Just stumbled across this thread. I have yet to get a Kill-o-watt type of reader but one of the first things I did when I built my current workstation is go into the bios and drop the voltage little by little on the CPU (testing stability in between). I managed to shave off a good ~30 Watts during peak (which is what it runs 24/7) by doing that according to HWMonitor.
I also tightened the memory timings while I was in there, and with relatively cheaper memory have been running stable undervolted and at 7-7-7-22
This AMD rig is still leaps and bounds behind the Intel side of things when it comes to efficiency at load though and thus my next addition to my BOINC fleet will be an i2500k system. I hope AMD can improve their architecture and keep Intel competitive.
For this set of efficiency charts, I thought I would also post curves for stock processors comparing the new Ivy Bridge versus Sandy Bridge.
Here are some key stock numbers for the S6LV Gravity Wave (GW) jobs:
2600K: 4 GW jobs = 13100 sec @ 140 watts;
3770K: 4 GW jobs = 12230 sec @ 112 watts;
2600K: 8 GW jobs = 20093 sec @ 153 watts;
3770K: 8 GW jobs = 16230 sec @ 122 watts;
The stock i7-3770K is running on an ASUS board which seems to automatically overclock without user intervention and CPU-Z reports that it is running at 3.7 GHz versus the stock i7-2600K running at 3.4 GHz. Most components were the same between configurations, the key difference was the use of Windows 7 on the [3770K +Z77 MB] versus Linux for the [2600K +H67 MB], turbo disabled.
I also plotted the efficiency curve for the CPU’s tuned to balance performance and power use. Key numbers below:
3770K: 4 GW jobs = 10810 sec @ 118 watts;
3770K: 8 GW jobs = 14943 sec @ 128 watts;
I was surprised by the voltage (1.15v) I had to apply to run at 4.0GHz on the 3770K, reputable review sites seem to use a much lower voltage, but when I set the initial voltage at 1.10v, I got an immediate blue screen. Maybe the review sites had hand-picked samples.
Notice the much larger difference between stock and tuned for the 2600K versus the Ivy Bridge 3770K deltas. Excuse the mishmash of clock frequencies in the charts, but it’s been busy this spring.
As a side note, I've added a nVidia 550 Ti to this box to process the BRP4 jobs. My efficiency estimate for this configuration, 4.0 GHz @ 1.15v, 4 CPU jobs and 1 GPU job, is 27000 RAC / ((201 watts/1000) * 24 hours) = 5597 RAC / KWh. Compare that against the 3779 RAC/KWh rating from the chart above for just the CPU.
Interesting data--thanks for sharing it. I have one speculation and one nit-pick.
Speculative Interpretation: The rather large separation of your stock and tuned curves for the 2600K hints strongly that your particular sample of 2600K had a great deal of speed/voltage margin running at stock. Unless your 2600K has a nasty minimum voltage problem in some circuit, it seems likely you could alternately operate it at a quite substantial undervolt at stock speed (which would also give a nice efficiency improvement over stock). It seems likely your particular 3770K sample had rather a lot less margin at stock conditions. Various web site postings seem consistently to call the Ivy Bridges poor overclockers, which may mean that the real speed capability of the parts is less generous compared to the desired selling speed points than has been true since Conroe came out. (Lest we forget, the last few pre-Conroe generations were gasping to meet their spec points, and one did not routinely see successful massive overclock with little artificial aid).
Robert wrote:
3779 RAC/KWh
Nit-pick: RAC is a rate, while KWh is not. So to get a pure ratio, one might like either to use rates, as in RAC/KW, or else to use time-integrated quantities, as in Credit/KWh.
@ Robert
The ASUS board isn´t overclocking the cpu but the cpu itself. The i7 3770K has a much higher turbo multi than the i7 2600K and also a much higher tdp headroom. The i7 2600K can only get 3.5GHz on all cores (multi +1) but the i7 3770K can go to 3.7GHz (multi +3) (everthing on stock).
Because of the low tdp of the i7 3770K it will always run on 3.7GHz as long as the temperature is below 80-85°C.
@ Robert
The ASUS board isn´t overclocking the cpu but the cpu itself. The i7 3770K has a much higher turbo multi than the i7 2600K and also a much higher tdp headroom. The i7 2600K can only get 3.5GHz on all cores (multi +1) but the i7 3770K can go to 3.7GHz (multi +3) (everthing on stock).
Because of the low tdp of the i7 3770K it will always run on 3.7GHz as long as the temperature is below 80-85°C.
I agree, my i7-2600 and can run 1 core @ 3.8Gig, 2 @ 3.6, but all 4 (HT=ON), 3.5 max., with stock settings.(INTEL DP67BG mobo)
Have 2 systems, older, C2Q|X(treme)Q6600+GTX470. and Host 2. . The Q6600+470 has a COOLERMASTER ??? housing and gets far to hot.
Second has no casing, optimal cooling for the CPU (3.56GHz.) and 480 GPU.
Second host, CPU + 480GPU, has the best efficiency. (So much for PC-Casings?!)
Both have an ASUS P5E mobo and a 650 Watt PSU.
(Can't get the url on the i7-2600 + 2 ATI 5870 GPUs, SETI's down for maintenance).
I also use Kill - a - Watt meters, on 2 rigs, the fatest ones (2).
@Fred
Your main contribution is done by the gpu and the Q6600 (when it isn´t overclocked) is just limiting your gpu. Try overclocking the Q6600 and results should be quite similar. 3.6GHz shouldn´t be a problem with a good air cooler.
I built a fusion reactor in
)
I built a fusion reactor in my garage that uses annoying neighborhood children and pets as fuel.
Is this with monitors off at
)
Is this with monitors off at all times? (except when checking tasks)
Instead of the old "blank" SS?
Looking at some of these shows me I have paid for quite a bit of power since 1999
Of course the nVidia cards are hungry critters.
RE: I built a fusion
)
I use the Tesla Coil for trespassers at my place.
Just stumbled across this
)
Just stumbled across this thread. I have yet to get a Kill-o-watt type of reader but one of the first things I did when I built my current workstation is go into the bios and drop the voltage little by little on the CPU (testing stability in between). I managed to shave off a good ~30 Watts during peak (which is what it runs 24/7) by doing that according to HWMonitor.
I also tightened the memory timings while I was in there, and with relatively cheaper memory have been running stable undervolted and at 7-7-7-22
This AMD rig is still leaps and bounds behind the Intel side of things when it comes to efficiency at load though and thus my next addition to my BOINC fleet will be an i2500k system. I hope AMD can improve their architecture and keep Intel competitive.
Watch for the ARM RISC cpus
)
Watch for the ARM RISC cpus which HP shall use in its Moonshot supercomputer. They use very little power.
Tullio
For this set of efficiency
)
For this set of efficiency charts, I thought I would also post curves for stock processors comparing the new Ivy Bridge versus Sandy Bridge.
Here are some key stock numbers for the S6LV Gravity Wave (GW) jobs:
2600K: 4 GW jobs = 13100 sec @ 140 watts;
3770K: 4 GW jobs = 12230 sec @ 112 watts;
2600K: 8 GW jobs = 20093 sec @ 153 watts;
3770K: 8 GW jobs = 16230 sec @ 122 watts;
The stock i7-3770K is running on an ASUS board which seems to automatically overclock without user intervention and CPU-Z reports that it is running at 3.7 GHz versus the stock i7-2600K running at 3.4 GHz. Most components were the same between configurations, the key difference was the use of Windows 7 on the [3770K +Z77 MB] versus Linux for the [2600K +H67 MB], turbo disabled.
I also plotted the efficiency curve for the CPU’s tuned to balance performance and power use. Key numbers below:
3770K: 4 GW jobs = 10810 sec @ 118 watts;
3770K: 8 GW jobs = 14943 sec @ 128 watts;
I was surprised by the voltage (1.15v) I had to apply to run at 4.0GHz on the 3770K, reputable review sites seem to use a much lower voltage, but when I set the initial voltage at 1.10v, I got an immediate blue screen. Maybe the review sites had hand-picked samples.
Notice the much larger difference between stock and tuned for the 2600K versus the Ivy Bridge 3770K deltas. Excuse the mishmash of clock frequencies in the charts, but it’s been busy this spring.
As a side note, I've added a nVidia 550 Ti to this box to process the BRP4 jobs. My efficiency estimate for this configuration, 4.0 GHz @ 1.15v, 4 CPU jobs and 1 GPU job, is 27000 RAC / ((201 watts/1000) * 24 hours) = 5597 RAC / KWh. Compare that against the 3779 RAC/KWh rating from the chart above for just the CPU.
Interesting data--thanks for
)
Interesting data--thanks for sharing it. I have one speculation and one nit-pick.
Speculative Interpretation: The rather large separation of your stock and tuned curves for the 2600K hints strongly that your particular sample of 2600K had a great deal of speed/voltage margin running at stock. Unless your 2600K has a nasty minimum voltage problem in some circuit, it seems likely you could alternately operate it at a quite substantial undervolt at stock speed (which would also give a nice efficiency improvement over stock). It seems likely your particular 3770K sample had rather a lot less margin at stock conditions. Various web site postings seem consistently to call the Ivy Bridges poor overclockers, which may mean that the real speed capability of the parts is less generous compared to the desired selling speed points than has been true since Conroe came out. (Lest we forget, the last few pre-Conroe generations were gasping to meet their spec points, and one did not routinely see successful massive overclock with little artificial aid).
Nit-pick: RAC is a rate, while KWh is not. So to get a pure ratio, one might like either to use rates, as in RAC/KW, or else to use time-integrated quantities, as in Credit/KWh.
@ Robert The ASUS board
)
@ Robert
The ASUS board isn´t overclocking the cpu but the cpu itself. The i7 3770K has a much higher turbo multi than the i7 2600K and also a much higher tdp headroom. The i7 2600K can only get 3.5GHz on all cores (multi +1) but the i7 3770K can go to 3.7GHz (multi +3) (everthing on stock).
Because of the low tdp of the i7 3770K it will always run on 3.7GHz as long as the temperature is below 80-85°C.
RE: @ Robert The ASUS board
)
I agree, my i7-2600 and can run 1 core @ 3.8Gig, 2 @ 3.6, but all 4 (HT=ON), 3.5 max., with stock settings.(INTEL DP67BG mobo)
Have 2 systems, older, C2Q|X(treme)Q6600+GTX470. and
Host 2.
. The Q6600+470 has a COOLERMASTER ??? housing and gets far to hot.
Second has no casing, optimal cooling for the CPU (3.56GHz.) and 480 GPU.
Second host, CPU + 480GPU, has the best efficiency. (So much for PC-Casings?!)
Both have an ASUS P5E mobo and a 650 Watt PSU.
(Can't get the url on the i7-2600 + 2 ATI 5870 GPUs, SETI's down for maintenance).
I also use Kill - a - Watt meters, on 2 rigs, the fatest ones (2).
@Fred Your main contribution
)
@Fred
Your main contribution is done by the gpu and the Q6600 (when it isn´t overclocked) is just limiting your gpu. Try overclocking the Q6600 and results should be quite similar. 3.6GHz shouldn´t be a problem with a good air cooler.