DID INTEL SEND REVIEWERS GOLDEN CPUS? Retail vs Review Samples
When Intel’s Alder Lake CPUs launched, I had a few comments and heard a few rumours that the chips us reviewers got were… well.. Let’s say “special”. They were the “creme of the crop”, the best, the highest binned, the best overclockers, the lowest power consumption – basically a cherry picked “Golden Sample”. That means what you, the hard working consumer, would get isn’t what we as tech reviewers received, and that’s a big deal.
Now, to be clear, these sorts of allegations are far from new. Back when the 8700K was launched, it was suggested that the review samples sent to reviewers were cherry picked, and in fact the retail samples would perform considerably worse. In fact, this practice has been alleged (or outright caught) for decades. AMD were accused with the launch of their R9 290X GPUs, EVGA were caught sending a ‘polished’ version of their power supply for review, and the list goes on.
But, why? Well on the face of it, it’s pretty simple. The company, in this case Intel, sends media publications a ‘slightly special’ sample that’ll perform a little better, maybe draw less power, etc, then we as media get to report “Intel’s new CPU destroys AMD!” and then you, the consumer, take that information and make a purchasing decision based on it. Now there’s a pretty big risk that if the samples are too much better than retail that’ll be obvious, but a little performance bump might be enough to swing some conclusions.
So, did they? Well the lovely folks at Cyberpower sent me this system, their Infinity X125 Pro PC with an i5 12600KF, RTX 3070, DDR4 board and RAM and a 1TB SSD, as well as both the i7 12700KF which I’ll be checking out in it’s own video soon, and an i9 12900K. That means I can test these retail chips against both the i5 and i9 Intel sent me as part of their press kit prior to the launch. I’ve retested all four of these chips in this system with this GPU and DDR4 for a proper comparison.
So are my review samples rigged? Well looking at Cinebench R23 single threaded…. Technically speaking, both retail samples are a fraction FASTER than my review samples. It’s under a single point on the i5 – AKA completely even – and it’s only a couple points on the i9 at around 1% faster so again close enough. In multi-threaded the retail i5 stays a fraction faster, but this time my review i9 is a hair higher. We are only talking about half a percent though, which is easily within margin of error.
Just to confirm this isn’t just boosting in Cinebench I also ran the Blender BMW scene and.. Yeah. Almost identical. My review i9 just about scraped a second faster but that’s a single percent faster so nothing worth writing home about. Interestingly, the peak power usage on the i5 was consistently around 5 watts less than my review i9, which could be because this is the KF meaning it doesn’t have integrated graphics but at no point do I use the iGPU so maybe not. My review i9 on the other hand runs over 10 watts less at peak, which seems significant but I’ve had what I believe to be power reporting issues with this chip in particular and looking at the stable power level it’s back to within margin of error so I wouldn’t be too worried.
What about gaming? Well, in CSGO both retail chips outperformed the review samples in all metrics. The i5 ran around 5% faster, and the i9 was around 2.5%, and even in the 1% lows both retail chips bested their review sample counterparts.
In Watchdogs Legion the chips almost completely tied with the only catch being a bit of variability in the 1% low figures and the 12600KF netting 2 FPS average more, but again that’s pretty spot on and isn’t helping prove the null hypothesis wrong.
Shadow of the Tomb raider isn’t helping either, with both retail chips taking the very, very slight edge in every result. Even the CPU render average and 95% figures are generally within a percent or two.
So, is Intel sending us reviewers cherry picked, rigged, “golden sample” chips? Well it sure doesn’t look like it, although a sample size of 4, a conclusive study does not make. But, even if you take an average of every reviewer who tested with DDR4 which is actually a shockingly low number compared to the number of reviews thanks I’d imagine to the complete lack of motherboards available to test on, you’ll still find those reviews are under 1% faster than my retail samples in Cinebench, both single and multi-threaded. Again, that’s still not a conclusive scientific investigation but it’s the best I can do.
That’s also not to say this sort of thing couldn’t, or doesn’t happen. It certainly can and does, but in this case it doesn’t look like that’s happened here, and in fact a number of those “cherry picked” accusations were caused by things like the “Multi-core enhancement” setting Asus has had enabled by default for years which disables the stock Intel power limits allowing chips to push run harder for longer, and the AMD 290X issue was a fan curve bug they fairly quickly issued a driver update to fix on their reference design cards. It’s good to keep a skeptical eye on this sort of stuff, with the catch that you shouldn’t just assume any review samples are ‘tampered with’, especially without any direct evidence and when allegations like this come out it’s worth getting as much data as possible before making your mind up about it.
https://www.tomshardware.com/reviews/radeon-r9-290-review-benchmark,3659-20.html
https://bit-tech.net/news/tech/graphics/amd-denies-cherry-picking/1/
