We recently published our AMD vs. Intel Dual Core comparison, which led some to wonder why we didn’t compare their quad-core chips instead. As I stated in the review, one reason was that we simply wanted to write a CPU review discussing products that more people can afford; even the high-end dual-core processors we looked at were all under $200. The cheapest quad-core is in the $300 range, both for Intel and AMD. The other problem was that we simply didn’t have a Phenom to test. We have to buy our AMD processors for review, and I hadn’t gotten around to that. I originally wanted to wait until they got their 2.4 GHz part fixed, so I could test the full Phenom lineup.
How wrong I was.
As it turns out, the Phenom launch has been plagued with nothing but blunders, and it may go down in history as one of the worst PC hardware launches in history.
First of all, it wasn’t just the 2.4 GHz part that is affected by the erratum – it’s the entire lineup of Phenom processors, along with all Barcelona Opterons. Basically, the problem is with the CPU’s translation lookaside buffer (TLB) and L3 cache, and can cause system crashes and data corruption. The specifics are as follows:
Erratum 298 will be described as follows: “The processor operation to change the accessed or dirty bits of a page translation table entry in the L2 from 0b to 1b may not be atomic. A small window of time exists where other cached operations may cause the stale page translation table entry to be installed in the L3 before the modified copy is returned to the L2. In addition, if a probe for this cache line occurs during this window of time, the processor may not set the accessed or dirty bit and may corrupt data for an unrelated cached operation. The system may experience a machine check event reporting an L3 protocol error has occurred. In this case, the MC4 status register (MSR 0000_0410) will be equal to B2000000_000B0C0F or BA000000_000B0C0F. The MC4 address register (MSR 0000_0412) will be equal to 26h.”
The reason they were able to proceed with the launch is that this particular erratum is fixable via a microcode update (done using a motherboard BIOS update). Basically, the part of the L3 logic that causes the problem is disabled. However, running the fix instills a performance penalty, because of how the L3 cache’s throughput and latency is affected. There was no way to know exactly how performance would be affected, except for the “around 10%” estimate that has been floating around. Fortunately, our friends at Tech Report have obtained a retail Phenom 9500, and tested its performance before and after applying the patch on an MSI motherboard.
The results were even worse than expected. When looking directly at cache memory performance, bandwidth dropped by as much as 38.7%, and latency slowed down by over 50%!
Of course, what really matters is how the actual programs we use every day are affected. Tech Report ran a full suite of Worldbench tests, to find that the performance drop ranged from around 5% to as high as 50% in programs like Firefox. The average drop in performance in their tests ended up being 13.9%.
So if you bought a Phenom, you have a broken CPU. If you want to fix it, it will cost you. Not in cash, but in performance. It’s basically like paying for a 2.3 GHz CPU but really getting a 2.0 GHz.
It is expected that future revisions of the Phenom chip will fix this error in the silicon itself, so there should be no performance hit at all. If you were considering buying a Phenom processor, you might want to hold off until the next revision (B3) comes out, and is confirmed to work at full capacity.
But that’s not all.
What really got me irate was another finding Tech Report made when doing these tests. When they plugged in their retail boxed Phenom 9500 CPU, they noticed that the northbridge was clocked at 1.8 GHz, rather than 2.0 GHz on the engineering sample 9500 that was provided to them by AMD. This would be fine if AMD had mentioned that the northbridge is not running at retail speed, but according to Tech Report, they were told that 2.0 GHz was the correct speed for the Phenom 9500. Obviously, that is not the case, and once again we are seeing AMD try to ‘trick’ hardware sites into reviewing overclocked products as normal retail products.
The overclocked northbridge helped increase the Phenom’s benchmark scores across the board too, which artificially made the CPU look more competitive to Intel’s competing products.
It’s interesting to see AMD come full circle from when they blacklisted us for giving them trouble over launching ‘paper’ products to review sites that never made it to retail, to kicking Intel’s ass during the PressHot days, to launching a lineup of broken CPUs that not only perform 14% lower when fixed, but are slower than initial reviews indicated to begin with.