Some thoughts on Windows 1903 + AMD's chipset drivers (Patreon)

Hi all,

We've been doing a lot of CPU benchmarking over the past two months. Unfortunately, we keep throwing away data because updates come out that seem important to test. We always try to pre-test as much as we can before new CPUs show up, but sometimes, that does mean we have to do the work twice if important updates launch.

We recently retested Windows 1903 without the new chipset driver, finding that all changes in performance were within margin of error and usual test variance. Because of this, after validating on 3 AMD CPUs and 1 Intel CPU, we decided to keep our existing test data.

After that, we saw that the new chipset drivers had been posted, and so again re-ran the tests with 1903 and the new chipset driver. AMD claimed major gains (upwards of 15% in a very specific scenario) as a result of improved clock responsiveness, but in our testing, we have been unable to reproduce these results in any of our normal games or production applications we test. All results are within margin of error and variance.

We are now evaluating how we want to approach the upcoming reviews, because sticking with 1903 means we'll have fewer CPUs on our charts (we'll have to throw a lot of data away and won't have time to update it all), but moving entirely to 1903 means it's updated. If we saw meaningful gains, we'd definitely make the move. For now, we are considering keeping our existing data (+/-1FPS in all games tested between 1903+chipset and the previous results), but then migrating the new CPUs to 1903+chipset drivers and validating their performance on the old test setup. This would allow the most CPUs on the chart while still getting the updates loaded for the CPUs that may be more likely affected. It's always a challenge of how many CPUs we can test versus how many changes happen between that pre-testing and launch.

We'll figure it out, but just wanted to let you all know that we've done the testing and have seen no benefit as of now. Intel is about +/-2FPS AVG, AMD is about +/-1FPS AVG in our testing. Production results are within normal error margins.

Edit: One more important thing - we will validate this again once the chipset driver is re-pushed by AMD. For now, only a few partners have uploaded it, so we do want to make sure it's the right one and not some mistake by partners.