Phase Vapor Super Cooing, what it gets you

Matt Simis

For those considering more extreme cooling, the following maybe useful:

Ive just finised testing two Prometeia (www.chip-con.com) Super Coolers on both Intel P4 and AMD Athlon platforms:


Intel P4 2.26 (533mhz FSB based) overclocked to 3280mhz.
Sisoft Sandra CPU bench, MIPS: 6269
Multimedia Bench: 12809

AMD Athlon TB (A) 2200 overclocked to 2310mhz. (XP2920)
Sisoft Sandra CPU bench, MIPS: 6363
Multimedia Bench: 12666


The P4 system scores higher (10%) in Q3 and 3DMark. The P4 is running at 1.92V and AMD 2.0V. Temperatures in the compressor unit are -29C at the highest (under load) on the AMD system and -34C on the P4 system. CPU temps on both systems peak at +12C under load. Dropping voltage on either system would drop the peak Temp to 0C.



Matt

mackanz

Matt,

Would you sell one of them?

They are 1 fine piece of kit.

Mac

Matt Simis

Lol, they are mere weeks old and cost me an arm and a leg to source and purchase!

Offer me 1000 and we'll talk.




Matt

mackanz

Thanks for the quick answer, but they go for 800 shipped as new, even less.

How did you find them?
Temps on the amd setup?

Mac

Matt Simis

They do indeed go for about that, but if I sold you mine, Id have to replace it and make it worth my while.


The first one I got was a first gen model from a Dutch Distributor. It had poor insulation and was damaged in shipping. I would recommend you buy direct from Chip-Con if possible or from a Disti you know gets stock updated regularly.


As I only have the CPU in-socket probe for temps on the AMD, I generally refrain from putting numbers on the temps. Highest under load temp I saw was around +7C, which I imagine was 0 or below on the CPU die itself. Reports on the internet would seem to confirm this. I intend increasing the pressure on the P4 clamp tonight, to reduce temps further. The reviews I have read claimed better performance than my current setup.


Personally I was more impressed by the AMDs overclock. This is the "bad" TBred, the one some reported overclocked to 1888 max. The 500Mhz+ increase on an AMD is quite impressive, as its competitive scores vs the P4 show. (The P4 is my main system btw). The next gen 2400 and 2600 Tbred Bs should do even better again.



Matt

mackanz

Looking 4-ward to those results, m8!

I know a few who runs the Prometeia on the Radeon, and Phase Change Chiller on the cpu and they have gotten excellent results.

A friend of mine runs his XP 1700 at 2.2GHZ at 2.15 V, nice overclock there!

Mac

Gerry

Excellent stuff matt, but a few questions. The amd overclock is more impressive, I'd agree there. The p4 is going to be pretty stretched on the old fsb front (192mhz), ( well not the chip, but the board ) What platform are you running it on? And what timings for the ddr? If its anything less than fastest timings, its going to be really starved of bandwidth.

From a pure cpu speed point of view, you should be able to get more out of a 2.5ghz C1 stepping chip, approaching 4ghz I'd say.

mackanz

Up until now 4GHZ has only been achieved by LN2 cooling which is another ballpark. But still, 3.5-3.8 is in the range.

Mac

Matt Simis

I increased the pressure on the P4 CPU, load temps are now +4C.

You're correct Gerry, the motherboard is getting very stretched at this point. Its still the trusty old BD7-II. The ram Im using is TwinMos 3200 DDR (Komplett), which I can honestly say is the best ram Ive used (I tried Samsung, Corsair and Crucial... Crucial have sunk waaay low in my books after trying several sticks). The settings are the fastest bar one, CAS 2, 6-2-2 timings (fastest is 5-2-2 on this board). Memory is running at 1:1 ratio, the next one up would require DDR 450. Little bit more memory speed would be nice..

Im hoping to move over to a C1 CPU with a higher multi (this one is only 17x!), then I could use a higher memory ratio too.

Mackanz, 3.8GHz is approching 4GHz!



Matt

Gerry

Yeah, as matt says I wasn't saying 4ghz. 4ghz would require the extra "kick" of LN2 I'd say. What you are really up against here is the curve of max speed versus cooling/voltage.
As you get closer to absolute max speed, you are throwing more and more cooling and/or voltage for every 100 mhz. The curve has gone from being close to linear to being exponential.

A new stepping for the chip can give different kinds of results. It could purely be to fix yield problems, say there are a few known problem areas which are marginal, it could fix them, and so bring up the average speed of the chips. However it might not raise the max speed of the whole chip, so you might find that ( rough example ):

More chips make it to a 50% overclock, but even with LN2, the absolute top end is still the same for a good chip out of this bunch.

Theres another, more involved stepping change which affects the whole chip, and can probably raise the max speed the chip will do overall, with every type of cooling.

This seems to be what amd did with the B stepping of the thoroughbred. It may also be what intel did with the c1 stepping PIV, but we will have to see.

Its quite possible that a C1 with LN2 may not go any higher than 4ghz.
What is fairly certain though, is that 3.2 ghz is possible with the Stock intel cooler !!! . With an alpha, 3.4ghz stable has been seen, which is rather amazing for air cooling.

Matt, I'd say you are actually getting pretty good memory bandwidth. If you get another chip with a higher multiplier, you will be worsening the ratio of cpu speed to memory bandwidth. I have heard rumours of some samsung memory being able to do 220mhz ddr on fast settings, however it was supposed to vary wildly between batches, and I'm not sure the bd7-II would be capable of that. Does your cooling system cover the motherboard chipset?

mackanz

What i meant was that it is a hell of a difference between 3.8 and 4 ghz when it comes to overclocking.

So far, only a Few finns and a Japanese or 2 has achieved that with LN2/R-chiller in the system. 3.2-3.4 is a number that almost all Intel user can hit theese days and amd is clearly slipping longer and longer for each days that goes by without the T-bred "B". The question is, are they even going to release that chip now that they will announce the 166 version?

Mac

Gerry

Thats what I was implying. Though the mid 3ghz range is getting much easier, its debatable as to whether the absolute max limit will increase.

Matt Simis

Originally posted by Gerry
Matt, I'd say you are actually getting pretty good memory bandwidth. If you get another chip with a higher multiplier, you will be worsening the ratio of cpu speed to memory bandwidth. I have heard rumours of some samsung memory being able to do 220mhz ddr on fast settings, however it was supposed to vary wildly between batches, and I'm not sure the bd7-II would be capable of that. Does your cooling system cover the motherboard chipset?

The cooling system does not cover the chipset, although I have lapped the Intel Chipset cooler and added a 12v Fan. The Samsung 2700 ram I had (I tried 2 sticks) all got to 220 with the slowest settings. The Twinmos 3200 on the other hand reaches 220 easily, with good memory timings (all at 2.7v). I got the 512MB stick from Komplett, fairly cheap. I have tested the BD7 and 2 BD7II's (three boards in total) upto 195mhz FSB and 225mhz ram, they seemed ok. I expected as much however, being an Intel chipset.

My plan is to run a 2.5GHz C1 at 25x multi, 151MHz FSB, 204MHz Memory, for just over PC3200 speeds. This is a slight improvement from what I have now.



Matt