TL:DR - My friends and I were playing around with Linux and accidentally submitted a 9700X3D score, which got written up in the news. I'd like to set the record straight: The 9700X3D isn't real, and we should all learn from this. Remember, all benchmarks can be faked!
A weeks ago, my friends and I were talking about the inner workings of Zen 5. We were talking about how the CPUID instruction works, and how AMD MSRs are technically editable if you ask the processor nicely. One of us realized you could mess with Linux's /proc/cpuinfo to change your CPU to whatever you want, and we were wondering whether benchmark software would detect this... so, to test, one of us took a heavily PBO'd 9700X and changed /proc/cpuinfo to be a "9700X3D" and ran a Passmark run to see if the software would be fooled...
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+7+9700X3D+8-Core&id=6993
...It turns out that Passmark not only didn't notice that /proc/cpuinfo didn't match the CPUID, it actually submitted the result to the real live Passmark database... which is how we got here. Fast forward to today, and I google "9700X3D" out of curiosity. TPU, Videocardz, Tom's Hardware, Notebookcheck, igorsLAB, KitGuru, TechSpot, OC3D, and countless others... all wrote articles about a single unverified Passmark test. Shoutout to Videocardz and KitGuru who (as of the time of writing) corrected their articles.
You can preorder one here, apparently. Crazy.
So, uh, here we are now. I'm writing this post partly to set the record straight that this CPU is not real (as far as I know?), but I'm also writing to tech fans and journalists everywhere, to say: DO NOT TRUST ONE-OFF ONLINE BENCHMARKS!
In this case, we used /proc/cpuinfo to fool the test suite. /proc/cpuinfo is very easy to spoof because it's just an inode (see code below), but it's still possible to spoof any other part of the system too, even the hardware-level AMD64 CPUID instruction (either using a VM or by editing the MSRs using AMD's debugging system) which means that Windows isn't safe either. To be clear, this isn't a problem specific to Passmark, it's just a fact of computing, that there is no real way to 100% guarantee a benchmark is accurate.
You might think "Benchmark companies need to be more careful about accepting results!", which is true, but even if Passmark had checked if cpuinfo matched CPUID, a bad actor might still get away with it by simply changing both.
Really, the only solid takeaway here is that we all need to do better at double-checking any rumours. Many redditors correctly pointed out that the clock speeds were much higher than even a 9800X3D, which is correct. If we had actually been trying to fake a listing, we might have noticed that, but we are doofuses messing with Linux and we were just curious if it would even work.
Some media outlets even started making things up in an attempt to seem informed. TechPowerUp wrote: "Current rumors suggest it will feature a 120 W TDP, targeting the same $400-$450 range as its predecessor." which isn't specified anywhere at all. VideoCardz also suggests a 120W TDP, but they also correctly recognise that the clocks are way too high. To be clear, I have no idea how the 120W rumour started, but it scares me that it only took less than a week before people started making facts up.
... continue reading