LittleDevil's Rotary Cascade

[gosmeyer]

WOW, enough said!

[LittleDevil]

Awesome job LittleDevil! Have you tried a higher load ? Maybe 300W ~ quad core?

250W on that load tester is just enough for quard core..... i think that quard core doesn't produce 300W I will try to get one and test it.

And thaks for really nice comments guys

Regards

[Brettbeck]

250W on that load tester is just enough for quard core..... i think that quard core doesn't produce 300W I will try to get one and test it.

Regards

Ah right. Everyone seems to throw lots of wattage figures about saying quad core is this, that etc... so I still don't know what is right . Do you think it can hold under -100°c still with 300W?

[kayl]

Very nice work Little devil, your cascades just keep getting colder by the unit

Ah right. Everyone seems to throw lots of wattage figures about saying quad core is this, that etc... so I still don't know what is right . Do you think it can hold under -100°c still with 300W?

Very nice work Little devil, your cascades just keep getting colder by the unit

Ah right. Everyone seems to throw lots of wattage figures about saying quad core is this, that etc... so I still don't know what is right . Do you think it can hold under -100°c still with 300W?

Brettbeck there is a lot of vaibles to take into account when comparing load tester as heatloads. What thickness cold plate you use, if its a resistor load/ heater cartridge, how well the load tester was constructed, also if load tester is insulated or not, room temps, list goes on
On one of my load testers at 260w wasnt enough to simulate my QX6700 vmoded, on my new load tester 245w was as hot as a QX6700 @4.6ghz and 1.85v.

[Brettbeck]

Brettbeck there is a lot of vaibles to take into account when comparing load tester as heatloads. What thickness cold plate you use, if its a resistor load/ heater cartridge, how well the load tester was constructed, also if load tester is insulated or not, room temps, list goes on
On one of my load testers at 260w wasnt enough to simulate my QX6700 vmoded, on my new load tester 245w was as hot as a QX6700 @4.6ghz and 1.85v.

Yeah I see what you mean. What I meant was; lots of people seem to say X watts is the equivelant to a quad core, where the value is different almost every time I see someone post it .

[Ming@s]

I liked to see now, already so that it serves these manometers?

[n00b 0f l337]

As I said on OCX, you should place your first stage probe not on the suction line, but on the liquid output of the second stage.

[mytekcontrols]

On the subject of heat load I've just got to throw in my 2 cents worth.

My feeling is this...
If the evaporator and load resistor share the same insulated space, and that insulated space is of a high R value, then the only place where the energy that is put into the load resistor can go is into the evaporator. Of course there will be some leakage to the outside world, because most insulating methods wont be perfect, but with decent insulation this should be minimal (maybe 5-10 watts at most). If this isn't so, then where is the heat energy going?

The only other possibility for discrepancies with a dummy heat load vs. actual heat load from a CPU, would relate to the CPU contact method employed, the fact that only one side of the CPU is available for heat extraction (how much is heading towards the PCB side underneath), or an inaccuracy in the wattage measurement instruments.

Of course this is only my opinion, and as I said at the beginning of this post, it's only 2 cents worth

[bazx]

very nice work as always LD


there are limits to how useful dummy loads are when setting up a cascade

constant loads do not represent the the varible load given off by a cpu when it is working say spi 32m calc

when my cascade is first put under load from a spi 32m calc

(quad core with all 4x cores @ 4700mhz with 1.80v or duel core @ 5100mhz 1.80v)

a few things happen

the shock of the sudden load first wacks up all the pressures this then drives down all the hx temps these temps keep falling untill the cascade catches up
with the load it has taken at the evap

the hx temps begin to recover

by now the 32m run is about 5 lines in when the calc steps up a gear and the hx temps start to fall again with the increased load from the 32m calc

this "up and down" load from the cpu continues right tho the bench

it is here in the middle of the 32m calc that you need to tune a cascade
for its best performace

by adding a lil refrigerant to the different stages as they show there weakness

if only used a dummy load to setup my cascade it would not work in a real world bench test

my 2cents

[n00b 0f l337]

When I bench under subzero, I run F@H or a busy program to create load first so that the cascade or single stage has already caught up

[bazx]

When I bench under subzero, I run F@H or a busy program to create load first so that the cascade or single stage has already caught up

nol i sure would like to see the unit you refur to holding such a load and the results you achived

can you link me

thanks baz

[gosmeyer]

bazk,
Thanks #59 Said Alot

[n00b 0f l337]

Let me go looking, I'm only recently now getting r23 (and soon r14), and I have not had ethylene, with co2 I pulled -58C (stuck at the triple point it would seem).
As for quad core ready single stages, I'll post up results of Tyler dustins unit in the current page when I finish up tuning this week. The quad core discussion thread had the results of Harvsharks unit (With a SC12MLX) pulling -38.2C at 270 watts.

[bazx]

Let me go looking, I'm only recently now getting r23 (and soon r14), and I have not had ethylene, with co2 I pulled -58C (stuck at the triple point it would seem).
As for quad core ready single stages, I'll post up results of Tyler dustins unit in the current page when I finish up tuning this week. The quad core discussion thread had the results of Harvsharks unit (With a SC12MLX) pulling -38.2C at 270 watts.

for me there is only one type of load for tunning and that is the cpu/gpu

static dummy loads just dont do it for me unless you want to under tune for say a 24/7 rig where the unit is to be held well within its ability range

for the out and out benching cascade where performance is key then you must tune on the cpu/gpu under load if you want to get the best posible temps and then bench results after all that is why we build them

[n00b 0f l337]

I agree a 100% that that is the best way to do it, but I personally cannot afford a QX6700 or the like and setup to beat up for load testing.

[bazx]

I agree a 100% that that is the best way to do it, but I personally cannot afford a QX6700 or the like and setup to beat up for load testing.

this is true a QX700 costs quite a bit but then thats why i build cascades
so i can overclock and bench

[n00b 0f l337]

Ah, I build to build and have fun with cold temps. That and I find it an engineering challenge.
Back to the matter at hand though; pickup a quad core LD and lets see some benches!!

[SoddemFX]

Having the temperature sensor on the liquid line won't tell you an accurate enough temperature of the heat exchanger when the first stage is cooling down, maybe the heat exchanger is at -50C but the liquid line temperature point (which at this time contains no liquid) might only be showing -10C as it's just relying on conduction through the small diameter copper tube to get its "coldness"

Having it at the exit of the heat exchanger won't be accurate under load but at least it'll be indicative of when it's safe to start the second stage for the end user. If i was building a cascade to sell to an end user with only one first stage temperature point this is where i'd have put it also.

With the plate heat exchangers used in these cascades subcooling is always minimal so i'd just go off the saturation pressure to decide liquid temperature because the liquid line is generally very small and well insulated.

NoL - there's something seriously wrong with your cascade if you can't break -58C, speak in PM if you like to talk more

Superb cascade LittleDevil, your work is amazing and keeps getting better

Tom

[mytekcontrols]

All good points, and I agree with post #59. However it still doesn't detract from the dummy load vs. real load argument (watts is watts no matter how it's created). The difference is in the method of simulation.

If the CPU represents a variable load, then simulating this variability on the dummy load should yield the same results. So it would appear that what is needed is a programmable load controller. This would allow for simulating wattage change vs. time, and should satisfy the remaining discrepancy, that being variability of load.

I understand that the ultimate test will be if it works in the "real" application, but this doesn't preclude us from being able to precisely mimic this with a dummy load. It'll just take a bit more ingenuity.


Since this has gone a little off topic, we should perhaps look at starting a separate thread on this for future discussions.

[gosmeyer]

True enough, That's the bottom line!

[LittleDevil]

Some more pics of finished cascade.

Without flash:





Flash:







Best Regards

[sacha35]

Wow it looks as good as it performs mate, well done on a good job.

[elfy]

that's a sexy looking machine, very nicely done

[gosmeyer]

Sweet!!

[K404]

Gotta hand it to ya, that does look hella sweet