AMD A6 3600 vs 3620 -- does it matter?

[leuce]

G'day everyone

I just bought an Acer Aspire X3470 for EUR 500. On the box and on the site it says that the CPU is an AMD A6 3600, but SIW tells me the CPU is an A6 3620. The A6 3600 has a cpubenchmark.net score of 4271 whereas the A6 3620 has a score of 3740 (i.e. 500 points fewer). I couldn't find the price of A6 3620 on Google to compare it with the A6 3600 price.

I'm going to use the computer for office work (which includes database stuff), and not for games or multimedia. I hope to use this computer for at least 5 years, and I probably won't upgrade my OS during that time.

When I bought the computer I thought it would have a score of 4271 and I was suitably impressed, but now I see that what's inside the box is a mere 3740, which is rather less spectacular.

My question (and I want your opinions) is whether the discrepancy in CPU should be a problem for me. In other words, should I send the computer back to the store (an online outfit) and ask for my money back (as I doubt that they will have a replacement with the actual A6 3600 CPU in it), or do you think this difference is not worth fighting over?

Thanks
Samuel

[wonderwrench]

The A6-3620 is a step up from a A6-3600 by 100 MHz in both base clock and turbo so you got a slightly faster CPU. I'd guess the reason your results look low compared to A6-3600 results is they are most likely inflated do to overclocking. Comparison link
Also make sure you are using the 64 bit version of PT and disable any down clocking done by power management by setting windows power profile to High Performance. Remember to set it back to balanced when done testing.
Bill

[David (PassMark)]

Yes, the AMD 12h (Llano) family of chips has been causing us a bit of pain.

AMD claims that some of these chips are locked. That is to say the multiplier can't be changed to overclock the CPUs in this family (except for the 'K' variants).

But we have few of these chips in house now and it turns out the locking wasn't done very well.

On some motherboards is seems like the BIOS does the locking. On others it is possible to change the multiplier even if the chip is supposed to be locked.

Overclocking the locked parts seems to result in two different behaviors.

Some chips like the AMD A4 3400 report that the chip is running at a higher speed, but the benchmarked performance doesn't improve. See,
http://www.passmark.com/forum/showthread.php?t=3580

Other chips like the A6-3650 we saw different behavior. Increasing the 'locked' multiplier in BIOS improved the benchmark results, but didn't change the reported clock speed in PerformanceTest (nor in CPU-Z for that matter).

We have been in contact with AMD, as it seems like a CPU bug. But haven't gotten any response as yet.

We have an idea for a hack that might work around the suspected bug, but it hasn't been implemented as yet.

For the OP. The A6-3620 is at least as good as the A6-3600

Update: Further testing & research showed that A) The 2nd behavior was an apparition. Benchmarks don't improve. B) There is a CPU bug.

[David (PassMark)]

AMD never got back to us about the suspected bug. The bug seems to be with these two MSR (Model-specific registers) of the Llano CPUs.

MSR0000_00E7 Max Performance Frequency Clock Count (MPERF)
MSR0000_00E8 Actual Performance Frequency Clock Count (APERF)

The data being returned is in reverse to other AMD CPUs, and doesn't seem to match AMD's documentation.

So we have released a new version of PerformanceTest that attempts to work around this problem.
V7.0 Build 1028 - 23rd/March/2012
http://www.passmark.com/products/pt_history.htm