Today, Rage3D pulled the AMD Radeon HD 6850 Crossfire results from our first article covering the Sapphire HD 6850 and AMD Radeon 6800 series introduction (review link). Why? Well it turns out the Sapphire Radeon HD 6850 we received wasn't representative of the HD 6850 product you can buy off the shelf. It was based on an early engineering sample sent to partners for board validation purposes, and was never meant to see the light of day. The key difference between the press sample Sapphire Radeon HD 6850 and the retail edition is the number of STREAM processors - 1120 instead of AMD's specified 960. Basically it's a downclocked AMD HD 6870. This invalidated our Crossfire performance testing, as we'll detail below. For those of you excited about the possibility of 'unlocking' your HD 6850, it's not going to happen; the disabling is done at an ASIC level, not BIOS. These cores were provided as basically 'throwaways', so boards could pass whatever testing for safety and fit they required. They might be indicative of late product design changes; was the 6870 originally destined to be lower than 900MHz? Likely not, but it's fun to speculate.

AMD Radeon HD 6850

Sapphire Radeon HD 6850

The story starts last week, on October 14th. Rage3D was in Los Angeles, California, invited courtesy of AMD to attend their Northern Island tech day, and flew in the previous day. We learned about (amongst other things ;-) ) the AMD Radeon HD 6800 series, codenamed Barts. Friday afternoon the assembled press and literati went on their merry way. The Caveman headed home, press samples of the AMD Radeon HD 6870 and AMD Radeon HD 6850 clutched tightly in hand. Airport security was fun. Due to departure times, delays and general homeland security shenanigans, the Caveman got in around 1AM Saturday morning, to sweet home Alabama. Later that morning, the Caveman had to perform his cavedad duties and coach kindergarten soccer, as well as get 'reacquainted' with the Cavewoman. Late afternoon Saturday, we got the chance to fire up the brand new products in our test system. Drool.

Sapphire Radeon HD 6850

Sapphire had emailed us the previous week that they would be sending us a sample of their HD 6850 for launch date - which you now know was October 22nd at 10PM EST. So we're at Saturday afternoon, giving us until Thursday evening to get our performance and feature testing done, and talk about the new bits; plus compare performance to relevant products. We knew we were going to need to compare against the GTX 460 1GB, as well as the HD 5850. Additionally, we had hoped to test the 5830, 5870, and 5970 - the last against Crossfire 6850s. Reality set in, when the Caveman remember he had a day job, and taking another four days off wouldn't be acceptable; working for Rage3D is cool 'n junk, but it ain't putting bacon in the pan or clothes on backs. Yet. Sapphire estimated that the card would be with us at the beginning of the week, which worked out fine.

The weekend passed by, with the 6870 and 6850 testing completing pretty normally. We started with the 6800s as (a) we were excited about them and (b) it gave us the most time to find and potentially resolve issues with drivers or applications, before launch. Monday came, and with it the Sapphire HD 6850. At this time the Caveman was benching the HD 5850, having completed the 6800 series tests over the weekend. At this point, it was decision time - redo all the 6850 benchies, or just move straight to 6850 Crossfire? The two cards were supposedly identical - no factory overclock, no special sauce; just a non-reference cooler. We decided to get the temperature readings from the new cooler in single card use, then move on to 6850 Crossfire results. Benchmarks and game testing continued, we got the GeForce card done, and moved on to write up on Thursday.

The review is published, we're sickly and tired, but it's done. Whew! Time for bed. The Rage3D crew variously awake Friday morning to see comments and discussion about orders, etailer price gouging - the usual stuff. Giggity. Until the afternoon, when BenchmarkReviews.com's executive editor Olin Coles posted their notice about the Sapphire, XFX and Powercolor HD 6850s with 1120 stream processor count, and gives us a heads up; shortly after confirmed by AMD themselves:

AMD: Apparently a small number of the AMD Radeon HD 6850 press samples shipped from AIB partners have a higher-than-expected number of stream processors enabled.

AMD: This is because some AIBs used early engineering ASICs intended for board validation on their press samples. The use of these ASICs results in the incorrect number of stream processors. If you have an HD 6850 board sample from an AIB, please test using a utility such as GPU-z to determine the number of active stream processors. If that number is greater than 960, please contact us and we will work to have your board replaced with a production-level sample.

AMD: All boards available in the market, as well as AMD-supplied media samples, have production-level GPUs with the correct 960 stream processors.

You may have also seen published an email from Sapphire about press sample HD 6850s with improperly applied TIM causing a recall. This email is genuine, not a 'smokescreen'' for the switched ASIC's; our press sample was not one of the serial numbers listed in the recall, but did feature the 1120SP core.

Oh FFFffffuuuuuu

Mea Culpa

At this point we Picard facepalm and utter the traditional expletive, as demonstrated above (courtesy of long-time Rage3D member, Jonz). Now what? What's affected by this? Well, all the Crossfire results are bogus. Do we rename them 6850 + 6870? No, the clocks are wrong. Best to just get rid of them. OK, that's simple thanks to our XML driven flash charts. What about the 6850 results? Well, they're solid - they were performed on the AMD press sample, with correct clocks and stream processors. So they're good. We can't rebench the Sapphire card, it's got the wrong number of shaders, but what about our product recommendation? Well, the only thing that might be different in our results vs. a retail sample would be the power draw and temperatures; both of which would tend to be lower in retail than our extra stream-processor equipped press sample. On that basis, we're not changing the award we gave. Hopefully this transparency into our review and test procedures will continue to give you confidence in our reviews. Rest assured, in future we'll remember the old adage about assuming the product is what it should be. Our bad. Got questions, comments or concerns? Please share them with us in our Review feedback thread, here.