Printable View
Hi All. I've been busy lately, with EVGA and'all ;-)
Images are worth a thousand words. I tried to put everything on one screen so it's all self explanatory. There is more to come though.
The pic is very big, sorry, but it's the only way that one can read all the details clearly.
http://www.swiftnets.com/assets/imag...10-fullres.jpg
Next will post Unigine, then xtreme water-cooling setup (since even this dedicated dual rad is kind of mild for the folks around here ;-)
wow max temps decreased nearly by half on water. looks great :up:
temps seem pretty good, considering 10C ambient/water delta
RDM1225S is rated @ 2,000rpm, producing 81.3cfm at 36dBA. Pretty beefy fans.
as promised the Unigine tests under the same setup
http://www.swiftnets.com/assets/imag...10-fullres.jpg
Next we are testing ultra extreme setup with dedicated (2) MCR320 & (6) RDM1225S & (2) MCP655's in series. We'll note the temps first, and depending on how much drop in temp we get, we'll see how much more of a stable overclock we can get. So far our 890 is the highest on record (under furmark xtreme) on water, isn't it? we were unstable at 897, and full crash at 900. So we dropped to 890 which seems fully stable after 1 hour of Furmak xtreme..Quote:
Originally Posted by GTOViper
i need your bios please....
use gpu-z to make a backup.
please.
Might be just that particular card? I've seen EternalFantasy's GTX480 w/EK block go up to 900 with less core volt bump @1100mV and slighter higher temps at 54C.Quote:
Originally Posted by gabe
Another explanation might be that the VRegs are not getting as cool as they could be since it's cooled by heatpipe? :shrug:
Give another card a try I guess with the same setup.
He is using Kombustor. GPU load is lower in Kombustor than in Furmark Xtreme burn mode. he's also testing for 12 to 15mn. We test a full hour. An immediate crash when you go from say 895 to 900 indicates a GPU failure, nothing else.Quote:
Originally Posted by GTOViper
and to drive the point home, we ran the card at 900 under Kombustor using the Xtreme bench, with no problems for 13mn as shown below BUT the card crashed under furmark Xtreme burn after just 1:32mn - this demonstrates that 1/this particular card might not be as good as EternalFantasy; 2/lower GPU temps affect the O/C (but we already know that), and most importantly that 3/Furmark Xtreme Burn places a higher stress on the card, thus triggering a faster crash.Quote:
Originally Posted by GTOViper
http://www.swiftnets.com/assets/imag...10-fullres.jpg
Prior to this, we ran Furmak xtreme burn at 890 with the xtreme bench (2mcr320+2MCP655's), and found GPU temps 6~7C lower (45C). The theory says that we should be able to increase our OC by about 2% for every 10C drop. Since we only improved 6~7, we should be able to increase our O/C by 1.2%, or 890x1.012= 900. We now know that 900 was in fact a no-go at 900, so I think that we should be good in the 897 neighborhood, which would still validate the theory. Let's verify this next.
Just gonna put this here.
http://i225.photobucket.com/albums/d...Extreme1H1.png
:yepp: typically cards & processors that require less voltage than others are better silicons, and thus can run a little faster. Congrats! You lucked out. You'd be able to O/C even higher if you had better cooling ;) ..Quote:
Originally Posted by eternal_fantasy
but don't sweat it too much though, as I'm about to show how much of an OC gain we can get from a 6~7C drop in temps in the next few minutes...
this is really strange u are reporting temps that are far cooler than the koolance and DD blocks not to mention GPU only blocks and those were cooled by 360 rads
go higher ive seen people reach 940
Remember that this is a dedicated loop. but yes, in general Swiftech blocks always have lower GPU temps because of our pin matrix. But I can tell you with absolute confidence is that these temps are NOT lower than GPU only block for having tested this in depth. I just finished testing a 5850 using all 3 cooling solutions, Komodo, MCW60 and MCW80, and the differences between F/C and single GPU are sizeable. In any case, any difference between these particular GTX480 results and other users could come from many things, different bench setups, different GPU quality, etc..Quote:
Originally Posted by owned666
I will publish GPU only tests with this board next week... MCW60 and MCW80.
Oh, and one last thing, and it's no joke: I always get the dogs! that's one the reasons why I got into the cooling business to begin with 10 years ago.. My procs were never too good, so I had to cool them to beat my pals ;-)
Thanks! :up: Just wanna see if my card can stand up to a more "industrial standard" test as opposed to 10~ minutes of Kombustor, and it seems to be able to. And the "So far our 890 is the highest on record on water, isn't it?" and the following forum posts felt like a challenge. :DQuote:
Originally Posted by gabe
I do see the Swiftech are better at GPU cooling then the EK full cover. However they are not available to purchase at least on this side of the pond. ;)
Curious, is it possible to point an infra-red thermometer at the VRM section of the Swiftech? Wondering how well the heatpipe works to cool them.
VRM's are only relevant to OC in as much as they dictate an absolute operating limit. Let me give you an example: say the card manufacturer has a spec of 120C for their VRM's. They would hardcode the card to shutdown if the VRM's exceed this temp to prevent damage. Now, if you are running under this maximum spec, then the limiting factor for your O/C is the GPU, not the VRM's.Quote:
Originally Posted by eternal_fantasy
In other words, as long as VRM temps remain under their specified limit, they have no effect whatsoever on O/C. As to the card's life, we know (*) that component life is doubled for every drop in 10C. So we want to keep these component to run lower than stock in order to improve the shelf life of our cards, right? So let's calculate this. Say the stock temp is in the high 90's. I don't know what the shelf life of the component is, but let's say for the sake of argument that it's at least the same as the manufacturer warranty, which is what? 3 years? So let's do the math: if we drop this component temp down to 80C, we are at 6 years shelf life, down to 70C we are at 12 years, etc..
(*) drop in 10C is either equal to ~2% frequency increase, or ~double the life at same frequency
Being someone that watercools their motherboard's VRMs, overclockability is second to cooling what is hot in their computer. ;)Quote:
Originally Posted by gabe
Anyway, as far as I know, when an electrical component such as VRMs heat up, they are less efficient at what they do, which is to provide the right voltage to the components they are responsible for. A hot VRM might provide the power "within specification" of the GPU up to their rated maximum temperature, they have to work harder, outputting more heat and becoming less efficient then if they are cooler. Also their voltage output when hot have slightly higher ripple then if they are cool. Of course if these properties add up to reduce overclockability is another issue, but you have to agree when talking about electrical components cooler = better. :)
And yes, GPU cooling > VRM cooling.
EDIT: OMG Gabe holyninjaeditbatman!
What you say is correct in general. But you need to go into specifics to analize what makes sense and what doesn't. The engineers at AMD, nVidia and Intel, design their products with very rigorous specifications, that are meant to produce a good balance between performance, cost, and reliability (wich is a liability for them). In the VGA business, knowing that the stock cooling solutions are very limited in size and air flow, the engineers have to use components that can sustain high heat without failing. Doing otherwise would extend their liability to unsustainable levels. So, these components are designed to run hot (relatively speaking), and the after market cooling solutions that are available all do enough of a job to extend their shelf lifes much beyond what the stock cooling does. Thus, it really doesn't matter if these components run at 50 or 60or 70. With max specs comprised between 120 and 150 (depending on models), they are so far away from their max operating point that it becomes irrelevant. Any good engineer will tell you that good engineering is SUFFICIENT engineering. Overkill is a waste ;-). We are interested in max O/C, and extended reliability compared to stock, and that's exactly what we are getting here.Quote:
Originally Posted by eternal_fantasy
This being said, let me finish the last chapter of this round of testing by posting the best stable O/C for this GTX480 under furmark xtreme burn (1hour), and using an absolutely extreme cooling bench:
http://www.swiftnets.com/assets/imag...10-fullres.jpg
What is interesting to notice, is that for a 6C drop in GPU temperature (51 down to 45), we gained 6 Mhz in stable performance (less than 1%). Note btw that I tested at 897, and furmark crashed after 30 minutes. What is also interesting to notice is that in order to do so, we went from an MCR220+MCP655 pump, to two MCR320 rads with two MCP5655 pumps. The cost of cooling more than doubled, for a gain of only 6C, and an improved OC of only 6 Mhz.
I thought I would do the above to bring things back into prospective, and allow everyone to make an informed choice when it comes to cooling!
Gabe:
Availability of the 480 HC FTW was/is too limited, and thus I decided to pull the trigger on two EVGA GTX 480 SCs. I was very impressed with your 5970 block results, and would be excited at the prospect of a full cover 480 block from Swiftech.
Any thoughts on why you went with the hybrid design on the EVGA HC FTW? Can we expect any stand-alone 480 blocks from you outside of the EVGA HC FTW edition?
Thanks again for the test data and feedback!
WoW! 2 D5 and 2 rads for one video card! It's mental illness :confused:
nah, That is Xtremesystems. Gabe said it in is first post to give us a idea what the temp would be with a overkill setup. Gabe, do you have any temp probe or infrared thermometer lying around so you could see what are the vrm temp with your heatpipe solution. Sadly the gtx 480 dosent seems to have a probe for that or gpu-z is not recognizing it.
hmmmm.Quote:
Originally Posted by snoro
http://hwbot.org/signature.img?iid=3...se&iehack=.jpg
extreme?
LN, cold water or dry ice - extreme. 2 pump and 2 rads for a card - IMHO madness, but temp is perfect.
You havent seen Naekuh rig have you ? He have 2 360mm rad and 2 mcp 355 pump for 2 hd 5970. We can call that Xtreme not only for xtreme form of cooling but also for massive overkill. And Gabe, i agree with you,for a engineer overkill mean he have done is job badly,very badly. But for us, overkill is opur motto. Well, atleast for a couple of person over here.
Honest testing and reporting to the public from the manufacturer is much appreciated, especially when the result may not be in great favor to the product you are selling. :yepp: :clap:Quote:
Originally Posted by gabe
You have proved your point though, that the Swiftech Hydro-Copper is amazing at cooling the Core temperature of a GTX 480. :up:
wow, temps, volts and clocks are all over the place for the GTX 480's
It's great right? :up: you'd actually have a good chance of a great scaling overclock, as opposed to generally average overclocks on the 5870, even on water.Quote:
Originally Posted by Helloworld_98
Average OC on the 5870 > GTX480 right? :shrug:Quote:
Originally Posted by eternal_fantasy
If the GTX 480 is not oc'ed. But I'm seeing a ton of 480s that are hitting 20% oc even on stock air cooler. 900Mhz is pushing 30% oc and thats rare, even for watercooled gpus.Quote:
Originally Posted by Vinas
You guys with water setups are getting very lucky this round with nVidia gpus and overclocking numbers.
I am out in the sand box but my FTW card arrives tomorrow, I need to wait till I can order my second card so I can setup the first loop for GPU's. Good to see some data on the the block.
Your in Iraq and building a computer?Quote:
Originally Posted by H20Cooled
mh i thought 2 triple is kinda normal :D
but rly nice temps...mb i should get some for testing too
Nope, I am currently south of Iraq, I'm in the UAE right now for work. I am typing this from Abu Dhabi. I will be back state side next week once I finish up my meetings. I have to come out here 3-4 times a year for work. Might have to spend a month or so out here shortly depending on how my meeting goes in the am.Quote:
Originally Posted by =[PULSAR]=
I have a project in Kuwait that has been pushed back till next year as well.
damn water temps r piss water lol my water temps (23.5)r lower then that mayb becuz i using 3 swiftech rads ..Quote:
Originally Posted by gabe
Anyone can connect up 5 quad radiators with 20 delta fans and bench in a 10C room. That was not the point of this thread.Quote:
Originally Posted by xslix
Design choices: Sorry, but I cannot comment in details as it would reveal too many technical choices that could be emulated by my competition. What I do want to say in general terms is that 1/ from an industrial design standpoint, this choice offers the best VRM TIM joint one could possibly have given the card's layout, and that 2/ the water-cooled heatpipe removes more than enough heat to allow overvolted VRM's to operate substantially (by several ten's of degrees C) below their maximum operating point.Quote:
Originally Posted by TradeWind
stand alone 480 blocks: well you already know I presume that the MCW60 is compatible with the G80 adapter, and there are reports already of the upcoming MCW80.. so yes, it's coming.
Gabe:
No worries, I understand. You have to make a living! :up:
I will stay tuned for more updates on the MCW80. In the mean time, I have two EVGA 480 SCs, inbound from NE in a couple of days, that will need something to keep them cool. Based on my experience with your 5970 block, and your presence here on the forums, Swiftech will always enjoy front-runner consideration in my loops.
Thanks again for the guidance.
As promised (admittedly late), here are the comparative test results for MCW60 with stock back plate, MCW80 with stock back plate, and the EVGA FTW cooling solutions. The Delta T GPU/Air is the key indicator.
Ranked by order of performance, lower to higher:
The MCW60: 23.2C
http://www.swiftnets.com/ASSETS/IMAG...-1600x1200.jpg
EVGA FTW (Swiftech) cooling solution: 21.6C
http://www.swiftnets.com/ASSETS/IMAG...-1600x1200.jpg
and the MCW80: 20.3C
http://www.swiftnets.com/ASSETS/IMAG...-1600x1200.jpg
As expected, the MCW80 pulls ahead by a relatively narrow margin of 1.3C compared to the EVGA FTW cooling solution. The 60 is clearly outdated (retirement was well deserved), and the EVGA FTW performed remarkably well.
What about VRM temps with MCW60/80 and stock cooling plate?Quote:
Originally Posted by gabe