- Joined
- Aug 30, 2004
- Location
- Santa Barbara, CA
Now that my Opty 165 is here its time to play! This is going to take a while, so please be patient! This small write-up is intended for people who are fairly new to overclocking and who have never owned a dual-core CPU before. Hence, expect it to be wordy, heavy on screenshots and light on terminology.
Current Status: CPU OC settled at 2.7GHz. Sold all my RAM, because I'm broke.
I'll run a few SLI tests instead to see if this CPU alleviates any bottlenecks.
System at hand:
Opteron 165 CCBWE 0534 MPMW 131606AK50026
DFI NF4-Ultra D (modded to SLI) BIOS 6-23-3
Fortron Blue Storm AX-500A
512 x 2 Corsair PC3200 Plat v1.1 (BH-5)
XFX 6800GS
I've cooked up a few tests. Please let me know if you would like me to add anything else to this. I hope this would turn out to be some sort of a review and I'm counting on your inputs, suggestions and criticism
Test list:
First impressions:
VCore defaults at 1.33V in the BIOS (although CPUz reports it as 1.312V).
Make a note of the cache latancies. For the A64 Venice the L2 latency is 18 cycles , whereas for the Opty it is 13 cycles (1024 Mb block) and a max of 17 cycles. Still, this is lower than the Venice. It would be interesting to see what kind of SuperPi time's one can get with similarly clocked A64's and Opty's.
By golly! I will never ever go back to a single core CPU again! Working with a DC is absolutely wonderful! Running programs is a seamless process, with no jitters or lagging whatsoever! I bet that is why those who go SMP never look back
Burn-in phase:
Burnt it in at stock settings overnight. My preferred program is StressCPU.
Device manager indicates 2 CPU's, Task manager shows 2 CPU's and I'm running two instances of Stress CPU. If you notice, temps are a bit too high for my liking. 40C at load, at stock, doesn't look very promising. Ofcourse, the heatsink although,lapped doesn't have the perfect mirror finish and AS-5 takes about a week to cure. So, there may be nothing to worry after all.
In case you are wondering how to run a particular process on a particular core, the answer is, by setting the program affinity to a particular CPU. This ss should be self-explanatory. An interesting test would be to see if one can obtain better performance by manually setting affinity instead of letting Windows do it.
Now let us see how far I can push it without bumping up the Core Voltage.
At this point, I'm at 2600 MHz. I'm going to take this nice and easy. In case you missed it, I will be following a strict burn-in protocol. The details can be found HERE (first post).
If you noticed, I have a stepping similar (same week-code) to the one posted HERE. As you can see, the author's final results are tad lower than my initial tests. I would attribute this to the start-up VCore. In my discussion with Reefa_Madness, we were speculating that BH-5 which boots up at low VDRAM are excellent chips. Taking a step in the same direction, I would speculate, that, lower the default VCore better the CPU would OC.(*Edit* Apparently, this was well known in the old school days.) Also, I believe burning in at stock voltage has its merits. Ofcourse, I don't have a twin CPU to test this out, so maybe some of you with more than 2 CPU's can test this for me.
I seem to have hit a wall at 2.6GHz,1.33V. This despite a 16 hr burn-in at these settings. Moral of the story, optimum burn-in time per step < 8 hrs. Time to bump up the VCore. Bumped it up by 0.025V. Now, I'm stable at 2.7GHz, 1.508V.
Here is what I have gleaned thus far, regarding burning in. Burninig in should be used judiciously. One cannot expect to go from total instability (failure in a few seconds) to complete stability. In my case I hoped that would happen after a 16 hr burn-in, to no avail. The system must be in a half-way stable state for a burn-in to succeed. Last night, Stress CPU would fail after 120,000 iterations. After an overnight burn-in (12+ hrs), I'm stable at 1 million iterations!
I've observed that the weaker core usually "lags" the better core. An easy test would be to start two instances of StressCPU and setting the affinities/priorities. Suppose the initial differential is 10 iterations, after several hours, it would jump to 50 iterations etc. It doesn't mean that its unstable, just that it perpetually shows up as being weaker. Note that in RW situations, load is properly distributed by the OS, so this "lag" is not visible.
Another thing I noticed that aws the dip in temps as the burn-in progressed. If you look at the first SS, the temps were consistently at 49-50C (last night), however, this morning the temps were at a surprising low 46C! Now I believe my AS-5 would have finally settled in, to drop the temps, but you are free to draw your own conclusions.
Does burning in help at all?
Yes.Absolutely.
Current Status: CPU OC settled at 2.7GHz. Sold all my RAM, because I'm broke.
I'll run a few SLI tests instead to see if this CPU alleviates any bottlenecks.System at hand:
Opteron 165 CCBWE 0534 MPMW 131606AK50026
DFI NF4-Ultra D (modded to SLI) BIOS 6-23-3
Fortron Blue Storm AX-500A
512 x 2 Corsair PC3200 Plat v1.1 (BH-5)
XFX 6800GS
I've cooked up a few tests. Please let me know if you would like me to add anything else to this. I hope this would turn out to be some sort of a review and I'm counting on your inputs, suggestions and criticism
Test list:
- Lapped Stock HS v/s Lapped XP-90 and thermal performance. Is an aftermarket solution really necessary?
- High voltage RAM performance, BH5/UTT
- Overclocking performance and tests to see if Burning-in really helps.
- If application of DC specific software tweaks really yeilds a boost in performance.
- 3D-Gaming performance and Synthetic Benchmarks runs.
First impressions:
VCore defaults at 1.33V in the BIOS (although CPUz reports it as 1.312V).
Make a note of the cache latancies. For the A64 Venice the L2 latency is 18 cycles , whereas for the Opty it is 13 cycles (1024 Mb block) and a max of 17 cycles. Still, this is lower than the Venice. It would be interesting to see what kind of SuperPi time's one can get with similarly clocked A64's and Opty's.
By golly! I will never ever go back to a single core CPU again! Working with a DC is absolutely wonderful! Running programs is a seamless process, with no jitters or lagging whatsoever! I bet that is why those who go SMP never look back
Burn-in phase:
Burnt it in at stock settings overnight. My preferred program is StressCPU.
Device manager indicates 2 CPU's, Task manager shows 2 CPU's and I'm running two instances of Stress CPU. If you notice, temps are a bit too high for my liking. 40C at load, at stock, doesn't look very promising. Ofcourse, the heatsink although,lapped doesn't have the perfect mirror finish and AS-5 takes about a week to cure. So, there may be nothing to worry after all.
In case you are wondering how to run a particular process on a particular core, the answer is, by setting the program affinity to a particular CPU. This ss should be self-explanatory. An interesting test would be to see if one can obtain better performance by manually setting affinity instead of letting Windows do it.
Now let us see how far I can push it without bumping up the Core Voltage.
At this point, I'm at 2600 MHz. I'm going to take this nice and easy. In case you missed it, I will be following a strict burn-in protocol. The details can be found HERE (first post).
If you noticed, I have a stepping similar (same week-code) to the one posted HERE. As you can see, the author's final results are tad lower than my initial tests. I would attribute this to the start-up VCore. In my discussion with Reefa_Madness, we were speculating that BH-5 which boots up at low VDRAM are excellent chips. Taking a step in the same direction, I would speculate, that, lower the default VCore better the CPU would OC.(*Edit* Apparently, this was well known in the old school days.) Also, I believe burning in at stock voltage has its merits. Ofcourse, I don't have a twin CPU to test this out, so maybe some of you with more than 2 CPU's can test this for me.
I seem to have hit a wall at 2.6GHz,1.33V. This despite a 16 hr burn-in at these settings. Moral of the story, optimum burn-in time per step < 8 hrs. Time to bump up the VCore. Bumped it up by 0.025V. Now, I'm stable at 2.7GHz, 1.508V.
Here is what I have gleaned thus far, regarding burning in. Burninig in should be used judiciously. One cannot expect to go from total instability (failure in a few seconds) to complete stability. In my case I hoped that would happen after a 16 hr burn-in, to no avail. The system must be in a half-way stable state for a burn-in to succeed. Last night, Stress CPU would fail after 120,000 iterations. After an overnight burn-in (12+ hrs), I'm stable at 1 million iterations!
I've observed that the weaker core usually "lags" the better core. An easy test would be to start two instances of StressCPU and setting the affinities/priorities. Suppose the initial differential is 10 iterations, after several hours, it would jump to 50 iterations etc. It doesn't mean that its unstable, just that it perpetually shows up as being weaker. Note that in RW situations, load is properly distributed by the OS, so this "lag" is not visible.
Another thing I noticed that aws the dip in temps as the burn-in progressed. If you look at the first SS, the temps were consistently at 49-50C (last night), however, this morning the temps were at a surprising low 46C! Now I believe my AS-5 would have finally settled in, to drop the temps, but you are free to draw your own conclusions.
Does burning in help at all?
Yes.Absolutely.
Last edited:
) .
