Help needed - M77 232 917 celebration build

[kladner]

6700k (3200 DR ram) - 1.85 results per day (this is my main system which is in daily use, so may have impacting timings)

How is the ideal ram bandwidth determined? The system above is the same as mine, except I am running the CPU at 4300 MHZ.

[jbpace]

Originally Posted by mackerel

What I figured out was:

Score = (ideal ram bandwidth in GB/s) / (all core AVX active clock in GHz * number of cores * CPU correction factor)

Ideal ram bandwidth is approx. rated MT/s * channels * 8 / 1000
CPU correction factor is 1 for Skylake+, 0.88 for Haswell, 0.82 for Broadwell, 0.58 for Sandy Bridge, and I'm assuming 0.5 for Ryzen but not tested in depth.

If I'm correctly scoring a stock i9-7900X @ 3.3 GHz paired with quad channel DDR4-3200:

score = [ 3200 * 4 * 8 / 1000 ] / [ 3.3 * 10 * 1.0 ]
score = 102.4 / 33.0
score = 3.10

Which mackerel finds to be about 90% the performance of unconstrained memory.


Applying that scoring method to Mystical's OC'd test run:

Originally Posted by Mysticial

Core i9 7900X @ 4.0 GHz AVX (11% overclock)
Quad Channel DDR4 @ 3200 MT/s (20% overclock)

I don't have the bandwidth graph on hand to show it. But basically, the system was maxed out at 70+ GB/s for the entire test without ever dropping below that.

score = [ 3200 * 4 * 8 / 1000 ] / [ 4.0 * 10 * 1.0 ]
score = 102.4 / 40
score = 2.56

Where mackerel's testing showed things "... dropped off quite steeply below 3." Assuming an i9-7900X can typically be OC'd to 4GHz, quad channel memory would need to be at least DDR4-3733 to score 3 (2.99, close enough ;-).


Since I need PCIe lanes for other uses (video coding is a bear!), I'll just order a 7900X. Thanks for keeping me from wasting $400!!

[jbpace]

Originally Posted by kladner

How is the ideal ram bandwidth determined? The system above is the same as mine, except I am running the CPU at 4300 MHZ.

An i7-6700K @ 4.3GHz and dual channel DDR4-3200 RAM would be:
score = [ 3200 * 2 * 8 / 1000 ] / [ 4.3 * 4 * 1.0 ]
score = 51.2 / 17.2
score = 2.98

You're also getting about 90% as much production as you could with unconstrained memory.

[kladner]

Originally Posted by kladner

How is the ideal ram bandwidth determined? The system above is the same as mine, except I am running the CPU at 4300 MHZ.

Doh! I finally spotted the information in CPUZ.
EDIT: (I have now seen your response. I really appreciate your walking me through this). It seems, then, that I might be wasting energy with the CPU OC, if I understand the score correctly.

[LaurV]

You (general you) always waste energy with overclocking. CPU, GPU, running horse, whatever, it does not matter. You get the tings done faster (like in a little bit faster) with more energy expense (like in a lot more energy). This was discussed (with numbers) many times here around. My GPUs will work 20% or 30% faster if I overclock them, but that will be spending 50%, or 80% or even more energy. Beside of the little parasitic capacitance coming to effect (high frequency passes through small capacitors, so if you have few parallel wires driving your CPU clock, they act as a short-circuit, producing heat, if you increase the clock enough), you also have to invest higher power in cooling, etc. The waste of the power grows exponentially, if the clock increases linear, after a limit.

[ixfd64]

Originally Posted by LaurV

You (general you) always waste energy with overclocking. CPU, GPU, running horse, whatever, it does not matter. You get the tings done faster (like in a little bit faster) with more energy expense (like in a lot more energy). This was discussed (with numbers) many times here around. My GPUs will work 20% or 30% faster if I overclock them, but that will be spending 50%, or 80% or even more energy. Beside of the little parasitic capacitance coming to effect (high frequency passes through small capacitors, so if you have few parallel wires driving your CPU clock, they act as a short-circuit, producing heat, if you increase the clock enough), you also have to invest higher power in cooling, etc. The waste of the power grows exponentially, if the clock increases linear, after a limit.

Not to mention that overclocking increases the chance of getting bad LL results.

[preda]

IMO:
take care to use 4 sticks of RAM, not 2.

You might save some money without losing LL performance by going with 16GB (i.e. 4 x 4GB) instead of 32GB (4 x 8GB) total.

[mackerel]

Overclocking doesn't necessarily have to negatively affect power usage... overall we should be looking at performance per watt of the system, and a faster CPU can help offset static power consumption elsehwere. Also if you overclock without changing the CPU voltage, you only scale for clock, not voltage, so it is effectively free.

Also on my scoring, now I've looked it up again, I got the numbers a bit offset.
6 - 98%
5 - 95%
4 - 90%
3 - 82%
2 - 65%

There is quite a bit of noise, particularly at the lower % area, so don't take these numbers too exactly. I'm sure there are factors not considered into this that will have an impact. Also note I'm using the ideal ram bandwidth as indicator. I tried to correlate the performance with measured ram bandwidth (aida64) but didn't see a good relationship there. I know I'm missing something... I'd speculate something to do with the ram access pattern behaves differently under different configurations.

[bgbeuning]

Originally Posted by jbpace

The motherboard has already been sourced, and I'm not inclined to swap (unless there's a really compelling reason), so assume an ASRock X299 Taichi XE.

Most chip sets on most motherboards support 2 memory channels.
The LGA 2066 on your motherboard support 4 memory channels.
That should mean it can handle twice as many cores before memory
is the bottleneck.