George's dream build

[Prime95]

Originally Posted by rleshane

My question (which I'm not sure if anyone can answer) is would the difference in ram (2133 to 3000) yield 15% faster processing (for example on LL tests)? If so, then it seems it would be worth the faster ram.

3000 vs 2133 is 40% faster RAM. Will that give a 15% increase in throughput? If you are running 4 workers, estimates from my Haswell and Sandy Bridge benchmarks are you will see a 17-18% performance boost.

Prime95 is the rare application that does gain significantly from memory overclocking.

[Madpoo]

Originally Posted by Cruelty

Out of curiosity my home-built Xeon E5 v3 @ 2.5GHz 10c/20t (105W) + 4x4GB 2133 DDR4 SR ECC-R. Below excerpt from FullBench:

Timings for 4096K FFT length (10 cpus, 10 workers): ...  Throughput: 283.06 iter/sec.
...
Timings for 4096K FFT length (10 cpus hyperthreaded, 10 workers): ...  Throughput: 262.55 iter/sec.

Looks like the hyperthreading is indeed slowing things down, as would be expected.

[Fred]

George, thought you would find this particularly interesting since it sounds like we're pursuing similar builds around the same time. I finished my second build and found some interesting info.

Both builds: Mini-ITX with 80+ Bronze PSU

Build 1: i5-6500, 2x4gb DDR4-2133

Bench:
Timings for 4096K FFT length (4 cpus, 1 worker): 6.37 ms. Throughput: 156.97 iter/sec.
Timings for 4096K FFT length (4 cpus, 2 workers): 12.66, 13.04 ms. Throughput: 155.68 iter/sec.
Timings for 4096K FFT length (4 cpus, 4 workers): 25.68, 25.71, 24.64, 24.65 ms. Throughput: 159.00 iter/sec.

Build Cost: $356.83 http://pcpartpicker.com/p/N2rMBm

Watts at wall: 71 (full throttle)

Build 2: i5-6600, 2x4gb DDR4-3200

Bench:
Timings for 4096K FFT length (4 cpus, 1 worker): 5.42 ms. Throughput: 184.62 iter/sec.
Timings for 4096K FFT length (4 cpus, 2 workers): 10.76, 10.65 ms. Throughput: 186.78 iter/sec.
Timings for 4096K FFT length (4 cpus, 4 workers): 21.29, 21.30, 22.19, 21.20 ms. Throughput: 186.16 iter/sec.

Build Cost: $439.84 http://pcpartpicker.com/p/7WrMBm

Watts at wall: 93 (full throttle)

So, going from the 6500 DDR4-2133 to the 6600-DDR-3200 gives me:

Upfront Cost: $356.83 to $439.84 (23.3% increase)
Power Consumption: 71W to 93W (30.1% increase)
Highest Throughput: 159.00 to 186.78 (17.5% increase)

Not what I was hoping for, but it's looking like if I'm going to add a few more systems (which I plan to), the i5-6500 with DDR4-2133 may be the best bang for the buck.

[Prime95]

Originally Posted by Fred

Not what I was hoping for, but it's looking like if I'm going to add a few more systems (which I plan to), the i5-6500 with DDR4-2133 may be the best bang for the buck.

Great info!! The big surprise for me is the difference in power consumption. It is more than enough to make the cheaper system the better choice.

Two questions if you don't mind:

1) Can you enable the "DDR4 non-Z OC" feature on the H110 board and does that make a difference in throughput and power consumption?
2) If you put the i5-6500 in the Z170 board, what is the throughput and power consumption?

Note: The throughput numbers are less than I was expecting when I was extrapolating from Haswell/DDR3 numbers. I had estimated 197 vs. 189 and 165 vs. 159. It is my understanding that DDR4 has worse latency than DDR3 and perhaps that explains my optimistic estimates.

[Prime95]

A couple other things:

1) Did you try the included Intel heatsink/fan? If so, what were the temps? What are the temps now?

2) Can you tell if there is any turbo boost when prime95 is running 4 workers?

[Mark Rose]

I am also curious how the i5-6600 compares to the i5-6500 with DDR4-2133 RAM.

[Fred]

Originally Posted by Mark Rose

I am also curious how the i5-6600 compares to the i5-6500 with DDR4-2133 RAM.

This one I can easily answer because I already tested it. I didn't save the results, but there was virtually no change in throughput. If limited to 2133 ram, I definitely think it's worth saving a few bucks and going with the 6500.

[Mark Rose]

Originally Posted by Fred

This one I can easily answer because I already tested it. I didn't save the results, but there was virtually no change in throughput. If limited to 2133 ram, I definitely think it's worth saving a few bucks and going with the 6500.

That makes me wonder how the 6400 would compare. The 6500 gets a quad core boost clock of 3.3 GHz. The 6400 gets a quad core boost of 3.1 GHz. It should be possible to lock the 6500 or 6600 at 3.1 GHz to make a comparison.

If there's no difference at 3.1 GHz, that would be another $15 in savings per system.

[Fred]

Originally Posted by Prime95

A couple other things:

1) Did you try the included Intel heatsink/fan? If so, what were the temps? What are the temps now?

I didn't try the stock cooler. I had been having heat problems in my desktop computer with Prime95 running (i5-4670k - not OC - clean - temps hitting 90c). I put a cooler master on it and it brought my temps into the 50s. Right now (with the cooler master on both mini-itx setups) they are both running in the 40s. Since I want them to run flawless for many years, I just sprang the extra $25 for the coolers, but I expect I could have probably gotten by with the fan Intel ships.

Originally Posted by Prime95

2) Can you tell if there is any turbo boost when prime95 is running 4 workers?

Yes, but not to the full "turbo" rate. The 6500 is listed at 3.2GHz, turbo of 3.6, but is running at 3.3 under prime (very steady and consistent 3.3). The 6600 is listed at 3.3, turbo of 3.9, and is running 3.6 under prime (again, very steady 3.6). These numbers are the same with 4cpu on 1 worker and 4cpu on 4 worker.

[Fred]

Originally Posted by Prime95

Note: The throughput numbers are less than I was expecting when I was extrapolating from Haswell/DDR3 numbers. I had estimated 197 vs. 189 and 165 vs. 159. It is my understanding that DDR4 has worse latency than DDR3 and perhaps that explains my optimistic estimates.

Timings on my DDR4-2133 are 15-15-15-35. The DDR4-3200 are 16-18-18-38.

[Fred]

Originally Posted by Prime95

Two questions if you don't mind:

1) Can you enable the "DDR4 non-Z OC" feature on the H110 board and does that make a difference in throughput and power consumption?
2) If you put the i5-6500 in the Z170 board, what is the throughput and power consumption?

I'm curious about these myself and will check them out as soon as I have the time. Right now though I better get some work done. Have been playing in Prime-land all day. Keeping job = disposable income = more systems = discovery of 50th Mprime.