AMD last week teased a press photo of a 7-nanometer Vega product, the Radeon Instinct, aimed at machine learning and artificial intelligence workloads. The company didn't get into detailed specs or performance, both of which are of interest, considering that 7nm Vega GPUs could potentially end up in consumer graphics cards at some point. But in the absence of official data, some benchmarks of the new card have popped up in Futuremark's 3DMark 11 database.
Before getting to the benchmarks, one thing to keep in mind is that it's likely an engineering sample and not necessarily the final design that we're looking at. It's probably close to it, but until the GPU/card actually starts shipping to customers later this year as AMD indicated would be the case, the company could still make some minor tweaks. Either way, we're also looking at early drivers that are not fully optimized or mature. So, take the results with a grain of salt.
Okay, let's have a look at the numbers:
The above results pit the Radeon Instinct based on Vega 20 (left) with a Radeon RX Vega 64 Liquid (right). We see the two cards trade blows, in the individual tests, and the Radeon RX Vega 64 Liquid coming out a little bit ahead in the overall score. That's either encouraging or disappointing, depending on your perspective. It's disappointing if you wanted to see a performance boost right away from a 7nm GPU, and encouraging if you take into account what the Radeon Instinct was built for, along with using drivers that are not optimized for gaming.
One interesting thing to note is that the 3DMark listing confirms that the Radeon Instinct is armed with 32GB of HBM2 memory. The listing also shows the card running at 1,000MHz, though we suspect 3DMark is misreading the clockspeed, just as it did with Vega 10 before it was released. Both sets of scores above were recorded on a testbed with a Ryzen 7 1700 processor.
AMD is planning to ship its 7nm Vega cards to customers this year. Beyond that, it's not yet clear if AMD will build a gaming card based on the same GPU, or if we'll have to wait for Navi.