My Sandy Bridge Experience (Q6000 -> i7-2600K)

[Stereodude]

So, I'm in the process of putting together a Sandy Bridge based PC to replace my Q6600 @ 3.0gHz as mentioned in this thread.

I went with a i7-2600K + Gigabyte P67A-UD3P + Cooler Master Hyper 212 Plus.

So far it's together and running at stock speed (3.4gHz) with the 2x2GB of 1333MHz DDR3 I borrow from another system since UPS hasn't delivered the 4x4GB 1600MHz G.Skill memory intended for the build. I loaded XP on it temporarily for the best comparison between the two system (so they're both running the same OS).

BD Rebuilder was the primary reason I wanted to upgrade since a 2 pass highest quality compression job was taking longer than overnight on the Q6600.

So far it seems like a mixed bag. In the first pass the i7-2600K @3.4gHz is not significantly faster than the 3.0gHz Q6600. Processing the same movie the i7-2600K shaved 16.56% off the first pass completing it in 77m 55s instead of 93m 23s. This is because the first pass does not max out all 8 threads. Average CPU usage is around 50% on the 2600K (pretty close to 100%). CPU usage is higher on the Q6600 (only 4 threads) on the first pass.

There's a larger difference on the second pass, but it's not done yet so I don't have the final numbers. It looks to be around 70% faster. x264 (the compressor used) seems to be able to saturate all 8 threads.

...to be continued...

 

[CougTek]

Well, regarding MY i5 2400 experience, I hope Speedfan readings are erronous because if they are right, I wasted good money on a third-party heatsink. With all cores at 100% (Folding), Speefan reports temperature between 23C and 27C. Apparently, the retail heatsink would have been just fine. I installed this one and it's a serious overkill. I haven't overclocked the CPU, but one thing being sure is that overheating won't be the limiting factor.

And for those who weren't completly sure, LGA1156 and LGA1155 are exactly the same regarding the position of the heatsink holes. Heatsink compatible with the LGA1156 will also be with the LGA1155.

 

[snowhiker]

Any issues with that Gigabyte board? Seems the first round of Sandy Bridge Gigabyte boards are less polished. Love the color scheme of the board. Very sexy.

2600k = Overclock that sucker!

 

[time]

1. The Sandy Bridge Gigabyte board I used - GA-H67MA-UD2H - has been rock solid so far. I've been running every version of 3DMark, a few other benchmarks, automated backups and surfing the web.

2. I used the stock cooler, which mostly just ticks over and only speeds up slightly when the CPU is stressed long enough. However, my example has an annoying whine - not enough to bother 99% of people, but a consideration if building a truly silent PC.

 

[Stereodude]

Well, regarding MY i5 2400 experience, I hope Speedfan readings are erronous because if they are right, I wasted good money on a third-party heatsink. With all cores at 100% (Folding), Speefan reports temperature between 23C and 27C. Apparently, the retail heatsink would have been just fine. I installed this one and it's a serious overkill. I haven't overclocked the CPU, but one thing being sure is that overheating won't be the limiting factor.

What does CoreTemp report?

 

[Stereodude]

Any issues with that Gigabyte board?

Well, I have had it refuse to boot twice. Both times after being reset decide the BIOS was corrupt was it reflashed itself back to the original BIOS. I'm not sure what that was about. The 3rd time I flashed it to F6A from Windows using their @BIOS utility instead of from DOS (USB boot key). We'll see if it happens a 3rd time.

2600k = Overclock that sucker!

That's the plan after I do a few more benchmarks. :twistd:

 

[Stereodude]

BD Rebuilder update.

The second pass of the test movie end being 80.79% faster. It completed in 4:32:48 on the stock i7 as opposed to 8:13:12 on the 3.0gHz Q6600.

So, in summary... 3.4gHz i7-2600K vs. 3.0gHz Q6600:

Pass 1 - 2600K was 19.85% faster / shaved 16.56% off the Q6600 time
Pass 2 - 2600K was 80.79% faster / shaved 44.69% off the Q6600 time

...to be continued...

 

[Stereodude]

Some Foobar2k FLAC to V1 MP3 (LAME 3.98.4) and FLAC to q.5 AAC (NeroAAC 1.5.4) benchmark results using the 248 FLAC files...

3.4gHz i7-2600K vs. 3.0gHz Q6600:

FLAC to MP3 - 2600K was 94.97% faster / shaved 48.71% off the Q6600 time
FLAC to AAC - 2600K was 113.21% faster / shaved 53.10% off the Q6600 time

...to be continued...

 

[Stereodude]

Some Foobar2k FLAC to V1 MP3 (LAME 3.98.4) and FLAC to q.5 AAC (NeroAAC 1.5.4) benchmark results using the 248 FLAC files...

4.5gHz i7-2600K vs. 3.0gHz Q6600:

FLAC to MP3 - 2600K was 150.32% faster / shaved 60.05% off the Q6600 time
FLAC to AAC - 2600K was 175.61% faster / shaved 63.72% off the Q6600 time


4.5gHz i7-2600K vs. 3.4gHz i7-2600K:

FLAC to MP3 - 4.5gHz was 28.39% faster / shaved 22.11% off the 3.4gHz time
FLAC to AAC - 4.5gHz was 29.27% faster / shaved 22.64% off the 3.4gHz time


...to be continued...

 

[ddrueding]

You really were able to turn it directly up to 4.5Ghz? Holy crap.

 

[Stereodude]

You really were able to turn it directly up to 4.5Ghz? Holy crap.

Didn't even change the Vcore. I just went into the BIOS and changed the multiplier to 45. I got a reboot partway though a test, so went back into the BIOS, found the memory had been set to "Turbo" changed that to "Normal" and changed the PLL voltage to enhanced (I think that's what it called) and booted back into windows. I've got 8 copies of CPU Burn-in v1.01 churning away for over 1 hour so far with no issues. :diablo:

 

[Stereodude]

Well, I have had it refuse to boot twice. Both times after being reset decide the BIOS was corrupt was it reflashed itself back to the original BIOS. I'm not sure what that was about. The 3rd time I flashed it to F6A from Windows using their @BIOS utility instead of from DOS (USB boot key). We'll see if it happens a 3rd time.

It happened a 3rd time. Pressed reset and complains about a bad BIOS checksum and flashes itself back to the original F2 BIOS. So, I flashed it a 4th time to F6a. Got a 4th lockup. This time instead of pressing reset I powered it off. Then powered it back on and it booted normally without all the craziness. I'm not really sure what it's all about. The power supply in the system is a cheapo Cooler Master Elite 460 (since that's what I had lying around). Not sure if that's a factor, or if the beta BIOS is an issue or what. :dunno:

I'll swap out the power supply later and perhaps try the F5 BIOS should it happen again before I contact Gigabyte. I haven't had the F2 BIOS do it, just the F6a.

 

[CougTek]

What does CoreTemp report?

Between 38C and 42C with 4 cores at 100%. More realistic, but it still shows that the SD1283 was overkill. Oh well, at least it was easy to install and it's quiet. 30$ for tranquility isn't that bad.

 

[ddrueding]

I'm really amazed at how quiet my Zalman CNPS9900NTs are. And how much noise F@H on GPUs can make them.

 

[Stereodude]

BD Rebuilder update.

4.5gHz i7-2600K vs. 3.0gHz Q6600:

Pass 1 - 2600K was 41.56% faster / shaved 29.36% off the Q6600 time
Pass 2 - 2600K was 130.47% faster / shaved 56.61% off the Q6600 time

4.5gHz i7-2600K vs. 3.4gHz i7-2600K:

Pass 1 - 4.5gHz was 18.12% faster / shaved 15.34% off the 3.4gHz time
Pass 2 - 4.5gHz was 27.48% faster / shaved 21.55% off the 3.4gHz time

 

[Stereodude]

FWIW, I also don't think the 3.4gHz numbers are really at 3.4gHz. I think it's really 3.5gHz since that's the default Turbo behavior of the Core i7-2600K for 4 loaded cores.

 

[Adcadet]

I've found that if I set my baseclock to 100 MHz (instead of the default 103 MHz), I can hit 100 MHz x 48.

 

[Stereodude]

I've found that if I set my baseclock to 100 MHz (instead of the default 103 MHz), I can hit 100 MHz x 48.

Nice... I'm struggling with some gremlins. I can stress test mine for hours, have it pass without issue. Then reboot, do the same stress test and have it blue screen in minutes. So, for the moment I backed it down a little.

What all have you tweaked / changed to hit 48 x 100?

 

[Adcadet]

I disabled spread spectrum (of the CPU and VRM, although I should probably turn the spread spectrum of the VRM back on) and turned TPU and EPU off. I did this initially because my board had trouble booting, and haven't bothered to turn on TPU or EPU yet.

I found that if I leave the baseclock at 103 MHz I can only hit at 46x multiplier using Vcores of up to 1.4V, but with a 100 MHz baseclock I can hit 48x. As I'm about to go from 8 to 16 GB of RAM, and there's little chance that I'm memory-limited, 100x48 would be better than 103x46 for stability purposes (theoretically).

 

[Stereodude]

I struggled to understand the who Vcore situation at first. In my P67A-UD3P, the default Vcore setting is "Auto" which apparently means the motherboard will change the Vcore as the processor's loading changes dynamically through the Vid voltage range set / specified in the processor. You can override that and set it to a fixed voltage. Or you can set it to Normal (which apparently is the same as Auto), but when set to normal you can specify a positive or negative trim which is added to the requested / used Vid.

My i7-2600k seems to have a max Vid of 1.3711V. When I was running 4.5gHz previously I did not have any extra voltage being added to the requested / used Vid. That is probably why it wasn't quite stable.

Have you run any stress tests on it at 4.8gHz?

 

[Adcadet]

Yeah, with each change I run Intel's Burn Test (runs Linpack) at "standard" and then when I think I've reached the limit I re-run using the "high" setting. I think it makes a difference - my 100x48 seemed fine on "standard" at a Vcore of 1.25V but not when tested at "high", hence I bumped up to 1.3V and was fine.

Here's a link: http://www.techspot.com/downloadget.php?id=4965&file=1&evp=7967b74b9dbf99853e4d4f4dfd60ca78

 

[Adcadet]

I used to use Prime95, but when I built my HTPC (http://www.storageforum.net/forum/showthread.php?t=8201&page=2&highlight=HTPC) I saw that Intel's Burn Test would consistently produce higher temps. Whether this actually translates into a better test, I don't know.

 

[Adcadet]

I still don't fully understand Vcore, nor how the Turbo Frequency vs baseclock fits together when OC'd. I think my 2600k gets an additional boost of up to 400 MHz above my baseclock (4.8 GHz when OC'd) when conditions allow.

 

[CougTek]

Sometimes, a system that is stable for a few hours isn't necessarily stable in the long run. For instance, my damned i7 950 completed three big units with the FAH client (that's more than a week time), but it froze at 97% (more than two days of work) on the fourth one and then it started crashing more frequently (less than an hour). It's not a SandyBridge, but the same principle applies.

I wouldn't increase the Vcore more than 10% above the default value on a 300$ processor. I'm sure it will affect its lifespan significantly.

 

[Adcadet]

Noted. Which is one of the reasons my PC boots up at what the motherboard thought it should run at - 103x44 with the default Vcore (1.2V, I think). If I don't feel the need, I don't load the more aggressive settings.

But at the same time, what are the odds that I will outgrow this CPU before it wears out? From the chatter on the net (which is perfectly accurate, of course), this is unlikely, especially since the Vcore gets reduced when loads are low.

 

[MaxBurn]

I haven't read one story anywhere that had someone fry a CPU that wasn't doing something like raising the voltage to insane levels or actually hurt the thing mechanically. There were some bad chips released from time to time but that's it. Theoretically if you shorten the life from 15 years to 9 will you care?

 

[Stereodude]

Most electrical failures are usually caused by a mechanical breakdown / failure of sorts. It just happens on a very small scale.

 

[Stereodude]

Well, I just reserved a Lian Li PC-A77F (in black) to put my Sandy Bridge Build in.

Microcenter has them in store for $147.95. (many stores are already out of stock) I did the online reservation for in store pickup thing. We'll see if I actually get it tomorrow when I stop by on the way home from work. I got the reservation ready e-mail, so I have my fingers crossed.

 

[ddrueding]

Wow, that looks like a beast.

 

[Stereodude]

I posted this in the other Sandy Bridge thread, but it really belongs here...

I found a nice bug in my Gigabyte motherboard. Running the RAM at 1600MHz (it's rated speed) instead of 1333MHz prevents the CPU from dynamically changing it's clock speed as the load on it changes.

The the RAM at 1333MHz the core will dynamically drop the CPU multiplier down slowing the CPU to under 2gHz (if the core is basically idle). With the RAM at 1600MHz the CPU multiplier just sits at whatever the BIOS is set to regardless of CPU load. :cursin:

 

[Stereodude]

Microcenter has them in store for $147.95. (many stores are already out of stock) I did the online reservation for in store pickup thing. We'll see if I actually get it tomorrow when I stop by on the way home from work. I got the reservation ready e-mail, so I have my fingers crossed.

Well, I was successful. I got the case for $147.95 + 6% sales tax. Not too shabby for a case Newegg sells for $299 + ~$25 in shipping.

 

[mubs]

Case looks really good. Quick check showed prices here in the $570 range. Nada!