Benchmarking the Latest Hardware is Fun, Unless...

[Rob Williams]

From our front-page news:

Do you think that benchmarking the latest hardware and writing about it is some sort of paradise? I admit, what I do for a living is fun, or else I wouldn't be doing it. But, there are occasions when things don't work out as planned, and the frustration levels go dangerously high. I think all of what I've dealt with over the past few days is rather crazy, so I decided to post about it.

With Intel's Lynnfield right around the corner, time is currently tight. Because of all the things that kept me from getting things done this month, we were behind on getting our AMD Phenom X4 965 Black Edition review posted. Since I received that CPU long before Lynnfield, the sensible thing to do was to shift its priority to a high level in order to get something posted prior to Intel's embargoed content.

That's fine though, right? After all, it's just a speed-bump. Pop it in, benchmark and post the review. Ahh, if only it were that simple! I tend to have a knack for things going wrong, and to spice things up, if one thing goes wrong, something else usually goes wrong right afterwards. The week began out with me building up our AMD test machine this past Sunday night. The first problem? I couldn't find the same RAM kit we have used in the past. RAM is tricky... it's thin, and can hide pretty much anywhere.

So with that being a bust, and our testing motherboard (Gigabyte MA790GP-DS4H) refusing other DDR2 memory I had, I ended up moving to a newer AM3 DDR3 board to get the testing done, even though that's hardly an ideal route to take. After setting that machine up, the problems didn't end. Performance wasn't really up to snuff, and even at this point I'm unsure why. Late Monday night though, I stumbled on the RAM for the previous board (last time I set a stack of magazines on a kit of RAM), so I decided to move back to the MA790GP-DS4H.

The problems were far from over. In order to verify that performance was on par with what we should be seeing, I installed the Phenom X4 810 to run a few benchmarks, to compare the results to our previously-published review. Not surprisingly, the results for each benchmark was about 8 - 10% lower than it should have been. After a very lengthy back and forth with our latest recruit, Robert Tanner (Kougar), we finally figured out that one of our installed drivers (I believe the LAN) had a memory leak, and as a result, up to 8% of the CPU's power was being hogged constantly. That explained the loss. Upgrading to the latest driver solve that problem.

With performance back to normal, all should have been good to go, right? Well of course, but that again wasn't the case. After I installed the 965, there was a serious issue. After running Cinebench, the performance didn't stack up, and after a couple of runs, the PC would shut itself off. Take a look at the above picture, and you can understand why. Mis-mounted CPU cooler? Nope. At this point I'm not entirely sure whether this is a CPU issue or a BIOS issue, but I'm leaning towards the latter.

As it turns out, the BIOS was applying way more voltage than it should have been. The BIOS detected the "default" voltage for the CPU to be 1.4v (still high, but workable), but for some reason, it was actually pushing near 1.5v, according to CPU-Z. So after manually adjusting the voltage to have the CPU sit at around 1.325v, I'm finally getting expected results, and the CPU hasn't gone above 60C.

Whew, what fun! Here's to hoping I don't run into issues like this when getting down and dirty with Intel's Lynnfield! You can also expect our review of the Phenom X4 965 soon (unless the computer blows up).

[2Tired2Tango]

Originally Posted by Rob Williams

As it turns out, the BIOS was applying way more voltage than it should have been. The BIOS detected the "default" voltage for the CPU to be 1.4v (still high, but workable), but for some reason, it was actually pushing near 1.5v, according to CPU-Z.

FWIW... this is something of a known issue with Gigabyte boards.

I always set the voltage 1 step below spec. when installing Gigabyte.

[Kougar]

Hey, those RAM modules are not meant to be used to bookmarks! Especially in Playboy magazines, you oughta know better!

A bit more seriously, you did remember to update the BIOS again after reverting it to that old version I hope! At 1.5v some people were getting some rather decent 4GHz+ overclocks on that chip.

[Rob Williams]

Originally Posted by 2Tired2Tango

FWIW... this is something of a known issue with Gigabyte boards.

That's interesting. I've had this happen to a mild degree before, but not like this. There's a HUGE difference between 1.325v and 1.49v. The most obvious issue I've had like that is with our Rampage II Extreme board. If the DRAM voltage is left to auto, it will set it to 1.69v. It's humorous, because if you manually set to anything above 1.65v, it warns you about it being too high, according to Intel spec.

Originally Posted by Kougar

A bit more seriously, you did remember to update the BIOS again after reverting it to that old version I hope! At 1.5v some people were getting some rather decent 4GHz+ overclocks on that chip.

Yup, reverted back to the latest BIOS. Finished benchmarking now as well, and it went well. Overclocking didn't go THAT well, but that's to be expected due to the overall heat of this room. Got it up to 3.80GHz, but it wasn't stable. I know it can get stable, but at the required voltages, I'm going to have the chip overheat.

[2Tired2Tango]

Originally Posted by Rob Williams

That's interesting. I've had this happen to a mild degree before, but not like this. There's a HUGE difference between 1.325v and 1.49v.

Basically if you want 1.49 set it to 1.325 ...

This I have confirmed with voltemeters in the past. A lot of motherboards do this to some minor degree, to compensate for trace resistance, but Gigabyte gets rather carried away with it.

Set your voltage one step lower than the one you want... You should be fine.

[Rob Williams]

Originally Posted by 2Tired2Tango

Set your voltage one step lower than the one you want... You should be fine.

I haven't tested much multi-meters, but sometimes it's not as simple as just going one step below what you want. To achieve 1.31v, I had to set it to 1.375v... and for 1.45v, I had to set it to 1.475v. It's just so random.

As a follow-up, it is the motherboard that caused an issue. I put the CPU into the other board (also Gigabyte), and it didn't exhibit the issue at all. It actually booted up at 1.40v like it should have.

[2Tired2Tango]

Originally Posted by Rob Williams

I haven't tested much multi-meters, but sometimes it's not as simple as just going one step below what you want. To achieve 1.31v, I had to set it to 1.375v... and for 1.45v, I had to set it to 1.475v. It's just so random.

As a follow-up, it is the motherboard that caused an issue. I put the CPU into the other board (also Gigabyte), and it didn't exhibit the issue at all. It actually booted up at 1.40v like it should have.

You're setting one step above... you should be going below... To get 1.375 you would set the board to 1.31...

The only way to explain the heat problem is that the actual voltage you are getting is higher than that you are setting... 1.31 gets you 1.375, not the other way around.

If you are getting the results you say, that board is a candidate for the trash bin.

[Rob Williams]

Originally Posted by 2Tired2Tango

You're setting one step above... you should be going below... To get 1.375 you would set the board to 1.31...

The only way to explain the heat problem is that the actual voltage you are getting is higher than that you are setting... 1.31 gets you 1.375, not the other way around.

If you are getting the results you say, that board is a candidate for the trash bin.

Well, that could be. The board is put aside now, so I can't say for sure. I did end up getting the right voltages after a while... just a lot of trial and error.

Originally Posted by 2Tired2Tango

If you are getting the results you say, that board is a candidate for the trash bin.

This is the last time I'll be using the board. Once we get our new CPU methodology in order, I'll likely begin using Gigabyte's MA790FXT-UD5P instead, which is a much more stable offering.

[2Tired2Tango]

Originally Posted by Rob Williams

This is the last time I'll be using the board. Once we get our new CPU methodology in order, I'll likely begin using Gigabyte's MA790FXT-UD5P instead, which is a much more stable offering.

Sounds like a reasonable precaution, Rob... When you start getting voltage and temperature problems it's a pretty sure sign the board is on it's last legs.

[Rob Williams]

That's no doubt the case quite often. This isn't a bad board... just that problem for the most part. Back when AM3 first launched, AMD sent these boards along with them (I think), so that's what I used from the get go. For all I know, there might be better boards than the MA790FXT-UD5P by now. I need to check that out...