GeForce 7800 GTX ***NOT*** dissipating 200W!

Tzupy · Post by **Tzupy** » Tue Jun 28, 2005 12:01 pm

@Edward Ng: I'm looking forward to see the results of your setup.
However, for most of us that aren't watercooling, it would be most helpful to know if the 7800GTX can be cooled by a VF-700Cu (since no Silencer is available for it yet). I'm not sure if some parts of the 7800GTX are cooled by the main stock cooler or separate heatsinks. There's also the issue of lenght: it seems to be quite long, and you may want to remove the second HDD cage of the P180 in order to make it fit properly. And, by the way, xbitlabs says it consumes MORE power than a 6800U (I guess it's a matter of load percentage in a particular test).

Edward Ng · Post by **Edward Ng** » Tue Jun 28, 2005 12:12 pm

That's the only site I know that shows 7800GTX pulling more power than the 6800 Ultra; several other sites did actual power tests and found the 7800GTX coming in less power hungry (not just quoting nVIDIA, but actually testing it out).

VF-700Cu--maybe, although there are several other solutions I'm personally more interested in trying before resorting to this item. I'll decide once I receive the parts, likely tomorrow.

I already moved my SP1614N to the lower compartment, and have the case prepped to receive the Raptor down there as well--i.e. I already took out the upper HDD cage, not so much in preparation to fit the longer 7800GTXs, but more to reduce air flow restriction from the front intake.

As I mentioned (I believe in another thread), as far as I can tell, I won't be water cooling this system. There's just no quick and easy method I can come up with to fit sufficiently sized radiators on this case without making permanent modifications, and I don't want to do that to my beloved P180.

-Ed

dexton · Post by **dexton** » Tue Jun 28, 2005 1:43 pm

Hey Edward. I have a question that you may already have answered: Is there a dust filter in the GPU duct? I was thinking that it would be truly useless if Antec didnt put dust filter on ALL air intakes!

EDIT: did you get the new Abit board with heatpipe cooling for the nForce4 chipset?

Edward Ng · Post by **Edward Ng** » Tue Jun 28, 2005 6:57 pm

No filter; not a big fan of that silly duct.

I have an AN8-SLI that I have in my silent rig, which is in an SLK2650-BQE with a Venice and 6600 non-GT; I have another AN8-SLI coming in tomorrow that will be installed into my P180 with a San Diego and a pair of 7800GTX in SLI.

-Ed

magic_p · Post by **magic_p** » Fri Jul 01, 2005 5:40 pm

dexton wrote:
Edward Ng wrote:I'm going to use two 512s for now and squeeze the most perf I can out of them until I start hitting games that are happier with over a gig; I'd rather get better perf in most of my games and sacrifice one than speed up one and lose perf in all the others.
hmm, Ive heard that C'n'Q wont work with 3 or more modules. Is that correct? Adding 2 more 512s later would take away the C'n'Q functionality.

Not true for my Asus A8N-SLI with 4000+ (Clawhammer). I have 4x 512MB sticks of Kingston ValueRAM and Cool and Quiet works just fine. I idle at around 33-35C and at load playing WoW I'm at 51-53C

wumpus · Post by **wumpus** » Fri Jul 01, 2005 10:56 pm

a gig of RAM

This is a mistake; Battlefield 2 actually uses almost an entire gig while running [at highest detail]. I can send you taskman screenshots if you're skeptical.

As of mid-2005, all high end gaming rigs should have > 1gb. And any gaming rig definitely needs 1gb minimum!

wumpus · Post by **wumpus** » Fri Jul 01, 2005 11:01 pm

Also, I have a 7800gtx, and while it is reasonably quiet for a single slot stock solution, you will definitely want to invest in two Zalman 700cu coolers. No doubt about it. For heat, noise, and overclocking reasons.

I have a ghetto cardboard duct I am experimenting with on the single 7800gtx, ducting from the unoccupied 2 slots above the card. It definitely helps by keeping the fan from ramping. It's decent at idle, but it'll ramp very aggressively under load!

I also find that the voltage circuitry at the "front" of the 7800gtx near the voltage connector-- which isn't cooled at all by the stock cooler-- gets EXTREMELY hot. I'm hoping my recently purchased Zalman 700cu will blow some air down and over the card to cool the voltage regulator heatsink and circuitry as well.

Tzupy · Post by **Tzupy** » Sat Jul 02, 2005 4:32 am

Well, this has nothing to do with quieting a 7800 GTX, but I was curious if anyone encountered this issue (read about it on Anantech forums): when overclocking some GTX cards, like 460 instead of 430, due to some BIOS / driver issues, the clock goes to 500 MHz. And 3DMark results seem to fall in line with the clock increase. I'm just wondering if this is 100% true, it could be an indication of the 7800 Ultra speed.

Edward Ng · Post by **Edward Ng** » Sat Jul 02, 2005 12:10 pm

wumpus wrote:
a gig of RAM
This is a mistake; Battlefield 2 actually uses almost an entire gig while running [at highest detail]. I can send you taskman screenshots if you're skeptical.

As of mid-2005, all high end gaming rigs should have > 1gb. And any gaming rig definitely needs 1gb minimum!

I never said there are no games that use a gig or more; I said I have more games that don't, and benefit more from the tight timings I can achieve with the amount of RAM I have now; there's no way to get 1.5 or 2GB down to the speed and timings I am getting with two 512MB pieces, and to sacrifice some performance in all games for the benefit of just one? This is not a one trick horse and I am not a one-game guy.

I figured I made this clear enough (in a different thread), but I suppose I should start posting the same statements to all my threads so I don't have to hear the same thing in the ones where I don't.

And you speak as if I can't just add more memory later when the majority of my games would be faster with over 1GB of lower performance memory than 1GB or less of faster memory.

Where's the mistake?

-Ed

wumpus · Post by **wumpus** » Sat Jul 02, 2005 8:55 pm

I said I have more games that don't, and benefit more from the tight timings I can achieve with the amount of RAM I have now

Do you have any benchmarks showing the better framerates you get from this "tighter RAM timing"?

I was under the impression that tweaking memory performance just wasn't that significant in this generation of CPUs. Maybe at 8x6, but particularly not at high resolutions where the video card is the bottleneck 99% of the time. And then there's the on-die memory controller of the A64..

I just can't recall seeing any data to support your assertion that memory timings are this important for high end gaming. Can you provide some?

Edward Ng · Post by **Edward Ng** » Sat Jul 02, 2005 9:34 pm

As just one example, take a look at the results from this recent test using the very same memory I bought, OCZ Gold VX. Notice that the timing settings are already quite close between all the tested samples in here, and the results show consistent and obvious differences; remember that with dual 7800 GTX in SLI, the CPU and RAM are more often the bottleneck than you'd think. Also realize that the difference in timings between two 512MB sticks and four 512MB sticks is going to be far greater than the difference in timings shown in this test as is, resulting in a far larger difference in performance than the comparison indicated. I particularly like this example because the various memories proved to show a decent difference even with very close settings; there's no way I'll get even close to these settings with four sticks of 512 or two sticks of 1024.

I could easily rattle off several more links with the data you're looking for, but I'm hoping this example is enough as is.

-Ed

Shining Arcanine · Post by **Shining Arcanine** » Sun Jul 03, 2005 7:23 am

lenny wrote:
Edward Ng wrote:I'll think of something...
Somebody once suggested using 4 reserators as the legs of a table - just a thought

Two for each of the graphics cards, one for the processor and another for the northbridge?

wumpus · Post by **wumpus** » Sun Jul 03, 2005 9:49 am

but I'm hoping this example is enough as is.

Well, 1999's Quake 3 Arena @ 1024x768 and a game based on that same engine @ 1024x768 are not exactly compelling examples of even remotely modern games. And would you really be running at 1024x768 with 7800gtx SLI? Somehow I doubt that.

Even considered standalone-- and these are highly questionable results for gamers-- these aren't dramatic differences. As the anandtech article you linked says:

All 4 memories can achieve similar top performance levels

Essentially you can get that same minor performance boost by overclocking the FSB by about 10mhz. Furthermore, the "best" performing memory requires 3.5-4 volts (!!):

This brings us to the handicap with VX, and it will be a huge one for many users. VX requires high voltage to stand out from crowd - voltages not generally available on standard motherboards. You will need to start about 3.0V and extend to at least 3.5V to 3.6V to get the most from VX memory. The good news is that OCZ still provides a lifetime warranty on VX even if you run it all day long at 3.5V.

Just doesn't seem worth it to me for such a tiny real-world perf gain in modern games.

Edward Ng · Post by **Edward Ng** » Sun Jul 03, 2005 9:56 am

First of all, RTCM is also in there.

Second of all, I am using specifically high voltages (3.1-3.5) which is why I got the VX.

Btw I've been playing BF2 on this machine for the last couple of days and I don't see where I'm losing this dramatic performance you're talking about for using "just" a gig, even though I have the settings in the game maxed out while playing at 1600x1200 with 4X AA.

-Ed

wumpus · Post by **wumpus** » Sun Jul 03, 2005 12:35 pm

I don't see where I'm losing this dramatic performance you're talking about for using "just" a gig, even though I have the settings in the game maxed out while playing at 1600x1200 with 4X AA.

You won't lose anything in-game, but you'll page to disk like a mother once you exit the game; bf2 is using nearly every bit of that gb of RAM at highest detail settings. So all the windows stuff that paged out during gameplay has to be loaded back into memory from disk.

I dunno, the idea of a "high end" system with only 1gb of RAM is pretty laughable to me. I wouldn't even consider building a performance oriented system with less than 2gb these days.

From an Anandtech battlefield 2 RAM performance test thread:

I put together a quick web page with the results of using the Windows Performance Graph to look at Available Physical Memory, Percentage of Swap File Used, Page Faults, and Disk Reads while playing BF2 online.

I think the results are consistent with what others are reporting. 1GB of RAM is enough to run this game fairly smoothly, but it is only just enough. I played on a 32 person server and was using just the demo, which may not have fully revealed how much memory this game can suck up. I can easily imagine that it wouldn't take much to increase the memory load on the system and push things in the direction of more stutter. For example, I wasn't using the voice communication program or running anything else in the background (other than the performance logging and Norton's AV). And, of course, playing the Demo means that there was no switching to additional maps or dealing with servers tracking players' stats.

[/url]

Edward Ng · Post by **Edward Ng** » Sun Jul 03, 2005 12:54 pm

wumpus wrote:I dunno, the idea of a "high end" system with only 1gb of RAM is pretty laughable to me. I wouldn't even consider building a performance oriented system with less than 2gb these days.

Then go ahead and laugh all you like.

-Ed

rpsgc · Post by **rpsgc** » Sun Jul 03, 2005 1:41 pm

Great off-topic eh guys?

Anyway, The Inq is saying that the 7800 GT is on its way. All things point to a simple clock speed reduction.

dexton · Post by **dexton** » Sun Jul 03, 2005 2:13 pm

Edward Ng wrote:As just one example, take a look at the results from this recent test using the very same memory I bought, OCZ Gold VX.....

...I could easily rattle off several more links with the data you're looking for, but I'm hoping this example is enough as is.

Im not a game freak, but those are old games at 1024x768. Translated into the newest games at 1280 or 1600, it will differ like 1-3 fps.

Whether you need <1gig of RAM for the current games, is not for me to answer. To get 2x1GB or 4x512MB doesnt seem to be a big difference today with Venice. I definitely doesnt think that the extra cost for expensive LL memory is justified though, from the tests Ive seen.

Tzupy · Post by **Tzupy** » Mon Jul 04, 2005 6:31 am

@people that already have the 7800 GTX:
Since you were so lucky (read rich) to get it, please post some results / pictures of your setups, for the rest of us that have it as an upgrade option. If you already posted those elsewhere, please post a link here.
Now, with the risk of restarting the flamewar, may I suggest this DDRAM configuration: 2x512 MB of PC4400 that will run at 255 MHz with 2.5-3-3-7 1T (255 x 11 = 2805, the target CPU frequency). Later another 2x512 MB of the same could be added and run at 2T - IIRC the performance penalty for 2T with games is about 3-5%.

Edward Ng · Post by **Edward Ng** » Mon Jul 04, 2005 6:48 am

What Tzupy is talking about is precisely what I'm already doing; HTT255 * 11.0 = ~2800, at 2.0 8-3-2 1T. When I have more games that benefit from >1GB than I do that don't, then I'm adding more and switching down to 2.5 8-3-3 2T.

And just to clarify exactly what the basic issue here is--many people are looking at it from the pure standpoint of just the improvement/gains made from tighter timings--that's not how things really work. Look at it from a silencing point of view; if you just change your mobo cooler to passive, what's next? Then there's the VGA cooler, then there's the hard drive mounting, then there's ducting cooler air to the PSU to prevent ramping then there's changing the CPU cooler/downvolting the CPU fan etc. etc. etc. If you took only one piece of that equation and applied it, how much does it help? Only so much. If you add it all together, what do you get? A quiet machine. From my standpoint, does going tighter make a huge difference? Of course not!

What happens when you add higher CPU clock rate+higher memory speed+tighter memory timings+faster GPU+faster graphics memory? Now your total gain amounts to something! It's this same reason why I'm not going to weaken a part of that chain just for one game. It's like saying only one game manages to overheat my passively cooled GPU, so now I'm going to add active cooling to the thing just for that one title when all my other titles are fine--weakening a link in the silencing chain for one title over the rest? No thank you.

-Ed

Edward Ng · Post by **Edward Ng** » Mon Jul 04, 2005 6:52 am

Re: photos--I'm still working on the best overall solution in terms of keeping max performance while dropping noise little by little and maintaining 100% stability. Once that's finalized, I'll be able to do a final clean up of my cabling, at which point I'm perfectly happy to shoot some photos.

As of the moment though, I think somebody will call the cablegami police on me for the mess I have going on in there.

-Ed

wumpus · Post by **wumpus** » Mon Jul 04, 2005 11:40 pm

When I have more games that benefit from >1GB than I do that don't

It just seems ridiculous (to me, YMMV of course) that you're building a balls out dual SLI 7800gtx system with a whopping 1gb of RAM. But it's your system, and I find your beliefs fascinating, if not entirely grounded in real world benchmark results.

Anyway, back on topic.. I had to fly home to NC over the holiday weekend which burned two days. I'll put the Zalman flower on my 7800gtx tomorrow for sure. It seems quite happy at 480/1300 so far.

wumpus · Post by **wumpus** » Mon Jul 04, 2005 11:50 pm

Also, here's some data from an Anandtech forum showing the whopping difference that 1T vs 2T memory timing-- these benchmarks were performed using real world games at real world resolutions. Sample:

Insomniak System: CPU - Winch@2400Mhz | Memory- 2.5-3-3-8 X @200Mhz

Battlefield Vietnam 1T- 106.5FPS
Battlefield Vietnam 2T- 102.1FPS
1T is 4.3% faster*

Source VST 1T- 164.7
Source VST 2T- 158.3
1T is 4% faster*

Super Pi 8M 1T- 6:49
Super Pi 8M 2T- 6:58
1T is 1.4% faster

3dMark03 1T- 9258
3dMark03 2T- 9187
1T is 0.8% faster

Far Cry 1T- 105.2
Far Cry 2T- 103.4
1T is 1.7% faster

Insomniak definity showed the most pronounced differences so far, had a couple games in the 4% range. As an average though he's still less than 2%.

Summary: you'll be lucky to get 4% faster, looks like 2% in most games.

jojo4u · Post by **jojo4u** » Tue Jul 05, 2005 12:57 am

http://www.hardtecs4u.com/reviews/2005/ ... index6.php
Also shows some TMPEG Enc and Winace benchs. The loss is around 2% as well.

Tzupy · Post by **Tzupy** » Tue Jul 05, 2005 1:25 am

@Edward Ng: you mentioned 'maintaining 100% stability' and I'm not sure which Asus A8N-SLI you are using... If it's the Premium, then you should have no problem running 255 MHz 1T memory, but for the others, like my Deluxe

, this is not an option, and it's not an issue fixable by a newer BIOS, AFAIK (even at 240 I can't get it fully stable with 1T). Oh, and on my machine I couldn't use more than 3.0V for memory, not that I wanted to, so the VX shouldn't be the best choice.

dexton · Post by **dexton** » Tue Jul 05, 2005 1:44 am

Tzupy wrote:@Edward Ng: you mentioned 'maintaining 100% stability' and I'm not sure which Asus A8N-SLI you are using... If it's the Premium, then you should have no problem running 255 MHz 1T memory, but for the others, like my Deluxe , this is not an option, and it's not an issue fixable by a newer BIOS, AFAIK (even at 240 I can't get it fully stable with 1T). Oh, and on my machine I couldn't use more than 3.0V for memory, not that I wanted to, so the VX shouldn't be the best choice.

I dont think youve quite followed the thread:

Edward Ng wrote:I have another AN8-SLI coming in tomorrow that will be installed into my P180 with a San Diego and a pair of 7800GTX in SLI.

and

Edward Ng wrote:What Tzupy is talking about is precisely what I'm already doing; HTT255 * 11.0 = ~2800, at 2.0 8-3-2 1T. When I have more games that benefit from >1GB than I do that don't, then I'm adding more and switching down to 2.5 8-3-3 2T.

AN8 SLI != A8N SLI

I agree, the manufacturer's product names are VERY similar and in som cases almost identical. Asus, MSI and Gigabyte all have boards that are named K8N, for example.

Johan

Edward Ng · Post by **Edward Ng** » Tue Jul 05, 2005 1:52 am

That's right, I'm not using any ASUS A8N at all; it's an ABIT AN8-SLI. This board allows up to 3.5V.

-Ed

dexton · Post by **dexton** » Tue Jul 05, 2005 2:11 am

Edward Ng wrote:That's right, I'm not using any ASUS A8N at all; it's an ABIT AN8-SLI. This board allows up to 3.5V.

and what voltage was needed to reach DDR510 @ 2.0 8-3-2 1T?

Edward Ng · Post by **Edward Ng** » Tue Jul 05, 2005 2:39 am

Tzupy · Post by **Tzupy** » Tue Jul 05, 2005 3:06 am

My mistake, sorry. I kept reading AN8-SLI as A8N-SLI, probably because I had a subconscient desire to see the test with 'my mobo'. I am worried that I'll have to upgrade my mobo to Premium in order to accomodate a long card, since the NB-1 on the chipset would interfere with it.
Anyway, if someone uses the VF-700Cu on a 7800 GTX, I'd like to know if the fan can be run lower than @12V, and still be effective. Or Zalman should work on an improved version (more likely IMHO). Maybe the 7800 GT will run at lower voltage (and speed, of course) and be easier to cool.