Geekbench was tricked: the Ryzen 7 7800X doesn't exist

mongeese

Posts: 643   +123
Staff
WTF?! For the last few days, the computer hardware microcosm of the internet has been awash with theories about an AMD Ryzen 7 7800X that appeared in the Geekbench database on October 27. It hadn't been seen, heard, or rumored anywhere beforehand, which is almost unprecedented. Had AMD really made a 10-core Zen 4 CPU without anyone knowing?

Nope. Chips and Cheese has come clean. They fooled Geekbench with a phony name by spoofing the CPUID on what was actually a Ryzen 9 7950X system. They also disabled six cores and reduced the precision boost overdrive clock by 350 MHz to make it look (and perform!) like a middle ground between the very real 7700X and 7900X. I wonder what AMD's engineers thought when they saw the 'leaked' processor.

The Chips and Cheese team admitted to changing the CPU's name with an internal benchmarking tool originally designed to find bottlenecks in CPU design. It's been available on GitHub this whole time and still is, so if you want to pull off a similar prank, tag us on Twitter when you do. Here are some fun, definitely-real processors that appeared in the Geekbench database this week.

Chips and Cheese's tool, called PMCReader, abuses the way benchmarking tools like Geekbench read a CPU's name on AMD systems. The CPUID for AMD processors is stored across six MSRs (Model Specific Registers) that can each contain eight ASCII characters (so no emoji, sadly). Programs can read the CPUID by accessing these six registers. These MSRs are publicly available in AMD's PPR (Processor Programming Reference) papers, which explain that the BIOS sets them at boot.

Some AMD CPUs as old as Bulldozer allow them to be written to arbitrarily, although some architectures are more okay with name changes than others. Some MSRs can be changed later by programs with admin privileges, including those six registers. With PMCReader, you can give your CPU any name with up to 47 characters: check out Chips and Cheese's hilarious example below.

Chips and Cheese say their tool can fool Geekbench, Cinebench, AIDA64, HWMonitor, the Blender Benchmark, and almost everything else they've tested. So far, the only exceptions to the rule have been HWiNFO and BenchMate (which borrows HWiNFO's tools) because they source the CPUID from a more fundamental level, much like the BIOS does.

It's actually been possible to spoof these benchmark tools for a long time with more than just silly names. Virtualization software like VMWare or CPUs from non-major manufacturers can falsify the CPU model, family, stepping, and manufacturer if you have the requisite expertise. However, with PMCReader, anyone can change the CPUID on an AMD CPU. From now on, it'll be tough to tell if a leaked benchmark result from an online database is real — which adds to the fun if you ask me.

Permalink to story.

 
Not so "fun," though if it would allow scammers to buy cheap CPUs and pawn them off as much more expensive ones. The savvy user might catch on ... the "average" one? I get that the article speaks of "gimping" a proc, but if this is possible I wonder how hard it might be to "glam" one.
 
Not so "fun," though if it would allow scammers to buy cheap CPUs and pawn them off as much more expensive ones. The savvy user might catch on ... the "average" one? I get that the article speaks of "gimping" a proc, but if this is possible I wonder how hard it might be to "glam" one.

Not really, CPU comes in physical form. You can read etched name on the CPU itself. No joy with scamming.
 
Not so "fun," though if it would allow scammers to buy cheap CPUs and pawn them off as much more expensive ones. The savvy user might catch on ... the "average" one? I get that the article speaks of "gimping" a proc, but if this is possible I wonder how hard it might be to "glam" one.
I don't think this is how the tool works: it only rewrites the MSR registers when the tool is used, when you boot you will always get the real CPUID.
 
A 10 core processor rumour never made sense anyway, AMD hasn't produced a Zen-based CPU with an odd number of cores in an active CCD, and there is nothing about the Zen 4 that would indicate AMD has suddenly made that possible. The alternate of a 6 core CCD and 4 core CCD is also extremely unlikely the absence of a 4 core Zen 4 part (likely because they aren't producing enough defective dies to warrant a 4 core SKU), and likely issues with how CCD's with differing active cores would interact with the shared L3 cache.

The only thing AMD needs to do to improve competitiveness with Raptor Lake is reduce the price, the 7950X covers the 13900, drop the 7900X $50 below the 13700K, drop the 7700X to cover 13600K and then the 7600X to cover the 13400 and below (most of which are just Alder Lake rehashes). Price the 3D parts $50-100 ($50 per CCD) more than their non-3D parts when they are released and they should hold the price and price/performance lead until Meteor Lake is released. Given what we know about the Zen4 CCD size and manufacturing cost, this would have a small but not massively detrimental impact on margin, assuming they are not already at max capacity from TSMC.
 
Just another conspiracy that really existed.
Prime example of why you don't believe everything you see on the internet and to avoid the silly rumors.

True. Conspiracies are actually very common. The moment you spot a discrepancy in society, and try to figure out what's really going on and who's behind it, you're engaging in conspiracy theorizing.

But my question is how do you determine that a rumor is silly?

This article claims the 7800x doesn't exist. Is that a silly conclusion? Or is it reasonable to assume that AMD is working on the 7800x?

Just asking.
 
Back