The core in Nvidia's upcoming PC processor achieves 'performance parity' with Intel and AMD's latest chips, but will it actually be any good for games?
The hardcore microarchitectrure analysts at Chips and Cheese have turned their beady gaze towards Nvidia's GB10 'Superchip' and concluded that its CPU cores are capable of achieving true desktop-class performance equal to AMD and Intel's latest chips. Which is intriguing because because no lesser an authority than CEO Jensen Huang has said that GB10 is the basis for Nvidia's upcoming CPU for the PC, codenamed N1X. But one major question remains, namely games emulation.
Chips and Cheese, as usual, has gone to town on GB10's CPU cores, found inside a Dell Pro Max sporting Nvidia's processor. They're actually Cortex X925 cores designed by Arm and licensed by Nvidia for the GB10 chip, which thus far has been marketed as a device for running local AI models, also including in Nvidia's own DGX Spark box.
Chips and Cheese explains that the Cortex X925 is a bit of a beast, with a huge 10-wide instruction decoder, plenty of cache memory, a powerful branch predictor and "few concessions to reduce power and area. It’s a core designed through and through to maximize performance."
The net result, according to Chips and Cheese, is that, "Cortex X925 in Nvidia’s GB10 achieves performance parity with AMD’s Zen 5 and Intel’s Lion Cove in their fastest desktop implementations."
The website's testing shows the Cortex X925 core very much trading blows with AMD and Intel's latest cores in the hardcore SPEC INT and SPEC Floating Point benchmarks. Exactly how relevant those tests are to real-world performance is open to question. But most impressive is that Cortext X925 does all that with a peak clockspeed of 4 GHz, well below the 5 GHz-plus of AMD and Intel cores.
However, the one question that Chips and Cheese doesn't address is x86 code emulation. For now, emulating x86 code is critical for running PC games. Most games aren't available with a native Arm codepath, so emulating the original x86 code is the only option for running on an Arm chip.
Notably, the main existing Arm CPU architecture available for the PC, Qualcomm's Snapdragon X CPU, has dedicated hardware for accelerating x86 code emulation in its Oryon CPU core. However, as far as I am aware, Arm has not documented any dedicated hardware features for accelerating x86 code on the Cortex X925 core.
It's also worth noting that the Prism emulation layer in Windows on Arm is specifically tuned for the Qualcomm CPU. Now, it seems likely that if and when Nvidia does release an Arm chip for the PC, Microsoft will likewise provide an update for Windows on Arm to support the chip.
Indeed, some reports have claimed that the Nvidia N1X's release has been delayed because that update for Windows wasn't ready. But assuming it does arrive, the question over the Cortex X925's hardware support for emulation will remain. Could it be at a disadvantage?
We know that Nvidia has already designed its own in-house custom Arm core for AI servers in the new Vera CPU. So, it's tempting to speculate that Nvidia may do the same in future for the PC and, if so, including hardware x86 acceleration seems like an attractive option.
It's all speculation for now. But however performant those Cortex X86 CPU cores are in terms of raw compute power, the outcome for Nvidia's new Arm chip for the PC, at least in terms of game performance, could hinge on x86 emulation. Watch this space.