NVIDIA GeForce 4XX Series Discussion

[mastrdrver]

US is already at the limit of the standard outlet, and power companies will not adjust just because a Computer draws way more power than an Oven does.

Screw the oven, my PC needs that outlet. Who uses an oven now a days anyway?

If Fermi turns out hot and power hungry like all reports suggest, then that just shows that the TSMC 40nm is leaky for this design (or nVidia didn't do their homework). While they may appear to be heading backwards are far as power consumption goes, it just because of bad design and/or process. Together they have made a "perfect storm" for Fermi.

All mention of TSMC 28nm is that it is better which will probably translate to Fermi taking flight on both power usage and performance (compared to 40nm Fermi). Sort of like the little jump that nVidia got from moving GT200 to 55nm from 65nm launch.

 

[TheMailMan78]

no your wires and plugs can likely handle more than the circuit is (in fact it has to handle more to be able to pass code). The bottleneck is the circuit breaker itself. You can always talk to a local electrician, many times 15 amp circuit breakers are put in normal rooms because the builder didn't anticipate that you would need more than that. In these cases there a 50% chance that you can upgrade the circuit breaker and be fine. Since it can go either way, ask a local electrician before trying anything rash.

you know t's sad, I came here to talk about fermi and instead have to resort to electrical talk because there's nothing new about fermi out. sigh.

My father was an Electrical contractor for 30+ years. When I first got into computers he ran a dedicated line from the breaker box to my room. Thats always another option too. Of course he did it for free so it might be pricey for anyone else but its an idea.

 

[Steevo]

Screw the oven, my PC needs that outlet. Who uses an oven now a days anyway?

If Fermi turns out hot and power hungry like all reports suggest, then that just shows that the TSMC 40nm is leaky for this design (or nVidia didn't do their homework). While they may appear to be heading backwards are far as power consumption goes, it just because of bad design and/or process. Together they have made a "perfect storm" for Fermi.

All mention of TSMC 28nm is that it is better which will probably translate to Fermi taking flight on both power usage and performance (compared to 40nm Fermi). Sort of like the little jump that nVidia got from moving GT200 to 55nm from 65nm launch.

Leakage is caused by design, not by process. If it were the case the 5XXX series would leak more too, they don't.

Leakage is actually good for massive overclocks it allows more volts to be pushed. But for everyday people running air or water it is no good.

Perhaps Nvidia is waiting for a 32, or 28nm revision to make this work for the power requirements? It is possible....

 

[cadaveca]

Steevo, HD5XXX IS a leaky chip as well. It DID NOT meet the design goals. The low voltage used on 40nm, added in with the increase in temps when volts are increased, says a fair bit about that.

Also, I linked an article many moon ago with one of nV's engineers stating that what he required from TSMC was ZERO leakage, but was not getting it.

Silicon is a semiconductor...all silicon leaks as temps increase...it's the nature of the device. Yes, the design can make the issue worse, but that isn't always the lone factor.

 

[Goodman]

no your wires and plugs can likely handle more than the circuit is (in fact it has to handle more to be able to pass code). The bottleneck is the circuit breaker itself. You can always talk to a local electrician, many times 15 amp circuit breakers are put in normal rooms because the builder didn't anticipate that you would need more than that. In these cases there a 50% chance that you can upgrade the circuit breaker and be fine. Since it can go either way, ask a local electrician before trying anything rash.

you know t's sad, I came here to talk about fermi and instead have to resort to electrical talk because there's nothing new about fermi out. sigh.

The raison for 15amp breaker on 110/120v is for fire safety because wires would not old on much more & can heat up & caught fire wich here in canada/québec your assurance won't cover you if your house caugh in fire if you have breaker higher then 15amps on 110/120v (14-2 wire) that is why it's illegal

 

[ToTTenTranz]

Dudes, those 120 and 240V y'all talking about the power sockets are alternate current values.
Every component in a PC uses only DC values.

In order to convert the AC voltage to the DC voltage that the PSU can actually use, you need to use the RMS values, which is about 85V DC for 120V AC and 170V DC for 240V AC.


http://en.wikipedia.org/wiki/True_RMS_converter



With a 15A breaker and using a 120V power socket, the maximum wattage you can retrieve is 85*15A = 1275W, which is why any power supply over 1000W is a complete joke.

Even with a high 85% efficiency, pulling 1000W from the PC would mean pulling 1000/0.85 = 1177W from the power socket, and that's already awfully close to burning a fuse somewhere.

So that's why a graphics card with over 300W TDP is actually a bit dangerous.
It's one thing to have a 1200W PSU to backup the e-penis, but actually needing 1200W in a computer setup is a whole other issue.

There's no way they can make a graphics card with two Fermis at the moment, and apparently they won't be able to do it any time soon.


EDIT: That said, it's highly unlickely that the Fermi cards will support Tri-SLI.

 

[GC_PaNzerFIN]

EDIT: That said, it's highly unlickely that the Fermi cards will support Tri-SLI.

I have already seen 2 SLi connectors. There is 99.9999999999% sure 3-way SLi support.

Only one who has facts of its power consumption and heatload is NVIDIA.

 

[ToTTenTranz]

I have already seen 2 SLi connectors. There is 99.9999999999% sure 3-way SLi support.

Only one who has facts of its power consumption and heatload is NVIDIA.

What you've seen is prototype cards, not production models.

Yes, that demo at CES was running 3 Fermis in SLI, but most probably because they were downclocked to cope with instability (as the demo crashed in the end).

Nonetheless, we'll only be sure about Tri-SLI support when the graphics cards are out.

 

[mastrdrver]

Leakage is caused by design, not by process. If it were the case the 5XXX series would leak more too, they don't.

What? So Intel has a dedicated line for XE chips? I don't think so.

Minimum leakage is always the goal since the chip will use level voltage, run cooler, and last longer. I won't disagree on the overclocking thing as I've heard it mentioned several times on XS that high leakage chips are good for overclocking at insane levels, but they run hotter. You also need extreme cooling to keep the chips cool at those levels/volts.

Look at the 5870. While being cool, it still runs pretty warm for 1.1v. Push it up to 4870 volts, about 1.27v, and temperatures get similar to that of the 4870. Fermi has a die size slight smaller than GT200 at 65nm. Then proceeded to stuff twice the transistors in that roughly same area. You think a GTX 280 was warm? People have not seen anything yet.

Realize: Fermi is not big. nVidia has built ones this big before this is nothing new. Also, if you design Fermi for leakage, you have huge problems since nVidia is marketing it mainly as a GPGPU. Hence, the target audience are business that would rather save power than increase it. Not to mention a plethora of other things.

I really do hope nVidia does pull this off with the Fermi. It will really decimate the 5 series in heavy tessellation and make AMD really provide a competitive 6 series. Its a good thing. If the 6 series doesn't perform, were at G80 all over again for DX11. I don't think anyone wants that.

 

[ToTTenTranz]

Realize: Fermi is not big. nVidia has built ones this big before this is nothing new.

The original GT200 was huge, the biggest GPU the world has ever seen. Fermi will be roughly the same size and you say it isn't big?
Besides, the GT200 came out in a fully matured process.
nVidia had launched an entire lineup of 65nm GPUs, and even an IGP.
What has nVidia launched in 40nm until now? GT215, GT216 and GT218. All launched 4 months ago.

There's a reason why nVidia has missed the Christmas season, and it's definitely something that's going wrong.
So don't convince yourself that this is "nothing new" to nVidia. It is, but let's just hope they solve it soon.

I really do hope nVidia does pull this off with the Fermi. It will really decimate the 5 series in heavy tessellation and make AMD really provide a competitive 6 series. Its a good thing. If the 6 series doesn't perform, were at G80 all over again for DX11. I don't think anyone wants that.

Better at heavy tesselation? ATI has been doing tesselators for over 5 years and all of the sudden nVidia beats them at their own game in their first try?
Now that's just cheap futurology or wishful thinking.

 

[GC_PaNzerFIN]

Better at heavy tesselation? ATI has been doing tesselators for over 5 years and all of the sudden nVidia beats them at their own game in their first try?
Now that's just cheap futurology or wishful thinking.

Don't be so sudden to judge. The fact they have had no competition in tesselation at all does not mean no one can beat them,.

 

[ToTTenTranz]

Don't be so sudden to judge. The fact they have had no competition in tesselation at all does not mean no one can beat them,.

I'm not judging, I'm not assuming that the HD5870's tesselator will be faster.

It's just that mastrdrver's assumption that Fermi "will really decimate the 5 series in heavy tessellation" is a bit ridiculous at this point.

 

[eidairaman1]

ah who cares the parts are still smoke and mirrors, We shall be the judges of the parts ourselves (those who pick them up).

 

[GC_PaNzerFIN]

I for one will put one or two of these against the HD 5870. Now that the card seems to be working better wit these beta drivers found from somewhere, this wait isn't so long for me anymore.

 

[Wile E]

The original GT200 was huge, the biggest GPU the world has ever seen. Fermi will be roughly the same size and you say it isn't big?
Besides, the GT200 came out in a fully matured process.
nVidia had launched an entire lineup of 65nm GPUs, and even an IGP.
What has nVidia launched in 40nm until now? GT215, GT216 and GT218. All launched 4 months ago.

There's a reason why nVidia has missed the Christmas season, and it's definitely something that's going wrong.
So don't convince yourself that this is "nothing new" to nVidia. It is, but let's just hope they solve it soon.




Better at heavy tesselation? ATI has been doing tesselators for over 5 years and all of the sudden nVidia beats them at their own game in their first try?
Now that's just cheap futurology or wishful thinking.

DX11's tessellation is different than what ATI was working on in past years. Their former experience doesn't help them currently.

 

[ToTTenTranz]

DX11's tessellation is different than what ATI was working on in past years. Their former experience doesn't help them currently.

But nVidia's complete lack of experience does?

 

[Wile E]

But nVidia's complete lack of experience does?

No, but it doesn't hurt them compared to ATI either. ATI's Tesselation experience as it's related to computer gaming = 5k series. This is their first release, and fermi will be nVidia's. The playing field is level until ATI's next release.

 

[ToTTenTranz]

No, but it doesn't hurt them compared to ATI either. ATI's Tesselation experience as it's related to computer gaming = 5k series. This is their first release, and fermi will be nVidia's. The playing field is level until ATI's next release.

I digress. DX11's tesselator is based on the X360's tesselator, which is the exact same unit (according to most interviews) that we find in HD2000 to HD4000 cards.

And maybe not in computer gaming per se, but AMD/ATI has been showing off real time tesselation demos for quite some time (4 years).
How many tesselation demos have we seen from nVidia?

 

[mastrdrver]

The original GT200 was huge, the biggest GPU the world has ever seen. Fermi will be roughly the same size and you say it isn't big?

Compared to Intel, AMD, and ATI (recently) yes it is huge. Compared to what nVidia has done in the past, and ATI, its nothing new. I made that comment on the fact that every article I read on the thing comments on the "hugeness" of the chip. You're correct that 65nm was mature though and I agree. Though, seeing as nVidia did make the 40nm chips you mentioned, I'm lean toward the thought that the reason nVidia is late is that 40nm is leaky, compared to pass processes. I do not think that TSMC 40nm will ever (comparably) become as mature as 65nm was.

Better at heavy tessellation? ATI has been doing tessellators for over 5 years and all of the sudden nVidia beats them at their own game in their first try?
Now that's just cheap futurology or wishful thinking.

Several on the B3D forum were talking about how the increase in geometry computation of Fermi will greatly help it as tessellation scales higher or something along those lines. I need to go back and reread that since its been a month or two since I look at the Fermi thread they have over there.

Great thread to leave your head spinning from.

 

[Wile E]

I digress. DX11's tesselator is based on the X360's tesselator, which is the exact same unit (according to most interviews) that we find in HD2000 to HD4000 cards.

And maybe not in computer gaming per se, but AMD/ATI has been showing off real time tesselation demos for quite some time (4 years).
How many tesselation demos have we seen from nVidia?

Everything I have read has suggested that DX11's tessellation engine is NOT based on the 360 unit. It's completely different. If that's the case, the 360 style of tessellation doesn't help ATI at all.

 

[Steevo]

Leakage is a byproduct of the GPU design, packing more transistors in a given area by thinning the insulating areas can allow electrons to move throughthe lattice. Prescott was a very high leakage part untill they could move to the 65nm preccess as the 90nm required much more electricity to work, and to keep the number of transistors on the die size required meant the insulators had to be thinned. the move to 65nm menat the transistor shrunk, while they could add thickness in the insulating boundry. It applies the same here, to fit the required number of transistors onto a certain size chip they have to sacrafice the insulating area thickness, contributing to thickness. They would have known this in the design process, but untill they had working dies and could get a reasonable sample set they woudn't know the exact amount of leakage, and voltage required for a working minimum number of dies per wafer.


Thus the reason they will be releasing the lower end part at the same time, die defects and more than likely the dies for the 480 are going to be very throughly binned for acceptance. This isn't my first rodeo with die shrinks, leakage, and the waiting. I owned a couple series of Prescott. Stil have them working today as a fact.

http://www.pcworld.com/article/113866/intel_changes_prescott_plans.html

Steevo, HD5XXX IS a leaky chip as well. It DID NOT meet the design goals. The low voltage used on 40nm, added in with the increase in temps when volts are increased, says a fair bit about that.

Also, I linked an article many moon ago with one of nV's engineers stating that what he required from TSMC was ZERO leakage, but was not getting it.

Silicon is a semiconductor...all silicon leaks as temps increase...it's the nature of the device. Yes, the design can make the issue worse, but that isn't always the lone factor.

I wasn't aware of a target voltage onthe 5XXX series, and it really isn't a leaky chip. Thus the reason it does run so cool. Perhaps you have read a review where one was in a oven while it was benig tested, but the new 5XXX's are runnign about 20C cooler than the previous generation while providing 25% more work, and the idle power savings is huge, in fact look at the charts they show a 45% improvement in performance per watt over the 4XXX series.

 

[department76]

Dudes, those 120 and 240V y'all talking about the power sockets are alternate current values.
Every component in a PC uses only DC values.

In order to convert the AC voltage to the DC voltage that the PSU can actually use, you need to use the RMS values, which is about 85V DC for 120V AC and 170V DC for 240V AC.


http://en.wikipedia.org/wiki/True_RMS_converter



With a 15A breaker and using a 120V power socket, the maximum wattage you can retrieve is 85*15A = 1275W, which is why any power supply over 1000W is a complete joke.

Even with a high 85% efficiency, pulling 1000W from the PC would mean pulling 1000/0.85 = 1177W from the power socket, and that's already awfully close to burning a fuse somewhere.

So that's why a graphics card with over 300W TDP is actually a bit dangerous.
It's one thing to have a 1200W PSU to backup the e-penis, but actually needing 1200W in a computer setup is a whole other issue.

There's no way they can make a graphics card with two Fermis at the moment, and apparently they won't be able to do it any time soon.


EDIT: That said, it's highly unlickely that the Fermi cards will support Tri-SLI.

sorry to inform you, but 120VAC on a wall outlet IS the rms rating. ac power is V*I*powerfactor, and since most psus have active power factor correction, that usually stays above 95% anyhow. that leaves damn near 1800 real watts at the disposal of any psu on a since ac line in the US.

 

[TheMailMan78]

This thread has lost all direction. Its become an ATI vs Nividia thread all over again. I wish we could all have some kinda Braveheart style nerd battle. Dan could play Longshanks and Ill play Stephen the crazy Irishman. I have no idea about the rest of the casting but the battle would be of epic nerdom!

 

[Lionheart]

Talking about tesselation and DX11, wats this crap about if you buy a DX11 video card, your games will run faster in DX11 and more multithreaded, more advertisement BS!


Alright then, I'll be William Wallace!

 

[TheMailMan78]

Talking about tesselation and DX11, wats this crap about if you buy a DX11 video card, your games will run faster in DX11 and more multithreaded, more advertisement BS!


Alright then, I'll be William Wallace!

I'm thinking Wile E, Erocker or Mussles for that part.