nVidia ceased production some while back thinking the current stock would be sufficient, only to find it's not?
Yeah it’s possible they miscalculated how bad the supply shortage would be. Next earnings report will be interesting for multiple reasons.
DavidGraham
Veteran
MSI Afterburner is now updated to directly support Smooth Motion.
Only under DirectX 11 though
It's unprecedented as far as I can recall. Would be bad for them if buyers were willing to try an alternative. I think they'll find most buyers will likely buy nothing and wait rather than buy something else. Also Radeon supply is pretty spotty as well.
Potato Head
Newcomer
I don’t believe this rumor about 40 series production stopping back in December. You can buy a Dell with 40 series cards right now and have them in a week.
There’s two components right now that appear to be delaying Dell shipments:
14900KF - 3 weeks
RTX 50 series - 7 weeks (last night 5080 were available, now those are gone)
DegustatoR
Legend
DegustatoR
Legend
It's not for all SKUs, and Dell being Dell may have accumulated ample inventory in advance.
Potato Head
Newcomer
Dell doesn’t carry inventory, they have been JIT manufacturing from their very beginning. For example in their last quarter they had $18.2B in sales but just $3.6B of inventory (including EMC). That’s less than three weeks of inventory for the company.
Having seen their PC plant in Austin they literally have semi trucks unloading components one site of the building and being brought directly to the line. Strait into a PC, tested and loaded onto an outbound truck on the other end of the building. That’s how they can get you a BTO system at your door in a few days.
It’s also the best way to get a video card when there are shortages. By a Dell PC, pull out the card eBay the rest.
Also just to reiterate for those who don't know what @Potato Head meant when he said "including EMC" -- that's their enterprise storage and backup subsidiary. A significant portion of their inventory is going to be consumed with enterprise- or carrier-class servers and storage and networking gear, and some of their workstation-class gear that enterprises use in their SMB lineup. That's where the real money is made for Dell, and hence where (what little there is) inventory is best kept.
The big companies were also trying to hedge, where they could, with advance shipments from Taiwan expecting the Trump tariffs.
Man from Atlantis
Veteran
3DMark Vantage Synthetics Benchmarks, no 5090 yet.
GeForce RTX 5080 GB203
GeForce RTX 4090 AD102
GeForce RTX 4080 Super AD103
GeForce RTX 4070 Ti AD104
GeForce RTX 4060 Ti AD106
GeForce RTX 3090 Ti GA102
GeForce RTX 3070 Ti GA104
GeForce RTX 3060 GA106
GeForce RTX 2080 Ti TU102
GeForce RTX 2080 Super TU104
GeForce RTX 2070 TU106
GeForce GTX 1080 Ti GP102
GeForce GTX 1080 GP104
GeForce GTX 980 Ti GM200
Arc B580 BMG-G21
Arc A770 ACM-G10
Radeon RX 7900 XTX Navi31
Radeon RX 7800 XT Navi32
Radeon RX 7600 XT Navi33
Radeon RX 6950 XT Navi21
Radeon RX 6750 XT Navi22
Radeon RX 6650 XT Navi23
Radeon RX 5700 XT Navi10
Radeon VII Vega20
Radeon RX Vega 64 Vega10
Radeon RX 590 Polaris 30
Radeon RX 480 Ellesmere
Radeon R9 FURY X Fiji
GeForce RTX 4090 AD102
GeForce RTX 4080 Super AD103
GeForce RTX 4070 Ti AD104
GeForce RTX 4060 Ti AD106
GeForce RTX 3090 Ti GA102
GeForce RTX 3070 Ti GA104
GeForce RTX 3060 GA106
GeForce RTX 2080 Ti TU102
GeForce RTX 2080 Super TU104
GeForce RTX 2070 TU106
GeForce GTX 1080 Ti GP102
GeForce GTX 1080 GP104
GeForce GTX 980 Ti GM200
Arc B580 BMG-G21
Arc A770 ACM-G10
Radeon RX 7900 XTX Navi31
Radeon RX 7800 XT Navi32
Radeon RX 7600 XT Navi33
Radeon RX 6950 XT Navi21
Radeon RX 6750 XT Navi22
Radeon RX 6650 XT Navi23
Radeon RX 5700 XT Navi10
Radeon VII Vega20
Radeon RX Vega 64 Vega10
Radeon RX 590 Polaris 30
Radeon RX 480 Ellesmere
Radeon R9 FURY X Fiji
Last edited:
TopSpoiler
Regular
People aren't getting 2x INT32 throughput on RTX Blackwell compared to Ada Lovelace. There's probably an issue somewhere in the compiler/driver/firmware.
forums.developer.nvidia.com
Blackwell Integer
4080 sm_89 Throughput: 1.447048e+13 IMAD/sec 5080 sm_89 Throughput: 1.522615e+13 IMAD/sec 5080 sm_120 Throughput: 1.542646e+13 IMAD/sec
forums.developer.nvidia.com
Nvidia's own documentation still claims half INT throughput for Blackwell: https://docs.nvidia.com/cuda/cuda-c-programming-guide/#compute-capability-12-0People aren't getting 2x INT32 throughput on RTX Blackwell compared to Ada Lovelace. There's probably an issue somewhere in the compiler/driver/firmware.
Blackwell Integer
4080 sm_89 Throughput: 1.447048e+13 IMAD/sec 5080 sm_89 Throughput: 1.522615e+13 IMAD/sec 5080 sm_120 Throughput: 1.542646e+13 IMAD/sec
TopSpoiler
Regular
It's probably a copy&paste error. The CUDA documentation is not up to date yet.
According to the whitepaper it should have the same throughput as FP32.
DavidGraham
Veteran
NVIDIA corrected the L2 amount of the full GB202 chip, it is now 128MB instead of 96MB (this was a typo).
Well, the total L2 size was known since the die shot of GB202 was published.
DegustatoR
Legend
Render Path-Traced Hair in Real Time with NVIDIA GeForce RTX 50 Series GPUs | NVIDIA Technical Blog
Hardware support for ray tracing triangle meshes was introduced as part of NVIDIA RTX in 2018. But ray tracing for hair and fur has remained a compute-intensive problem that has been difficult to…
developer.nvidia.com
Results in older benchmarks like Vantage can vary quite a bit depending on system configuration. I wasn't able to find what exact settings ixbt is using but anyway here's my go at some 5090 results.
These are from the 3DMark_Vantage_v113_installer (which still reports as Build 1.1.2 in the GUI) with the Advanced - not Pro - upgrade.
I used Windows 11 Pro with the desktop set to 1280x1024@60 Hz and G-SYNC disabled.
