Woot! First shot at overclocking looking good!

At least to me.

I decided when I planned my upgrade, I would go for a mild to moderate overclock. Yeah, right... Like when I'm on a bicycle heading down hill, I'm going to maintain a mild to moderate speed. (Just in case it's not clear, I'm not going to touch the brakes on a down hill unless I need to scrub speed to make a turn. )

I'm starting with a Phenom II X4 820 (stock at 2.8GHz) plunked into an MSI 880GMA-E45 with 4GB of what is probably G.Skill's cheapest DDR3 1333 RAM. This is all installed in an old Lian-Li case and with a CM hyper 212+ cooling the CPU itself.

Anyway, I followed Dolk's Guide to the Phenom II

I bumped the FSB up to 250 while holding "other stuff" more or less level by reducing ,multipliers. The system was stable there so I started bringing multipliers back to where they belong to get other frequencies to where I thought they were supposed to be. At one point I thought I had a stability issue (PEBCAK!) so I increased CPU voltage by 0.02V (IIRC) Anyway, things are looking good, I'm excited and I now have Prime95 running in torture test mode. I'm even running a couple minor benchmarks while doing this and getting surprisingly good results on those! But I'll run them again when the system is unloaded. And fully loaded, the CPU temp is reported at 43° C with ambient about 20° C.

So, 25% bump? Heck yeah!

I'm going to let the Prime95 run overnight and if all is well in the AM, I'll go back in the BIOS and update with specifics on the settings I've changed.

best,
hank

Looking forward to the results. We don't get many 820's here so it'll be interesting.

QuietIce said:

Looking forward to the results. We don't get many 820's here so it'll be interesting.

Click to expand...

It did not survive the night. Logs indicate it ran about 6 hours before locking up. I nudged processor voltage up a bit and turned off Spread Spectrum now that I know what it is!

This part of the testing could take some time. I know it ran six hours before it locked up.

I was surprised by the paucity of information specific to the 820. At $100 (US) it seemed like a bargain to me, $10 more than a 555 and about 2/3 of the 955. Both of those seem pretty popular.

Short story: A 2.8 GHz Phenom II is running well at 3.5 GHz.

The system has run for over 24 hours now. At present it's stable.(*) In that time, I've been running Prime95 stress test on all four cores. In addition to that and from time to time, I've run burnMMX on one or two cores, ripped and transcoded some DVDs and some digital video from a camcorder. I even built a recent kernel because past experience showed that this is a pretty good stress test for a Linux system (I probably should have mentioned that I'm running Linux.) I'm cautiously optimistic that the system is stable. During stress testing, temps varied between 42° C and 45° C.

What surprised me the most is how responsive the system was with Prime95 running. I suppose that's a testament to the progress that has been made with the scheduler.

At present, the following settings have been modified:

• CPU Phase control (under "Green Control") disabled. This should disable Cool and Quiet as well.
• C1E support enabled. (I'm not sure what this does, but if disabled, the processor runs at 100% clock rate at all times.)
• So called FSB frequency set to 250.
• CPU ratio - set to x14. (Where it is locked on this processor.)
• CPU-NB Ratio set to x8, resulting in CPU-NB Frequency of 2000 MHz.
• FSB-DRAM ratio set to 1:2.66 resulting in DRAM Frequency of 1333 MHz.
• HT Link Speed set to x8 resulting in HT Link Frequency of 2000 MHz.
• CPU Voltage set to 1.346V
• CPU-NB Voltage set to 1.214V

CPU voltage was the last setting I played with. With FSB at 250 and CPU at 1.344V the system was not stable. Dropping the FSB to 243 resulted in stability as did increasing the CPU to 1.346V. So I went with the increase in CPU voltage.

(*) How can I ensure stability down the road? Do I need to retest when the weather warms? Do people back off on the OC to provide some margin? Should I bump the CPU voltage up just a bit? Is this something I can set and forget or will it require attention?

Since I'm not running Windows, I cannot provide CPUZ screen shots, but if I left anything out, I can fill in any missing details.

Next, I need to run some of the benchmarks I ran with the baseline system.

Thanks again to Dolk whose guide I used for my roadmap.

-hank

Looking good!

Temps can run stable up to 50°C load core and often into the mid-50's, occasionally higher, so if you want to go for more I don't see any problem with it.

Many Crunchers (BOINC users) will run a higher overclock in the winter and lower it in the summer. If your ambients temps will be changing a lot from summer to winter you can do that as well. If you can save Profiles in BIOS than save one here for future use and keep working upward. If you don't stress the CPU 100% in normal usage than all you have to worry about is getting it stable at 100% load. Even if the ambient goes up a little you probably won't see an impact on day-to-day use.

Thanks again for the tips. As long as I'm happy with present results and temps seem reasonable, I'll stick with what I have. From time to time I'll rerun the stress test to make sure that things are OK. It seems like I've got headspace to bump the voltage up a bit to insure stability should that be needed. The BIOS does support saving sets of settings so I've done that at several points in the process just in case I had to back up to a previous known stable setup.

I did run some CPU oriented benchmarks and compared to a couple other systems. The ones I chose are:

• old - This is the mobo that the new setup replaced consisting of an Athlon X2 3800+ with 3GB of RAM. It actually benches slower than an Athlon 3400+ with 2GB of RAM because RAM runs slower when all four DDR slots are occupied. )
• T500 - My other 'fast' system. A 2+ year old Thinkpad T500 with Intel Core 2 Duo T9400 at 2.53GHz
• New - The system I upgraded to - AMD Phenom II X4 820 clocked at stock 2.8GHz
• New/OC - The same but clocked at 3.5GHz

Benchmarks I reported include:

• bogomips - a number produced by the Linux kernel and probably used for internal busy/wait timing. Most likely a tight loop which runs entirely in the CPU.
• mbw - a Memory Bandwidth benchmark which reports results for three different types of RAM copy. Probably more a RAM benchmark than a CPU benchmark.
• sysbench - Identifies itself as oriented toward database benchmark. I run it using 'sysbench --test=cpu run' which executes a prime number generation benchmark.
• unixbench - An ancient benchmark from a previous age, formerly known as the Byte Benchmarks. Anyone familiar with Byte magazine probably knows about this one.

Results are:


The interesting one is the sysbench. Even overclocked to 3.5 GHz, the new system could not beat a Core 2 Duo. I remember in the K7 days when an AMD CPU could out perform an Intel CPU even when running a slower clock. I guess that Intel learned from that and does not presently allow that to happen!

In an case, CPU, MoBo, RAM and CPU cooler came in at about $200US and I'm pretty happy with the bang for the buck. I had considered upgrading to a Sandy Bridge based setup. I would have gotten a lot more performance, but the processor alone would have cost me more.

thanks,
hank

Update:

I've been running Rosetta and SETI as long term stress tests. Just before I left town for a few days, SETI work units seemed to be flowing OK so I switched to SETI exclusively. At some point while I was out of town, the system started rebooting spontaneously about a dozen times/day. I attribute that to lack of stability due to overclocking. It appears that SETI (including the SSE3 Astropulse app from Lunatics) may stress the system in ways that Prime95 and Rosetta do not. I've bumped CPU voltage from 1.346V to 1.350V to see if that helps. Change in CPU temp seems negligible. I'll be watching uptime to see if this helps.

Rebooting can also be caused by a low cpuNB voltage, at least that's the problem I ran into with my 940BE. Seems that chip doesn't like high cpuNB speeds at all, at any cpuNB voltage I was willing to try, anyway.

Thanks for the tip. It turns out that the increase in CPU voltage did not result in stability. After about 12 hours it rebooted again. But that seems to be an improvement.

I did just bump CPU-NB voltage from 1.214V two steps to 1.219V to see if that helps. It may be a while before I know. I think the reboots may have been triggered by the Lunatics astropulse application. At least this last one happened while it was processing a WU and it makes sense to me that if it shaves about a quarter off the processing time, it stresses the CPU more. (And I just switched from SETI to Rosetta to help with the contest.)

I'm running the CPU-NB frequency at 2000. According to Dolk's guide it should be at 2222. The next step in the bios (8x => 9x) would be 2250. I wonder if I should bump it up. (I'll wait to see what the latest change does.)

Astropulse is a lot more memory intensive than the MB units so it wouldn't surprise me if the problem was in the cpuNB or RAM voltages. Personally, I like to give the vDIMM a little nudge, too.

Aside: If you really want a solid, long-term memory system test run Einstein for a couple of days. Those work units are real memory hogs at ~250 Mb each instance. I've got an HTPC running only 2x1 Gb of RAM on a Vista machine and I've had to limit BOINC's RAM usage on that rig. SETI has been running on it well over a year without problems but once the Einstein units kicked in last fall I had to tone them down and limit BOINC to 50% RAM usage, which made Einstein drop back to three WU's at a time instead of four.