NVIDIA GTX 595 (picture+Details)

[SKYMTL]

Will pcie 3.0 bring moar powa? Does anybody know?

Not that I know of.

I believe boards which fell under the HPCES v1.0 specification (remember, the PCI-E 2.0 base spec never included 225W - 300W inputs) will now be incorporated directly into the PCI-E 3.0 base spec. However, there weren't any increases in power delivery limitations.

This basically means that companies producing higher-end cards will no longer need certifications for BOTH HPCES and PCI-E but the maximum remains 300W.

[IvanAndreevich]

Who cares what the power consumption is? More important are the cost and performance.

[[XC] gomeler]

Who cares what the power consumption is? More important are the cost and performance.

We have standards for a reason.

[zalbard]

We have standards for a reason.

Makes it easier to spot outstanding devices?

[saaya]

awesome article on xbitlabs testing 560 SLI
http://www.xbitlabs.com/articles/vid...i_7.html#sect0

At the highest resolution the SLI configuration is comparable to the GeForce GTX 570 SLI tandem we tested earlier in scalability. Its performance increases by an average 70% over that of the single card. The maximum performance boost is as high as 105%. Its advantage over the GeForce GTX 580 and Radeon HD 5970 remains the same at 19-21%, though. Compared to that, the GeForce GTX 570 SLI tandem enjoyed an advantage of 42-43% over those cards. Thus, the GeForce GTX 560 Ti SLI delivers comfortable performance where the single GeForce GTX 580 can do the same and even fails in BattleForge and S.T.A.L.K.E.R.: Call of Pripyat.

So, a dual-GF114 graphics card wouldn’t be competitive as a flagship solution. It might be fast but only about as fast as an ordinary single-chip GeForce GTX 580.

so nvidia is in a tight spot...
dual 114 scales well, costs less and runs cooler than dual 110, BUT it reaches a perf spot their 580 already conquered months ago...

so they HAVE to go dual gf110... which means heat heat heat and some more heat... but how are they going to take care of the memory width issue? and where will they put all the memory chips? dual pcb again? phew!

[eXa]

*Cough* 1GB vram *Cough*

But still, i agree, dual GF114 aint fast enough.

[saaya]

i think they are... its just that gf110 is so fast, it gets there all on its own

[xdan]

Unfortunatelly from what i know GF104/114 supports maximum 256bytes bus memory, i wonder what more boost a 320bytes bus would do.
Could be just enough to surpass GTX 580/5970 by 30%.

[tajoh111]

awesome article on xbitlabs testing 560 SLI
http://www.xbitlabs.com/articles/vid...i_7.html#sect0


so nvidia is in a tight spot...
dual 114 scales well, costs less and runs cooler than dual 110, BUT it reaches a perf spot their 580 already conquered months ago...

so they HAVE to go dual gf110... which means heat heat heat and some more heat... but how are they going to take care of the memory width issue? and where will they put all the memory chips? dual pcb again? phew!

I think it would be more useful as a card, if NV clocked a gf110 card up to 1ghz mhz(950 has been done with some stock coolers) and used 1.4ghz memory and used 128TMU as original specified as texture and memory bandwidth seem to be a limiting factor on the gtx 580. You would have a card that is undoubtly faster than a 5970 by 15% or so and be a single chip.

Hardwarecanucks clocked the gtx 580 up to 950 and it used 40 more watts

http://www.hardwarecanucks.com/forum...dition-13.html

so 1 ghz seems possible with great binning and a better cooler. So a card that is about 25-30% faster than a gtx 580, with 3gb of ddr 5 at 650 dollars would be fair I think.

For me I would rather have a single chip that is 85% of the speed, than a dual chip that is 100%. AMD driver support for the 4870x2 is non-existance(and their performance is getting worse with newer drivers) and I am not sure NV is better for their dual cards.

[DTU_XaVier]

So what everybody's saying is that both companies will pretty much be stuck performance-wise until they can finally transistion to a newer transistor-tech?
I don't see them re-designing chips on the current 40nm node, so I guess they'll have to suck it up, that they've hit the ceiling... And if your current single-chip solution is pretty much using all of the 300w max, and it's as efficient as possible, a handicapped dual-chip card just doesn't make sense, as you would have to cut the chip in half performance-wise to fit two of them on a card... And then you're just back to square zero...
I'm looking forward to seeing what NVIDIA and AMD will do with their dual-chip solutions, but I think the reason we haven't seen them yet is the same: They can't make them perform admirably better than their single-chip counterparts...

Best Regards

[saaya]

Yeah 4870x2 was terrible when it came out driver wise... crisis was an artifact adventure... no idea if its better now
In my experience nv is doing a slightly better job with dual gpu drivers but I'd still prefer a single gpu, hill

But about tdp... when you push silicon to the limit you get 10% extra perf for maybe 20% more power... IF that's the case with gf110 then maybe a dual gpu card is able to get more performance out of the 300 w limit... if NVIDIA can pull this off I'd be very impressed!

[kadozer]

Not that I know of.

I believe boards which fell under the HPCES v1.0 specification (remember, the PCI-E 2.0 base spec never included 225W - 300W inputs) will now be incorporated directly into the PCI-E 3.0 base spec. However, there weren't any increases in power delivery limitations.

This basically means that companies producing higher-end cards will no longer need certifications for BOTH HPCES and PCI-E but the maximum remains 300W.

Damn that's disappointing. I was hoping to see Nvdia/AMD not have to tip toe around the 300W limit to unleash some real future video card beasts. I know there are a few limited exceptions like the past limited edition dual GPU from NVidia, but not appealing from a financial point of view.

[saaya]

yeah i dont get it either...
its not like the 300W limit actually limits power consumption of systems anyways...
people who want extra power get up to 4 300W cards...

actually it would be more power efficient to allow a single vga with 400W then, cause then people who want ultra performance would run 2x400 instead of 3x300 or 4x300...
and people who want this kind of power overclock anyways, so then the limitation doesnt matter at all...

300W is clearly not enough and limiting cards... so im surprised why nvidia and amd are not pushing for more...
think about it, if there wouldnt have been a 300W limit then nvidia could actually have launched a GTX480 Ultra with 512sps and a 350W tdp and a massive heatsink...

and ati could have launched the 5970 at full 5870 clocks instead of clocking it down...

ah well... if ati and nvidia dont care...

[Kimofil]

ok so the heck with the standards then right ... no standards = no compatible parts for us ... proprietary solutions from every company ... and this makes them cost alot more etc...


but hey your right .. having limits on things to a certain degree is bad ....


if you do think so im sure you must have an asus ares ... and an asus mars gpu or two sitting somewhere right

Yeah they must remove this stupid limit for practical reasons.
Overclocked cards consume more than 300W anyway....

[csatahajos]

From what little I know the reason that they won't raise the 300W limit is the OEM market. 300W heat is alreadyy pushing the nevelope regarding the amount of heat that a normal medium sized case can handle, not to mention that it is per PCI-E slot. OEM certifications (user safety, fire protection, shock protection, whatever) would be impossible to get with higher limits. Just imagine if soemone burns his/her fingers while touching the outside of the case which can get really hot with 1 or 2 "heaters" inside.

Also most OEM machine VRMs and other stuff are built to these limits to remain cheap.

We xtreme guys are only 1% or even less of the total market, the big business is in the OEM market, they are the ones ordering thousands of parts.

Also as far as I can tell for both AMD and Nvidia the discrete graphics card market's incomes are decreasing each and every year and now with the intorduction of APUs like Fusion it will get even smaller as 90% of this ever limited income is coming in from low/medium end cards, not from the beasts that most of us here own.

So unless the OEMs change their minds or AMD/Nvidia seriously lower there prices (neither likely to happen) the PCI-E powerlimit will remain.

[zalbard]

Yeah, the heat restriction makes sense.

so they HAVE to go dual gf110... which means heat heat heat and some more heat... but how are they going to take care of the memory width issue? and where will they put all the memory chips? dual pcb again? phew!

Going to be one hot card, yep. And memory chips? Just put them on the back of the PCB... The PCB is going to be bigger than 580's one, anyway...

[Olivon]

and where will they put all the memory chips? dual pcb again? phew!

http://bbs.expreview.com/thread-39523-1-1.html

edit : already posted by cold2010 page 3, sorry

[saaya]

http://bbs.expreview.com/thread-39523-1-1.html

edit : already posted by cold2010 page 3, sorry

That's gonna be an expensive pcb I think...
Must be many layers and lots of smd parts on both sides...
Wonder how much this will cost... at least 699 probably 999

[SKYMTL]

There is a way for NVIDIA and AMD to overcome issues with the PCI-E certification: allow their board partners to do the leg work. We've seen this done quite successfully with the ASUS Ares, ASUS Mars, Sapphire HD 4850 X2, XFX HD 5970 Black Edition, etc.

All they have to do is send a set of design parameters and specifications to companies who are able to implement their own solutions from the ground up and let them deal with the rest. It's a money-saving proposition, allows NVIDIA / AMD to introduce class-leading products and skirts the certification OEMs want attached to products.

[Olivon]

I won't be surprized if Asus brings an Ares like with 2 GF110, full specs, with 6 GB ram and 3X8pin power supply

[dpa]

I won't be surprized if Asus brings an Ares like with 2 GF110, full specs, with 6 GB ram and 3X8pin power supply

loving the idea

[Sn0wm@n]

exactly .. asus evga etc.. are the ones who are suposed to go beyond the limitations with their special omgwtfbbq edition gpus .. not amd or nvidia ...

[Sam_oslo]

Since partners got total freedom to do whatever they want with GTX560, is there any hope to see a tipple-GTX560 on a single PCB from some partners?. That could beat the hell out of 6990 too, LOL.

[zalbard]

Since partners got total freedom to do whatever they want with GTX560, is there any hope to see a tipple-GTX560 on a single PCB from some partners?. That could beat the hell out of 6990 too, LOL.

The PCB would be too large, heat and power draw - unmanageable.
Also, it doesn't make much sense to use a lot of smaller GPUs instead of a few bigger ones because you run into scaling issues.

[Calmatory]

yeah i dont get it either...
its not like the 300W limit actually limits power consumption of systems anyways...
people who want extra power get up to 4 300W cards...

actually it would be more power efficient to allow a single vga with 400W then, cause then people who want ultra performance would run 2x400 instead of 3x300 or 4x300...
and people who want this kind of power overclock anyways, so then the limitation doesnt matter at all...

300W is clearly not enough and limiting cards... so im surprised why nvidia and amd are not pushing for more...
think about it, if there wouldnt have been a 300W limit then nvidia could actually have launched a GTX480 Ultra with 512sps and a 350W tdp and a massive heatsink...

and ati could have launched the 5970 at full 5870 clocks instead of clocking it down...

ah well... if ati and nvidia dont care...

Right. AMD and Nvidia surely would sell more if the PCI-E spec allowed +300 W cards.

Those monster cards probably sell with loss, but have an marketing impact so they actually increase sales across the board. Thats the sole reason why the cards exist. The market is non-existent and half of the cards are sent around/given for free(press samples, competition prizes) anyway to gain as much PR as possible.