|
https://twitter.com/VideoCardz/status/1635948981174796291?s=20 Not too shabby
|
#
?
Mar 15, 2023 13:19
|
|
|
|
| # ? Jun 3, 2024 17:36 |
|
|
that's actually really loving good if true?
|
|
#
?
Mar 15, 2023 13:23
|
|
|
oh man if they start putting those on desktop APUs yowza *wolf-whistle* BOI-OI-OING!!!
|
|
#
?
Mar 15, 2023 13:31
|
|
|
Truga posted:are they getting any extra perf from that tho? On identical CPUs it's about a 5% score uplift over the next highest score and allows the core clock to hit around 3.3GHz in TSE and Speedway. Of course, the top 100 scores are almost always some exotic cooling or other fuckery, so I compared to my scores for a closer-to-the-mean comparison. I'm running a slightly different CPU (5950X vs his 5900X), but our systems and cooling are otherwise near identical custom water cooled MBA cards and his GPU scores are around 25% higher than mine which is equivalent to a ~20FPS increase.
|
|
#
?
Mar 15, 2023 13:36
|
|
|
Isn't a 1650 basically a 1060, a midrange card from seven years ago? I dunno how that compares to the current igpus but doesn't seem that crazy.
|
|
#
?
Mar 15, 2023 13:57
|
|
|
mobby_6kl posted:Isn't a 1650 basically a 1060, a midrange card from seven years ago? I dunno how that compares to the current igpus but doesn't seem that crazy. IIRC, that's a 100% improvement over previous iGPUs.
|
|
#
?
Mar 15, 2023 14:03
|
|
|
The current fastest iGPU for x86 PCs is the RDNA2-based Radeon 680M that's been appearing in some mobile chips for a year now. In the 65W chips, it's maybe between the 1050 and 1050 Ti in performance. This new RDNA3-based one will be maybe 25 - 30% faster, which is about what you'd expect going from RDNA2 to 3 at the same TDP. The 1650 it's being compared to is the mobile Max Q variant, so the 1060 is going to be quite a bit faster still. Maybe AMD's iGPUs will reach that level of performance with RDNA4 in a couple years. Dr. Video Games 0031 fucked around with this message at 14:15 on Mar 15, 2023 |
|
#
?
Mar 15, 2023 14:09
|
|
|
I comfortably ran Half Life Alyx on a GTX 980, which this seems to be faster than. A mobile SoC being able to run that game has some interesting implications for Valve's future in hardware, I guess it remains to be seen how much power it sucks down to hit those performance numbers though.
|
|
#
?
Mar 15, 2023 14:15
|
|
|
SCheeseman posted:I comfortably ran Half Life Alyx on a GTX 980, which this seems to be faster than. A mobile SoC being able to run that game has some interesting implications for Valve's future in hardware, I guess it remains to be seen how much power it sucks down to hit those performance numbers though. I sincerely doubt it will be. Honestly, this whole comparison is questionable because it's just a Geekbench OpenCL benchmark. The regular desktop 1650 scores higher than the 980 Ti in this test, but the 980 Ti is probably 50 - 60% faster in actual gaming performance. These scores are meaningless.
|
|
#
?
Mar 15, 2023 14:19
|
|
|
Dr. Video Games 0031 posted:The current fastest iGPU for x86 PCs is the RDNA2-based Radeon 680M that's been appearing in some mobile chips for a year now. In the 65W chips, it's maybe between the 1050 and 1050 Ti in performance. This new RDNA3-based one will be maybe 25 - 30% faster, which is about what you'd expect going from RDNA2 to 3 at the same TDP. The biggest applications are going to be cheap 1080p laptops, ultra-small desktops that you might want to do some gaming on, and Ayaneo etc gaming handhelds. They can keep pushing that they're faster than the Steam Deck by X%
|
|
#
?
Mar 15, 2023 14:20
|
|
|
mobby_6kl posted:Isn't a 1650 basically a 1060, a midrange card from seven years ago? I dunno how that compares to the current igpus but doesn't seem that crazy. difficult to get comparisons among laptop chips, but, for example: - Hardware Unboxed got 144 average FPS out of the GTX 1060 laptop GPU when running Tom Clancy's Rainbox Six Siege at 1080p Vulkan Medium - elsewhere in the review I'm referencing, we see the GTX 1650 Ti laptop competing with the GTX 1060 laptop; and we can expect that the GTX 1650 [non-Ti] would then be even a little bit slower than that - now, if we move over to Gamers Nexus, they were able to get, when running Siege at 1080p Very High: - 59.9 FPS from the Radeon 8 iGPU on the Ryzen 7 5700G - 49.5 FPS from the Intel Iris Xe G7 (the iGPU on a laptop SOC) - 45.4 FPS from the Radeon 8 on the Ryzen 9 5900HX (laptop SOC) - 39.6 FPS from a GT 1030 - 30.7 FPS from the UHD 770 graphics on an i9-12900K now, this is of course not at all an apples-to-apples comparison: we can't even see the GTX 1650 laptop directly, and HWUB and GN are using different quality settings across their respective benches of Siege, but given that the best desktop iGPU is performing at less than half of the kind of performance being alleged (i.e., better than a GTX 1650 [non-Ti]), that's still a big leap forwards.
|
|
#
?
Mar 15, 2023 14:29
|
|
|
Shipon posted:eh, a top end GPU might be a bad per-dollar value but it's still better than the NaN value a console offers (what games lol) 
|
|
#
?
Mar 15, 2023 14:35
|
|
|
gradenko_2000 posted:now, this is of course not at all an apples-to-apples comparison: we can't even see the GTX 1650 laptop directly, and HWUB and GN are using different quality settings across their respective benches of Siege, but given that the best desktop iGPU is performing at less than half of the kind of performance being alleged (i.e., better than a GTX 1650 [non-Ti]), that's still a big leap forwards. A big leap forwards from what? This is not succeeding the Radeon 8 in the 5700G, but the Radeon 680M found in current mobile APUs. And that iGPU is already a lot faster than the Radeon 8. But all of this is moot because again, this benchmark result has absolutely zero bearing on gaming performance. edit: These supposed leaked TimeSpy scores are probably more relevant to gaming performance: https://wccftech.com/amd-radeon-780m-rdna-3-igpu-comes-within-striking-distance-of-nvidia-gtx-1650-dgpu/ It's still quite decent performance, but people need to remember that the previous laptop parts weren't all that much worse, and it's definitely not gonna be anywhere near a 980 lmao. Dr. Video Games 0031 fucked around with this message at 15:30 on Mar 15, 2023 |
|
#
?
Mar 15, 2023 14:37
|
|
|
The 1650 max q is a turd for a lot of games anyway, so matching it in performance is kind of a non-story
|
|
#
?
Mar 15, 2023 16:10
|
|
|
It's not a revolutionary improvement but it is a nice one.
|
|
#
?
Mar 15, 2023 16:16
|
|
|
Kazinsal posted:the minute a top end GPU becomes more expensive than a mesa boogie I'm buying a loving console the Titan Z was $3000 9 years ago. i wouldn't measure based on a "top end" whale hunter GPU vs a fixed price, lower performance console.
|
|
#
?
Mar 15, 2023 17:07
|
|
|
FuturePastNow posted:It's not a revolutionary improvement but it is a nice one. Yeah, though I would say the last few generations of iGPUs have all had pretty big steps which is pretty awesome. Comparing RDNA3 with the iGPUs of like 7 or 8 years ago is pretty revolutionary.
|
|
#
?
Mar 15, 2023 17:42
|
|
|
Cygni posted:the Titan Z was $3000 9 years ago. i wouldn't measure based on a "top end" whale hunter GPU vs a fixed price, lower performance console. It wasn't targeted directly at the gamer though, there have always been high priced workstation cards
|
|
#
?
Mar 15, 2023 18:22
|
|
|
HalloKitty posted:It wasn't targeted directly at the gamer though, there have always been high priced workstation cards The Z definitely was, you can see it on their page (which is still up, lol) https://www.nvidia.com/en-us/geforce/graphics-cards/geforce-gtx-titan-z/
|
|
#
?
Mar 15, 2023 18:40
|
|
|
If you really wanna pay Nvidia some money you could always get into their current pro line
|
|
#
?
Mar 15, 2023 18:43
|
|
|
DoombatINC posted:If you really wanna pay Nvidia some money you could always get into their current pro line Out of curiosity where is the bulk of the increased cost in these render farm cards? Is it the extra ram or something else?
|
|
#
?
Mar 15, 2023 18:54
|
|
|
The cost comes from being sold to people with the kind of money that can afford those prices.
|
|
#
?
Mar 15, 2023 18:56
|
|
|
VelociBacon posted:Out of curiosity where is the bulk of the increased cost in these render farm cards? Is it the extra ram or something else? The increased cost is mostly because of the license, the software license on Nvidia drivers specifically prohibit gaming Geforce cards from being used in a datacenter at any time. Well, actually IIRC they made a specific carve-out to allow cryptocurrency mining, but only cryptocurrency mining. Edit: Realistically, you're also paying a price premium to get the blower cooler. Nvidia will not allow gaming GPUs to have a blower-style cooler anymore, but if you know a criminal who has some access to RTX 4090 GPUs, they are being made:  https://www.tomshardware.com/news/geforce-rtx-4090-blower-gpu-blows-hot-and-loud No warranty, support, and also you gotta pay the vendor in cash or crypto. Twerk from Home fucked around with this message at 19:01 on Mar 15, 2023 |
|
#
?
Mar 15, 2023 18:58
|
|
|
Double sided polygons means double the price
|
|
#
?
Mar 15, 2023 19:18
|
|
|
Twerk from Home posted:The increased cost is mostly because of the license, the software license on Nvidia drivers specifically prohibit gaming Geforce cards from being used in a datacenter at any time. Nvidia's data center driver will also only install on its data center cards, iirc. Also the ECC on the pro products is certified, even though most modern GPUs also have ECC memory and some will even let you turn it on (like the 4090), but it isn't certified. But like pointed out above, as always with any product the answer to "why does it cost this much" is "because they think people will pay it". The price increases between the gaming and pro cards are laughably small compared to the cost of literally anything else in professional land. Cygni fucked around with this message at 19:43 on Mar 15, 2023 |
|
#
?
Mar 15, 2023 19:40
|
|
|
Also you get full speed double precision floating point.
|
|
#
?
Mar 15, 2023 19:42
|
|
|
You're pretty much paying for the VRAM Those parts are typically slower than the x90 gaming cards
|
|
#
?
Mar 15, 2023 19:48
|
|
|
Shipon posted:eh, a top end GPU might be a bad per-dollar value but it's still better than the NaN value a console offers (what games lol) What games you think high-end GPU users are buying exactly, chances are they're console ports.
|
|
#
?
Mar 15, 2023 19:51
|
|
|
Also MSRP price or whatever is always the most one will pay. Not sure how it is with Nvidia but any other professional hardware you can haggle up to like 90% off depending on a lot of things.
|
|
#
?
Mar 15, 2023 20:05
|
|
|
hobbesmaster posted:Also you get full speed double precision floating point. only on the V100/A100/H100/etc datacenter-only parts, they've long since dropped double precision support from the silicon that's shared between gaming cards and everything-but-the-flagship professional cards
|
|
#
?
Mar 15, 2023 20:12
|
|
|
doomisland posted:Also MSRP price or whatever is always the most one will pay. Not sure how it is with Nvidia but any other professional hardware you can haggle up to like 90% off depending on a lot of things. H100s are $30,000 USD each and I haven't seen even a rumor of discounting out there.
|
|
#
?
Mar 15, 2023 20:14
|
|
|
Twerk from Home posted:H100s are $30,000 USD each and I haven't seen even a rumor of discounting out there. NDAs between NVIDIA and the maybe dozen customers for the H100 for pricing are probably a bit more tightly controlled than the technical details that are by necessity shared widely within AIBs.
|
|
#
?
Mar 15, 2023 20:31
|
|
|
Twerk from Home posted:The increased cost is mostly because of the license, the software license on Nvidia drivers specifically prohibit gaming Geforce cards from being used in a datacenter at any time. Forgive my ignorance, but what would be the point as far as functionality compared to the current air cooling solution? Being able to propel your Fractal Torrent FORWARD?
|
|
#
?
Mar 15, 2023 20:53
|
|
|
Jiro posted:Forgive my ignorance, but what would be the point as far as functionality compared to the current air cooling solution? Being able to propel your Fractal Torrent FORWARD? it allows you to do this  hearing protection sold separately
|
|
#
?
Mar 15, 2023 20:55
|
|
|
They don't have nvlink either so multiGPU is of limited utility
|
|
#
?
Mar 15, 2023 20:58
|
|
|
repiv posted:it allows you to do this  Whhhyyyyyy??? Those Noctua fans sure are going to dampen the jet engine sound that will be your PC I guess?  Also is the PC case hiding the diesel generator that you'll need to run all of those cards and your CPU?
|
|
#
?
Mar 15, 2023 21:07
|
|
|
Jiro posted:
Noctua has a series of industrial fans that would be appropriate there. Anyways that’d be a workstation for AI/ML, CFD or other workloads. Nvidia would probably prefer that you rent what you need from the cloud. If you must have a setup like that on site… well, you’ll have to make it worth it for them I guess.
|
|
#
?
Mar 15, 2023 21:17
|
|
|
repiv posted:it allows you to do this Note: Utility Planning Commission submittal for a new substation is still ongoing.
|
|
#
?
Mar 15, 2023 22:15
|
|
|
Arrath posted:Note: Utility Planning Commission submittal for a new substation is still ongoing. Lenovo will sell you workstation towers that would likely trip a 15A breaker: 
|
|
#
?
Mar 15, 2023 22:25
|
|
|
|
| # ? Jun 3, 2024 17:36 |
|
|
When I move the two gaming PCs to the new spot in the basement I’m going to get 20A service run as well as Ethernet, I think. (Will be 7800X3D/4090 and 5800X3D/3090 and 4 displays, but we use GPU power limits because we aren’t savages.)
|
|
#
?
Mar 15, 2023 22:29
|
|
