The NVIDIA V100 was a family of cards 100% focused on the professional anddata centers. If the card was formidable at the time and it cost an arm and a leg, today the situation is quite different since it can be found for around a hundred dollars on the second-hand market. So, what is this AI card worth compared to contemporary consumer solutions? This is the question that a colleague asked himself.
NVIDIA V100 vs RTX 3060 vs RX 7800 XT en IA, qui gagne ?Â
One solution that you need for… L’adaptation!
In order to carry out its comparison, our colleague therefore acquired the famous card, but in an SXM2 format. The only problem is that this type of format is not natively compatible with aPC standard, you must therefore start purchasing an adapter to be able to install the card on a slot PCIe classic. On the other hand, the card being intended for the domain ofdata centerwe find passive cooling. Printing a sheath is therefore necessary in order to mount a fan, here, a model from Noctua. In short, after all these adjustments, the bill is around $230, i.e. a selling price lower than an RTX 3060 and an RX 7800 XT!
Who’s the perfs:
Despite the great age of the card, the V100 performs seriously compared to its rivals. Faced with the RX 7800 XT on gpt-oss-20b, the old NVIDIA card managed to generate 130 token/s compared to 90 for the 7800 XT. Otherwise, compared to the RTX 3060 on Gemma4:e4b with 20b settings is 42% faster than the smaller RTX.
But beyond raw performance, we must remember excellent efficiency in this sector of artificial intelligence. In absolute terms, the server card displays high consumption: 293W compared to 235W for an RT 3060. However, coupled with performance, we find a token/s/Watt ratio of 0.37 for the V100 compared to 0.33 for the 3060. And if we limit the consumption to 100W, the Volta card displays a token/s/watt ratio of 0.55 compared to 0.39 for the RTX 3060, i.e. a performance/consumption ratio 41% higher than the general public green card.
Now you have to keep in mind that this is the 16 GB version of the card. It would be possible to obtain better performance or manage larger models with the 32 GB version of the card, however, the prices are no longer the same. We are talking about a bill of between $400 and $500.
