Quote:
|
Originally Posted by Hoot
That's a great article, but it leaves me in a quandry about something. I ordered a G5 based upon PHaestus article showing it yielded CPU temps ~7.9-8.1C above water temp at the right flow rate. I contrasted that to what he said the Cascade yielded ~9.5-10C above water temp at the right flow. Now, reading the previously linked thread, Cathar says he got the following with the Cascade:
Who's water to CPU temperature differential is the right one for the Cascade?
|
Hoot, there are large differences between the two examples.
Phaestus is taking an actual CPU-die to water-in temperature delta based upon a T'Bred B CPU (~88mm^2). It has been postulated that while Phaestus has quite accurately calibrated the on-die diode of the CPU die, that the diode is actually located in a rather cool section of the CPU die (towards a corner would be my guess, just as is true of P4 CPU dies), so the reported temperatures are somewhat "compressed" because sections of the CPU die will be a lot warmer than what the on-die diode is reporting.
The numbers I was calculating was based upon a projected estimate C/W curve for a Cascade block given a 100W 10x10mm heat source derived from a combination of Phaestus's, BillA's, and my own data. The curve that I used was not
directly emperically obtained, but rather a theoretical curve that attempts to remove Phaestus's postulated temperature compression from the results he obtained, and my own results, and massage it into something that is somewhat comparable to BillA's old waterblock data. It is a fantasy curve though - make no mistake. For what it's worth, the results that I obtain for my own blocks, after a bit of pseudo-calibration with a 101mm^2 Barton CPU die, closely reflect the same flow vs C/W results that BillA obtained on his 100mm^2 CPU die test-bed, however the equipment I use is nothing in comparison to quality kit that Bill and Derek are using, so frankly I consider any correlation to myself and Bill's result to be more a result of pure chance than anything else.
The values I listed are the CPU die to ambient air differential, and include the pump heat effect, as well as the impact of the flow rate of each pump on the efficiency of the radiator (Thermochill 120.2 model was used) and the waterblock (based on the derived performance curve).
Given the results that both Bill obtained with the White Water, and what I measured, and again with what I measured with the Cascade, I would "suggest" that the values that I gave are perhaps closer to an accurate picture of what the hotter sections of a fairly heavily overclocked CPU die are approaching, but again, this is a mish-mash of extrapolations and corrections using woefully inadequate equipment, yet which somehow still manages to serve the purpose well enough for me to develop waterblocks with a fairly high degree of (perhaps misplaced) confidence.