NVIDIA GTX 595 (picture+Details)

[Manicdan]

How many runs were done?

As many know, Vantage peaks in several different areas; many of which are less than a second long and may not be picked up by a standard power meter.

In addition, CPU usage is a HUGE factor and can increase / decrease number accordingly and in a non-linear fashion.

Looking at that chart, it seems like the calculations for some cards are VERY high while others are low. It could be that the monitor is picking up the areas where CPU + GPU peaks converge in some situations and registering situations of non-convergence in others.

whats your opinion if someone did a 1 hour loop of some game or graphical benchmark and looked at the total power consumption of say killawatt meter? do you think that a spiky consumption would result in horribly skewed total watt-hour consumption? while the display of a meter like that only updates every second or so, does the watt-hour part depend on those low update frequencies?

you take power readings to heart a little more than most sites do, so your opinion is quite valued.

[SKYMTL]

whats your opinion if someone did a 1 hour loop of some game or graphical benchmark and looked at the total power consumption of say killawatt meter? do you think that a spiky consumption would result in horribly skewed total watt-hour consumption? while the display of a meter like that only updates every second or so, does the watt-hour part depend on those low update frequencies?

you take power readings to heart a little more than most sites do, so your opinion is quite valued.

Let' break this down. First of all with benchmark selection.

In an optimal situation and in order to ensure the CPU's power consumption plays as small an influence as possible, a non-gaming benchmark should be chosen.

Granted, this may not show "real" power consumption but that's not really the point. What it does is show COMPARATIVE numbers which is ultimately the goal of any chart.

I have tested literally every possible application in order to determine the best possible combination of high GPU load and low CPU load. Furmark and a number of others tend to put a high load on the CPU so those were out. I ended up being left with two options that put a constant load on the GPU: 3DMark's Batch Size test and Vantage's Perlin Noise.

Unfortunately, Perlin Noise was trashed since Vantage reloads the benchmark every time which eliminates the "constant" load that is needed. NVIDIA's application detection has started throttling Perlin Noise as well so that was definately thrown out.


The Kill A Watt is a great tool but ultimately very dangerous to use for comparative power consumption testing when using games, etc. The reason for this is exactly what you outlined: it's polling rate is ~0.75 seconds which means it can completely MISS a peak if the system load wasn't at a near-constant level. Hence why using a game, or 3DMark's standard tests is a huge mistake. Even the UPM meter I use has a ~0.25s and I still don't think that is quick enough to accurately judge the peak power consumption of a game.


Now to the heart of your questions:

you think that a spiky consumption would result in horribly skewed total watt-hour consumption? while the display of a meter like that only updates every second or so, does the watt-hour part depend on those low update frequencies?

No. Not at all. In order to get a proper idea of watt hours, a large sample size needs to be taken. Say, over the course of 12 hours. In those 12 hours a little over 43,000 data points will have been logged and even if 1000 were absolute peaks, the result would carry the weight of averages.

Hope that makes sense.

[Manicdan]

^thanks a bunch, i completely forgot about the other 3dmark tests

do you think those make great use of the ram and core? or is the rams power consumption just so minor today compared to core and vrms

[highoctane]

Whatever, as i was saying, nvidia will have a hard time battling the 6990 within the same powerconsumption. Especialy when u look at how close the single GTX 580 is to 6990. Even the SLI 570 is above 6990 in terms of powerconsumtion.

It will be interesting to see nvidia's binned 580 cores compeeting against amd's binned 6970 cores.

Sure power consumption is going to be high, it is/going to be high from both parties at this level of performance. The power consumption is no worse than an SLI or crossfire configuration delivering comparable performance levels instead of two cards its all rolled onto a single pcb.

480's and 580's still sell despite having higher price tags and higher power consumption than the competition but they where also faster than the competing products.

[SKYMTL]

^thanks a bunch, i completely forgot about the other 3dmark tests

do you think those make great use of the ram and core? or is the rams power consumption just so minor today compared to core and vrms

The Batch Size test uses less than 5% of an eight core CPU at most as long as long as the detail settings are increased enough to ensure 30-40FPS. Higher than that and CPU cycles begin to be eaten up.

IMO, the memory really is minimal in its impact. What is a concern is HDD power though. Some tests and games tend to access the hard drive much more frequently and certain high capacity drives can consume 15W or more during certain read/write/seek operations. When you're talking about certain lower-end cards, that 15W can have a massive impact. Hence why any test chosen should not stress the HDD during rendering.

[Manicdan]

The Batch Size test uses less than 5% of an eight core CPU at most as long as long as the detail settings are increased enough to ensure 30-40FPS. Higher than that and CPU cycles begin to be eaten up.

IMO, the memory really is minimal in its impact. What is a concern is HDD power though. Some tests and games tend to access the hard drive much more frequently and certain high capacity drives can consume 15W or more during certain read/write/seek operations. When you're talking about certain lower-end cards, that 15W can have a massive impact. Hence why any test chosen should not stress the HDD during rendering.

i know EXACTLY what you mean with the HDD, my total consumption is 350w max from the wall with cpu in prime and gpu in furmark, idling at 68-72w, if i throw in my WD black, its an extra 10 watts.

if only someone built a test that uses up a gpu fairly (less sharders more textures), no cpu usage, and fills in the memory without having to access the drive, we could be getting somewhere.

[JohnZS]

I am already looking forward to the Hardware Canuck's GTX 590 review, it was refreshing to see 8X FSAA tests in their HD6990 review, I hope to see these sorts of tests done more often by reviewers and also not with the same games.
John

[SabreWulf69]

From Expreview -

NVIDIA GeForce GTX 590 To Roll Out On March 22nd

AMD Radeon HD 6990 graphics card has unveiled yesterday,what about NVIDIA GeForce GTX 590?We have just got the news that NVIDIA has confirmed the launch date of the graphics card.

According to NVIDIA,the release day of GeForce GTX 590 falls on March 22nd.GeForce GTX 590 packs dual-GF110 core,features 1024 CUDA Core,3GB GDDR5 memory,dual-8pin external power connectors and has TDP of 375W.

Coincidently,March 22nd is also the release day of Crysis 2.


Source

[sin0822]

so it is confirmed then its two GF110s from TPU:

NVIDIA GeForce GTX 590 Launch Date is March 22
The dust seems to have settled down, after AMD's launch of the Radeon HD 6990, extending the red-team's performance lead previously held precariously by the Radeon HD 5970, to the GeForce GTX 580. It looks like NVIDIA will challenge the performance leadership with GeForce GTX 590, a dual-GPU graphics card that uses two GF110 GPUs (the ones on GTX 570 and GTX 580), for an SLI-on-a-stick solution. Rumors of NVIDIA working on this card became concrete as early as in November 2010, when NVIDIA's reference board became public for the first time.

Latest reports suggest that NVIDIA has chosen March 22 as the launch day of GeForce GTX 590. Incidentally, that is also the launch date of EA/Crytek's much-hyped, initially DirectX 9 action/shooter game, Crysis 2. GeForce GTX 590 uses two GF110, though the shader configuration and clock speeds are not known. Since NVIDIA is chasing the top-spot, you can expect the most optimal configuration for the GF110s. A total of 3 GB (1536 MB per GPU system) on board, and NVIDIA's workhorse PCI-E bridge, nForce 200 will be the traffic cop and radio station between the two GPUs. The card will be able to do 3DVision Surround (NVIDIA's multi-display single head technology comparable to ATI Eyefinity) on its own, without needing a second card.

[Lu(ky]

Anyone want to guess the price of the card? I would think around $800-$900 does this seem about right?

[zerazax]

Only 3GB GDDR5... hmm I wonder how that plays out at 2560x1600 against the 6990

[Chickenfeed]

Anyone want to guess the price of the card? I would think around $800-$900 does this seem about right?

Im going with a MSRP of +/- $50 (649-749) of the 6990 pending performance given they are likely to be fairly close. Actual pricing on these products have historically varried so what they settle at is anyone's guess.

[InCredible]

depending on performance id guess msrp would be $799 ..probably being a little faster then 6990 and costing the premium to do so.. At least thats how it normally works, however this time it MAY be different with the main reason being power of course. Since amd has always went smaller efficient gpus vs a power hungry beast this MAY actually pay off and continue the performance crown..Since for nvidia to be at 375w they will more then likely have to low clocks and i DOUBT they will have a dual bios like amd did which is VERY intelligent to pull out a little extra performance and break the "limit" but yet still be compatible with it..

[HelixPC]

MSRP will be $799.99

[-Sweeper_]

MSRP will be $799.99

[highoctane]

Since for nvidia to be at 375w they will more then likely have to low clocks and i DOUBT they will have a dual bios like amd did which is VERY intelligent to pull out a little extra performance and break the "limit" but yet still be compatible with it..

Nvidia has a tendency towards delivering high OC'd as well as std clocked reference cards.

How often is that switch really going to be used and really how hard is it to simply OC any video card with one of the mainstream utilities or flashing. Dual bios is cool and neat but at the end of the day doesn't really do anything that can't already be done very easily from within the OS.

[Chrono Detector]

I just wish NVIDIA increased the VRAM to 6GB, but yeah that is overkill, because 1.5GB per core just isn't cutting it anymore, but I'm glad to hear that it will use the GF110 cores.

[Lu(ky]

Question would you be able to use this GTX595 + GTX580 for a SLI setup? Dang I just pulled the trigger on the 6990 and getting it tomorrow thinking this was far away from being available.. I really wanted to go with Nvidia on my next setup...

[RPGWiZaRD]

A swedish shop already lists GTX 590 price http://translate.google.com/translat...x-590-prislapp

Comparing current swedish prices of GTX 580 usually ranges from like 4000~5000 SEK vs ~$500 USD on newegg this GTX 590 will probably land at $799 USD price at launch (if lucky maybe down to $749 at cheap USD retailers).

[SpuTnicK]

From that source

It can be compared against competing Radeon HD 6990 with a price tag of around 6000 dollars.

WTF?

[Lanek]

Nvidia has a tendency towards delivering high OC'd as well as std clocked reference cards.

How often is that switch really going to be used and really how hard is it to simply OC any video card with one of the mainstream utilities or flashing. Dual bios is cool and neat but at the end of the day doesn't really do anything that can't already be done very easily from within the OS.

Ofc they will be overclocked version, even if i'm not sure they will be out from Evga the same days of the 595 release (maybe ), it's not like pull a 900mhz 560TI.

But this will be the case too for the AMD6990, AIB have allready annonced get there OC version, without the switch bios.. so at least we will see 880mhz version and 900mhz+ .

This is maybe where the real fight will be. 590 AIB OC retail version vs 6990 AIB OC retail version. they will surely cost an arm. But i think AIB will get some fun there.

[MrToad]

From that source

WTF?

Is $6k a lot? I get a bit confused when it comes to loose change in other currencies

Either way, the article has been updated and the etailer (Dustin is the name?) says that is all big mistake and blah blah blah...

[onethreehill]

Leaked specs?
Two GTX 580 chips
Clockspeed 60xMHz/12xxMHz/3400MHz
http://www.sweclockers.com/nyhet/136...as-den-22-mars

[InCredible]

Nvidia has a tendency towards delivering high OC'd as well as std clocked reference cards.

How often is that switch really going to be used and really how hard is it to simply OC any video card with one of the mainstream utilities or flashing. Dual bios is cool and neat but at the end of the day doesn't really do anything that can't already be done very easily from within the OS.

yea and those oc'd cards will be overpriced as well. they wont be the norm. just like any oc cards.
and i have a 6950 right now and i can tell you that the dual bios is AMAZING..between flashing to 6970 and trying to adjust voltages and having bad flashes etc, i flip the switch boot flip it again flash bios and its done..simple and easy...flashing a bios within a minute..love it..
as far as the easy oc, its not the point of being able to do it etc..its the fact that Amd's card can claim higher results because it can do it in stock fashion. Of course both will overclock from there to achieve even higher performance easily within windows.

Bottom line though review sites will be able to test the 590 at 6xxmhz/3400mem(if above is true) and the 6990 at 880/5000 since the cards are stock..surely they will have the 830 as well but the 880 will be there too. Also its a LOT easier to flip a switch for people who don't want to OC. They know it works or else amd wouldn't have put it there so some people feel safer and more confident in using it...of course all of us here can easily clock chips etc but XS isn't the only people that buy parts just saying.

[Lanek]

Leaked specs?
Two GTX 580 chips
Clockspeed 60xMHz/12xxMHz/3400MHz
http://www.sweclockers.com/nyhet/136...as-den-22-mars

old and based on the gpu-z who have been show some pages ago on this thread... but if it's only 6xx mhz, AMD have not worry to have...