cpu speed question

[tim]

So, I took guru's and speedraider's advice and bought a i5 4690. Except when it arrived it was a 4590. That's what I get for dealing with a minor name, it was antonline.com through newegg.com's website. I've always liked newegg, but this reseller lied about the cpu.

Anyway, it was pretty close in performance, so I set it up and ran it. It's a Dell, runs nice, very quiet, too quiet. Running prime95 with 4 cores at 100%, it's doing high 30s ms/iteration. My i5 2400 is low 20s ms/iteration, my athlon II x645 is high 40s. This i5 4590 should be high teens.

I put the win 7 performance settings all to high, I tried speedstep and turboboost off and on. No difference. cpuid hwmonitor and openhardwaremonitor both look normal, except the temps are way low for 100% load, AND the cpu is running at about 40 watts. Should be over 80. The fans are running, but slowly. On boot the cpu fan does speed up for a moment, so I don't think it's faulty. If it were it would heat up, I'd think. Here's what it looks like with both monitor programs. Any ideas?

[guru]

First mistake is getting anything Dell. They like to cripple perfectly good hardware to save a buck. It is possible the memory is holding the system back. Try some other benchmarks to see. Give rc5 a try as it is all CPU and no dependency on memory.

[pinhodecarlos]

Or you can try to run enigma, Helix is needing some help in there for FreeDC.
http://www.enigmaathome.net/

Optimized clients:

http://www.enigmaathome.net/forum_th...tyle=8&start=0

[tim]

I've not used boinc before, I installed and chose enigma, it wants an email and password. I pm'ed Helix through the free-dc pm function to see if he wants to share that info.

The hardware monitors both show the memory as ddr3 1600. It's weird that the cpu is running on 40 watts but showing full frequency. It could be a bug in the hardware monitoring programs that don't read Intel chip temps right, but the i5 2400 I have does show 80+ watts with the same programs.

[guru]

I would run some benchmarks on it and compare it to others to get a feel for the system. Can you post a link to the system you purchased? Also run the SOB benchmark and post the numbers.

[tim]

The box is http://www.newegg.com/Product/Produc...9SIA24G2HZ2404 Mine is the only review. I'm not sure what you mean by Sob benchmark. Here's what three boxes running Sob are doing. Top one is the system in question.

[guru]

The SOB client(mprime.exe) has a benchmark mode built into it. It runs different numbers of threads and different FFT sizes. That is what I used to get the data from my systems to compare the actual performance. Different work downloaded has different FFT sizes and can affect the time reported.

[tim]

Thanks so much for your help on this. I have a lot to learn.
Okay, here's the link to the page of previous benchmarks for this cpu: http://www.mersenne.org/report_bench...ic_cpu=4377447

and here's the mprime benchmark output from mine. I'm not sure how to interpret it. Do I compare the "best time for" section to the benchmarks, or the "Timings for " section, looking at 4 cpus running 4 workers?
And for what's it's worth, under linux, the screen output for normal work for FFT 2M, pass1 256, pass2 8K (all 4 cores the same) was from 31 to 39 ms/iteration.
results.txt

[tim]

Thanks so much for your help on this. I have a lot to learn.
Okay, here's the link to the page of previous benchmarks for this cpu: http://www.mersenne.org/report_bench...ic_cpu=4377447

and here's the mprime benchmark output from mine. I'm not sure how to interpret it. Do I compare the "best time for" section to the benchmarks, or the "Timings for " section, looking at 4 cpus running 4 workers?
And for what's it's worth, under linux, the screen output for normal work for FFT 2M, pass1 256, pass2 8K (all 4 cores the same) was from 31 to 39 ms/iteration.
results.txt

Okay, more data. I've downloaded Intel Extreme Tuning Utility, and it shows TurboBoost is disabled. But it's enabled in the bios, and Win7 power setting is on High Performance. So I don't get it. And yes, it is enabled in the bios. Any ideas?

Another edit: I reset the bios to factory settings, and under Intel XTU, it now shows TurboBoost is enabled. No difference in performance, though. I did the prime95 torture test modes, the highest heat mode got the watts up to 63 and the heat to 159, the highest power consumption mode was 44 watts and 129 F, the same as normal operation.

[guru]

The system could be throttling way too soon. Check the heatsink and paste. Possibly upgrade the cooling of the CPU and case. Try running cpu-z while the system is loaded and monitor the top speed of the processor.

[tim]

It's not heat throttling. I notice that at idle the processor frequency is 3.692 Ghz, which is "turboboost" for this cpu. When I turn on Prime95, the processor frequency drops to 3.29 Ghz. When I stop it, the processor frequency rises to 3.692 Ghz. Shouldn't it stay high under full load? You can see that the temps are not excessive and there is no thermal throttling. And the CPU Total TDP doesn't go above 45 watts, when under full load it should be 80+ watts. Under Intel Extreme Tuning Utility CPU stress test, the frequency also drops to 3.46 Ghz and the watts range between 45 and 49, with the temperatures between 129 and 135 F. Here's a screen grab from Intel XTU of turning Prime95 off and on:

[tim]

Another piece of the puzzle: it IS hitting the current limit and throttling. Under Intel XTU, with full load, it shows Current Limit Throttling at 100%. With no load, it shows 0%. I think Current Limit Throttling occurs when the motherboard is sending current above specifications. Dell has replaced the motherboard (no effect) and sent a second, identical computer (no different). I've let them know about the current limit throttling, and that they need to either send a motherboard that doesn't cause throttling, or refund my purchase. Dell's support has been super, but I'm afraid my question won't reach someone with expertise to address this issue.

[tim]

To Dell's credit, they are going to RMA the computer and refund, even though it was sold by a third party, antonline.com through newegg. Excellent customer service from Dell, though.

It looks to me like the current limit throttling from the motherboard was designed by Dell to not run the CPU too hard to prevent wear/failure and prevent Dell service calls.

[guru]

Yep, typical Dell. Sell you something with high specs then run it under spec to prevent other flaws from showing. No matter how good Dell support is they will never make up for the inferior products Dell sells.

[guru]

Tim, did you get the replacement? How is it doing?

[tim]

The replacement was exactly the same. They were both designed to not allow the cpu to run full-on for any amount of time. Other than that, nice computers. Did not fit my needs at all for crunching.
So I'm building a i5-4690k as was suggested. The first mobo was doa, waiting for replacement. I'm going to run linux on it. Time to brush up on my command line. And see if I can get it to work over my windows network here at home. I want to log in from my windows box and see the log file to look at progress. I have done it before, but that was years ago. Should be fun.

[tim]

Folks, I've built a few Intel core i5 4690s and I've noticed something odd:

days between finishing P-1 test and then finishing the PRP test and reporting "x is not prime" and then getting a new test:
Intel core 2 duo, 26 days
Athlon II x4 645 3.1 ghz, 18 days
Intel core i5 4690 3.5 ghz, 16 days
Intel core i5 2500 3.3 ghz, 9 days

Why is the i5 2500 faster than the i5 4690?

Whether linux or windoze, about the same speed. There's 3 or 4 boxes of each type cpu, the number of days is an average for that cpu. I've not done anything to tweak the prime95/mprime client. All cpus report 100% use, one client working per core. None are overclocked. Prime95 reports using FMA3 instruction set on i5 4690 while the i5 2500 reports using AVX. The 4690 was way more expensive to build, I would have thought they would be my fastest boxes. What is wrong here? What do I have set up wrong? The Dell I returned (4690) ended up being the same speed as my hand built 4690, so I don't think it was Dell. The problem must be between the keyboard and the chair. But I don't know what.