CPU prices starting to fall ( for some processors )

[Mark Rose]

Originally Posted by mackerel

The 8100 has 4 cores at 3.6 GHz, although I think we're going to be somewhat ram limited.

I don't think by much. My I underclocked my i5-6600's to 3.3 GHz to match the dual rank DDR4-2133 I gave them.

Coffee Lake supports DDR4-2400, and with some IPC gains, I think that will match nicely with the 3.6 GHz clock.

[mackerel]

If not limited elsewhere, there is no IPC difference between Skylake, Kaby Lake, Coffee Lake when it comes to FMA3 type tasks. That is, if you run small tasks one per core, it scales pretty much ideally.

My basis for being concerned about ram is that only my fastest ram Skylake system comes close to being practically unlimited. That was when it was running 4 cores at 4.2 GHz, with dual channel dual rank ram at 3200. Other similar clocked Skylake systems but with slower ram do measure below that. I suppose it comes down to how much difference it makes, since it doesn't scale linearly. I think I should repeat my earlier testing given both updated software, and I also have more recent hardware to try it on. I recently got some Samsung B-die ram despite its price premium. It is only single rank but I've tested stability to 3600C16. I've not managed to push it further yet. Not bad for a 3000 rated kit!

[kladner]

That was when it was running 4 cores at 4.2 GHz, with dual channel dual rank ram at 3200.

That reflects my current setup with a 6700K @4.3 GHz, with dual channel, dual rank ram at 3200MHz. I would have to look at benchmarks to see if there is a fall-off with the 4th core, but it has always delivered excellent timings in DCLL. (In the 40M range, with all four cores on one worker, it was hitting as low as 1.8 ms/it, if other continuous loads such as system monitors and multiple active browser tabs were shut down.)

[Mark Rose]

Originally Posted by kladner

That reflects my current setup with a 6700K @4.3 GHz, with dual channel, dual rank ram at 3200MHz. I would have to look at benchmarks to see if there is a fall-off with the 4th core, but it has always delivered excellent timings in DCLL. (In the 40M range, with all four cores on one worker, it was hitting as low as 1.8 ms/it, if other continuous loads such as system monitors and multiple active browser tabs were shut down.)

Originally Posted by mackerel

My basis for being concerned about ram is that only my fastest ram Skylake system comes close to being practically unlimited. That was when it was running 4 cores at 4.2 GHz, with dual channel dual rank ram at 3200.

So those numbers would indicate I'm still a bit RAM starved:

2133 DDR4 bank / 3.3 core GHz = 646 (me)
3200 DDR4 bank / 4.3 core GHz = 744 (kladner)
3200 DDR4 bank / 4.2 core GHz = 761 (mackerel)

I should see if I can underclock those systems further.

Other similar clocked Skylake systems but with slower ram do measure below that. I suppose it comes down to how much difference it makes, since it doesn't scale linearly. I think I should repeat my earlier testing given both updated software, and I also have more recent hardware to try it on. I recently got some Samsung B-die ram despite its price premium. It is only single rank but I've tested stability to 3600C16. I've not managed to push it further yet. Not bad for a 3000 rated kit!

Nice!

[mackerel]

I can't find the chart now, but I did create a total core GHz vs ram bandwidth efficiency chart from past results. The main take away point was that the shape looks like it could be scaled from tanh function. If you have unlimited ram bandwidth, you're CPU limited. If you have unlimited CPU, you're ram limited. Where we actually are is the in-between part.

Thinking more, is this a case where Amdahl's law might apply? To do a certain amount of work, you need X amount of CPU time, and Y ram transfer time. To optimise productivity, Y needs to be sufficiently small compared to X. I'll have to play about with this later and see if that works out more intuitive than tanh function...

[tServo]

Originally Posted by ewmayer

Sticking to AVX2-capable Intel chips, I get around half the per-cycle throughput on Ryzen vs Intel but Ryzen has double the core count, so the question boils down to total cost of ownership for the hardware + electricity. How does Ryzen's total power draw compare to a current Intel quad?

Interesting question, on hardwarezone.com I found a comparison article, 7700k vs 1800x, that showed @ idle, 1800x = 48w; 7700k was 51w. @ load, they were close again, the 1800x was sucking 276w & the Intel was 283w.
Here's the link:http://www.hardwarezone.com.sg/revie...up-performance

[M344587487]

Big price drop on 1700X and 1700 on ebuyer in the UK (down from £305 to £227 for 1700X, £265 to £207 for 1700). I got a 1700X to play with, will probably underclock and/or undervolt to see how efficient I can get it.

[M344587487]

Originally Posted by M344587487

Big price drop on 1700X and 1700 on ebuyer in the UK (down from £305 to £227 for 1700X, £265 to £207 for 1700). I got a 1700X to play with, will probably underclock and/or undervolt to see how efficient I can get it.

Scratch that, they decided to cancel my order. Thanks ebuyer.

[kladner]

Originally Posted by mackerel

On Skylake-X delid, if you don't mind doing it yourself, the tools are not that expensive. In the US I'd suggest the Rockit 99 for $40 (+ whatever liquid metal TIM you prefer) https://rockitcool.myshopify.com/pro...akex-kabylakex

I bought their Rockit 88 for 115x CPUs in the past and it has served well.....

BTW my Skylake-X 7800X seems P95 stable at stock voltage and all cores at 4.3 GHz, and that'll probably remain my standard overclock. The delid will allow me to explore what sort of clocks I can get for non-AVX workloads with elevated voltage. I have benchmarked at 4.9 GHz but that was into the thermal danger zone.

I may have said before, but I really appreciate this information. Is the gel super-glue on the Rockit site the sort of thing one should use to reattach the IHS?

[mackerel]

Originally Posted by kladner

Is the gel super-glue on the Rockit site the sort of thing one should use to reattach the IHS?

I've used super glue before on my 1st delid, but it was a bit messy and I don't bother now, relying on the socket retention mechanism to hold it together. High temperature silicone is popular if you did want to go that route.

Oh, the delid and liquid metal application on my 7800X was a little disappointing. Hottest core dropped 5C at 4.3 GHz 1.10v running small FFT, at same ambient. Was hoping for a bit more than that.

[kladner]

Originally Posted by mackerel

I've used super glue before on my 1st delid, but it was a bit messy and I don't bother now, relying on the socket retention mechanism to hold it together. High temperature silicone is popular if you did want to go that route.

Oh, the delid and liquid metal application on my 7800X was a little disappointing. Hottest core dropped 5C at 4.3 GHz 1.10v running small FFT, at same ambient. Was hoping for a bit more than that.

Thanks for the updates. I understand expecting a greater improvement, but even 5C would be welcome. It would get my 6700K@4.3GHz into the mid or lower 60s, depending on ambient.