even more efficient than I had hoped: I'm getting right at 90% the work at 110W that the card is capable of at 200W. Though, sneak peek at the fuller story, it looks like the 4070's stock power limits are very generous for the number of cores on the chip. My card spent a lot of time only drawing 155-185W when under full load at the stock 200W limit.
|
|
#
?
May 2, 2023 20:50
|
|
|
yeah I don't get why people are hating the 4070 so much, it's a bit overpriced but if it were a hundred dollars cheaper it would be an absolute steal of a deal imo
|
|
#
?
May 2, 2023 20:57
|
|
|
Holy crap frame generation is absolute magic, especially in Microsoft Flight Simulator. It's gonna be a game changer for gaming laptops and portables.
|
|
#
?
May 2, 2023 20:59
|
|
|
|
| # ? May 30, 2024 12:20 |
|
|
Shipon posted:yeah I don't get why people are hating the 4070 so much, it's a bit overpriced but if it were a hundred dollars cheaper it would be an absolute steal of a deal imo 12 GB isn't anything to write home about at the price point. It's just the price that's the problem. mdxi posted:I think the only thing stopping the 4070 from being a world-beater is that it costs $599 instead of $449. Exactly
|
|
#
?
May 2, 2023 21:12
|
|