The GeForce RTX 4090 is one of the best graphics cards for gaming, but consumers have found other uses for the Ada Lovelace flagship, including content creation, machine learning and scientific workloads. However, the GeForce RTX 4090 may have lost its appeal as an NVIDIA employee confirmed that his graphics card does not support P2P (peer-to-peer) functionality.
In short, P2P is a great technology that debuted on Nvidia graphics cards a few years ago. This feature basically works like a highway between two of his Nvidia graphics cards. P2P allows you to transfer data between graphics card memory, bypassing memory on your system. This is a great feature for users with CUDA programs because P2P speeds up memory accesses and transfers instead of the data passing through system memory.
Puget Systems (opens in new tab) We recently ran various multi-GPU benchmarks on AMD and Intel systems and discovered that P2P was partially broken on the GeForce RTX 4090. The publication noted that P2P-related workloads fail or are corrupted. The simpleP2P test fails if P2P is enabled. 1 his Nvidia owner documented the problem (opens in new tab) Back in November last year, we detailed that standard Nvidia tests failed when the system was loaded with two GeForce RTX 4090 graphics cards.A few months later, an Nvidia employee replied to the thread (opens in new tab) Confirmed that GeForce RTX 4090 does not support P2P.
“Hi. Sorry for the delay. Our feedback from engineering is that peer to peer is not supported on the 4090. Applications/drivers should not report this configuration as peer to peer capable. The report has been fixed. and future drivers will report instead,” an Nvidia rep wrote.
P2P works with PCIe or Nvidia’s NVLink. The advantage of using NVLink is lower latency and significantly increased bandwidth. Interestingly, even the GeForce RTX 10 series (Pascal) products supported P2P over PCIe. Unfortunately, the GeForce RTX 20 series (Turing) graphics cards were the last generation to support P2P. However, P2P support is only available via his NVLink bridge, effectively limiting the total number of graphics cards per system to two. Even the previous GeForce RTX 30 series (Ampere) graphics card he doesn’t support P2P, as proved by Puget Systems’ tests.
Thinking about this issue, it’s easy to understand why Nvidia has phased out P2P support on GeForce graphics cards. Chip makers have always disallowed consumers to use their mainstream products for purposes other than gaming. However, in Ampere’s era, some system integrators began using the blower version of the GeForce RTX 3090 rather than his more expensive Quadro alternative in order to offer cost-effective server offerings. Rumors spread. As a result, it didn’t take long for the GeForce RTX 3090 blower graphics card to disappear from the market.
If the $1,599 GeForce RTX 4090 beats out the $6,800 RTX 6000 Ada Generation at home, it’s certainly not good for business. We don’t want to sacrifice performance in our products to beat our rivals, so we need to find other ways to differentiate our mainstream products from professional and workstation parts, such as playing with feature sets. And the RTX 6000 Ada Generation, which also lacks NVLink, supports P2P. However, since the GeForce RTX 4090 and RTX 6000 Ada Generation are on the same Ada Lovelace architecture and use identical AD102 silicon, why one can do his P2P and the other can’t is not. Maker.
It’s not that Nvidia’s A-series products (previously Quadro) were bad. It has several selling points such as certified drivers for specific professional applications, better software compatibility and more VRAM. However, you’re going to pay a hefty premium for these features, and if P2P is important to your day-to-day operations, the RTX 6000 Ada Generation is the best choice if you have the cash. The previous RTX A6000, priced at $4,650, is still available and also supports NVLink and P2P.
