The GeForce FX may be a true sleeper

[noko]

Both the core and memory are running extremely hot!! 140F plus at full load. Even with the dust buster. If the card was running cool at 500mhz I wouldn't think the GPU had much left in it. In this case there is alot of room for better cooling for both the ram and GPU - - > I believe a huge potential for overclocking exist if someone figures out a more effective method of really cooling this baby. Its weakness points to some very positive possibilities.

[digitalwanderer]

But don't you think that nVidia would have thought of that too and mebbe offered a water-cooled option? It ain't like it's that esoteric anymore, albeit not exactly common either.

I don't know, I'm holding a "wait-n-see" attitude towards anything like that....and letting someone else do it first. ;)

[StealthHawk]

is this some kind of sick joke spawned from reverse land?

a hotter card has more room for overclocking?

[ragejg]

ha!
:D

"the hotter this baby gets, the more the silicon phase shifts, increasing conductivity since all crossed traces are crossed intentionally... @ 200 F this baby hits 750mhz! So if you crank your AA & AF off, and set mipmap detail to lowest, you can just imagine the arifacts are snowflakes, and everything'll be all right in your sea of soft N64ish textures."
:D
:)
i get what noko's saying, but i just couldn't help that blurb... it's stealthhawks fault for makin me think about "reverse lands"...

...start button on the right hand side? That'd be CRAZY!

[Filburt]

Hmm...I'm not sure I entirely agree, noko. I'm not surprised at the heat due to the absence of low-k increasing capacitance internally...thus requiring them to ramp up the voltage higher as the chip cannot operate efficiently enough internally for the lower voltages. Raising power means the already fairly high resistance (which I'm guessing is even higher as to avoid electron leakage between interconnects?) is going to cause a massive amount of heat.

I have a feeling this thing...even with LN2 cooling...wouldn't want to see beyond maybe 575Mhz.

[Filburt]

Originally posted by Harnagel
They've been running the core at 650Mhz and higher for a while now at Nvidia. The main concern seems to be the RAM. I think there is some room for a good overclock with water and a bit of work. but I'm not one to go trying such things either...

Are you absolutely sure about the 650mhz thing? You say that like it's fact...while when I assess this from an engineering standpoint...it sounds really far fetched. 650mhz on a 0.13 micron die with that many transistors...and no low-k...is erm...not very likely at all. In fact, it's atrociously unlikely...

[undercover]

Wasn't there some guy a couple of years ago who water-cooled his GF3 ? Let him loose on a GFFX!

[Steppy]

quote:
--------------------------------------------------------------------------------
Originally posted by Harnagel
They've been running the core at 650Mhz and higher for a while now at Nvidia. The main concern seems to be the RAM. I think there is some room for a good overclock with water and a bit of work. but I'm not one to go trying such things either...
--------------------------------------------------------------------------------


Originally posted by Filburt
Are you absolutely sure about the 650mhz thing? You say that like it's fact...while when I assess this from an engineering standpoint...it sounds really far fetched. 650mhz on a 0.13 micron die with that many transistors...and no low-k...is erm...not very likely at all. In fact, it's atrociously unlikely... Well, I'm sure every run of CPU's/GPU's has a few "golden" parts that run exceptionally high. Producing thousands of them however is an entirely different matter. I'd bet that like 1 out of 100 run at that speed at best. I'm sure ATI has a couple 550 Mhz R300's and AMD has a few 3.0 gig athlons, and Intel has a few 5.0 ghz P4's laying around too...

[noko]

As you cool the chip the resistance for the traces go down --> less of a voltage drop internally of the chip. What this means is that the signals/data transfers internallyare less degraded. I am pretty sure we are looking at a good to great overclockable chip here if properly cooled. Take your cpu and heat it up to 140F plus, or your system ram see what kind of overclock you will get, more exactly if it will even run at stock speeds.

To properly cool the GF FX gpu would be probably too expensive for any card manufacturer to market successfully. The ram though I think is the most limited component of this design but its high heat also can be controlled better for better performance.

I am sure this will be found out in the next couple of months, my bet is that the N30 core will exceed 600mhz if properly cooled. the ram will also go to 550mhz plus is my best guess.

This is a new breed of card for the high end, some people may actually push it to the stratosphere as well.

[-=DVS=-]

Originally posted by noko
This is a new breed of card for the high end, some people may actually push it to the stratosphere as well.

I totaly agree with you its cheap to build JET ENGINE with few mods we can fly to space with that thing :D
Such a small thing generating so much heat it should not be a problem ;)


Sarcasm at its max

[netape]

Am I the only one who thinks that these "Jet engine" comments are getting a little old? :confused: Maybe it's just me :o

[-=DVS=-]

Originally posted by netape
Am I the only one who thinks that these "Jet engine" comments are getting a little old? :confused: Maybe it's just me :o

Offcourse its funny when some people make comment like this

some people may actually push it to the stratosphere :rolleyes:

Don't worry i am sure in few days GFFX will be forgotten and we all move on to something else to blame/speculate

[noko]

Just remember you heard it first here. Cool it better, play it faster dude. I will be fooling around with my Tyan Tachyon G9700 Pro :) but would love to play around with a GF FX as well. My rig could definitely cool it down significantly, ~40c cooler to be more exact. Without even changing the cooling system of the GF FX:confused:.

[Filburt]

Originally posted by noko
As you cool the chip the resistance for the traces go down --> less of a voltage drop internally of the chip. What this means is that the signals/data transfers internallyare less degraded. I am pretty sure we are looking at a good to great overclockable chip here if properly cooled. Take your cpu and heat it up to 140F plus, or your system ram see what kind of overclock you will get, more exactly if it will even run at stock speeds.

To properly cool the GF FX gpu would be probably too expensive for any card manufacturer to market successfully. The ram though I think is the most limited component of this design but its high heat also can be controlled better for better performance.

I am sure this will be found out in the next couple of months, my bet is that the N30 core will exceed 600mhz if properly cooled. the ram will also go to 550mhz plus is my best guess.

This is a new breed of card for the high end, some people may actually push it to the stratosphere as well.

Noko...I still don't see how you are determining it will be a great card for overclocking. The chip is as overclocked at 500mhz as the 9700 Pro would be. In fact, it's *harder* to clock up a 0.13 micron without low-k than it is a 0.15 in some ways. What I'm looking at is that any cooling solution you can make for this card...when applied to the 9700 pro...would net a much more powerful result than with the GeforceFX. I'm well aware of what happens when you get the heat down...I've outlined it plenty of times. Unfortunately, this GPU is almost *less* scalable than the R300 because of the absence of the low-k dielectric. The above problem with performance vs. a similarly cooled 9700 I think is also very important to consider.

[Shinri Hikari]

I think the "Jet engine" crack is cheap, done to death, and tasteless.:clever:

[noko]

Once again, take a .13uc P4 and let it operate at 140F and try to overclock it. Lets say a 3.06ghz version. You probably won't get to far, if at all. Now cool that baby down to 50F or below and now that P4 is breaking 4ghz!! This is all basic overclocking. My R-300 does about 380mhz comfortably on near ambient tempertures which is kinda hard for me to do because I have a 5,000 BTU A/C unit blowing over the card. I drop the temperture down to -12c blowing on the card and I only get 20mhz higher. Why? I am not so much heat limited but core limited. Meaning my core is just not going to go to much faster as it is.

Now take the GF FX or NV-30 and you crank it up to 500mhz and it is just plain out producing some heat and making it hotter then just about any other computer part out there. Problem is the proper cooling would be too prohibited to market, so Nvidia does the next best thing they can. Regular cooler 400mhz, a larger cooler and this NV30 now does 500mhz. Still a hell alot of room for better cooling is possible meaning a likely very good overclock remaining.

This may all be a mute point based on the real limitation of the design, the 128bit memory bus. If the memory is cooled sufficiently then I also believe it will be able to go up high enough to relieve somewhat the weakness of the design. Some real serious and good Nvidiots will have to explore these options, but I believe they are there. The rumors of Nvidia labs having N-30's at 650mhz isn't to far fetched now that I look at what the chip is doing presently.

[Jethro]

I like your take Noko very well put indeed.

yeah maybe this card isnt a 9700 killer and not for everyone but it is blazing fast and will be sold in dustbuster free models that may become the next ti4200's and for us modding geeks that is an exciting prospect.

Im really looking forward to playing with one of these new toys

[Filburt]

Noko, the P4 0.13 process utilizes low-k.

The reason they are able to overclock as high as they do when bringing the temperature down is BECAUSE it utilizes low-k. Omission of low-k not only increases capacitance among other things...it also removes a rather important block against electron leakage among interconnects.

When thinking about overclocking the NV30...a better example would be to use the Thoroughbred-A core...and even then it'd be optimistic.

Again, the P4/NV30 analogy is simply fallacious. I don't mean to sound harsh, but to use such an analogy is to ignore a very crucial point of the design that will *prevent* such a scenario in your analogy from actually occuring.

[Jethro]

Filburt do u really think ther wont be any more revisions of this chip dood? G4 had a1 and a2 didnt it?

[Filburt]

Originally posted by Jethro
Filburt do u really think ther wont be any more revisions of this chip dood? G4 had a1 and a2 didnt it?

They can revise it all they want. It won't help anything until they utilize low-k. Given TSMC's process isn't the simpler spin-on method...it would likely require something more than *minor* revision to be able to implement low-k on the NV30. By the time they manage to do it...they'll be busy with the NV35 launch. Also, it would cost them a small fortune to do it...a small fortune that wouldn't pay off enough in the end to make it economically sound. It's not like they are going to do it just to please nVidia fans...regardless of it costing them money they wouldn't recover.

Yes yes, I already factored in the card selling better to OEMs and such due to removal of the FXFlow at 500mhz when low-k is implemented. But you guys have to remember by the then R350 will be out...and it would require then large sums of money invested in low-k, as well as throttling the clock even higher. You may see the need of the FXFlow yet again at the clock speeds they'd need to compete with the R350! This is completely senseless for them to do when they could simply take the money and put it towards keeping the company afloat as well as developing the NV31, 34, and 35.

[noko]

Filburt,

The comparison is based on what heat or temperture can do to overclocking a chip. Remember we are not talking about gigahertz here at all where electron leakage, capacitance and what ever else you want to talk about is a major problem as in a P IV. We are talking about low Pentium 3 speeds where low-K wasn't used. The problems you are mentioning I don't think is a factor at the speeds the NV30 has except for the high heat output due to the increase count of transitors. Which if cooled better would allow (I fully believe) a much higher overclock. Now what would a 700mhz NV30 do in performance? With a significantly increase ram speed?

If the GF FX 5800 can be picked up in the low $200 then that would be a very fun card to play around with. The core should be the same as a GF FX Ultra minus the cooling.

Once again I think some of these possiblities will be explored in the next 2-3 months where the full potential (which isn't really known now) will be realized. In addition I expect a 15-30% performance increase in benchmarks when the drivers mature more. Nvidia has always increase performance (better then anyone else I've notice) within 6 months from the release of a new GPU.

[YeuEmMaiMai]

problem is you are dealing with roughly 3x the transistor count so do not expect the speeds to ramp up much at all hardocp onlt managed 540Mhz with their card about 8% OC from stock while radeon 9700 Pro on the other hand managed to get to 400Mhz on air cooling. and there was one gut that got 500Hhz out of LN cooling as well

[Filburt]

Originally posted by noko
Once again...

Once again, you neglect the very important problem of the omission of low-k dielectric in the NV30. Your model for overclocking, the P4, is an entirely inaccurate representation of how an NV30 will be able to overclock. Again, a better comparison would be to the Thoroughbred-A core...where people were finding...no matter how cold they got it...they could not overclock it very much before it stopped working. I believe this was due to electron leakage at the gates or something of the sort. The NV30 will suffer from electron leakage as well...regardless of how cold you get it...once you start trying to really push the clock up.

The non-Ultra GFFX will hit 500mhz only with very elaborate cooling (even more elaborate than the FXFlow). It's likely you'll need watercooling to hit 500/500...and by then, you've exceeded the price of the Ultra itself because you will a very custom...and very expensive...water cooling system on it. So it essentially becomes you've bought a non-ultra GFFX just so you can have an nVidia card...not so you could have a faster, nor even really better card. That *seems* like a waste of money to me...but perhaps I don't understand the "brand loyalty above all else" philosophy of purchasing products. It just seems senseless when I could take the same cooling system and use it to overclock the R300 to such a point that it would outperform the NV30 counterpart by a rather wide margin, while still costing me less.

[noko]

OK Filburt, I will change my model then and throw out the low-dielectric stuff.

Take a Pentium III rated at 500 mhz and heat it up to 140F and see how much of an overclock you get. . . Now cool it down to less then 50F. Hitting 800mhz was usual in that case.

Still too many unknowns until tried, yep more transitors but that mainly gives you the heat problem. I expect good cooling (much better then the make shift dustbuster) will yield 650+mhz capability if not more. Once again that is a guess. The problems you are specifying relates more for cpu's in the Multi-gigahertz range and not sub gigahertz range of the NV-30. Heat is limiting the speed of the NV-30 in the GF FX Ultra. It is written all over it.

[Filburt]

Originally posted by noko
OK Filburt, I will change my model then and throw out the low-dielectric stuff.

Take a Pentium III rated at 500 mhz and heat it up to 140F and see how much of an overclock you get. . . Now cool it down to less then 50F. Hitting 800mhz was usual in that case.

Still too many unknowns until tried, yep more transitors but that mainly gives you the heat problem. I expect good cooling (much better then the make shift dustbuster) will yield 650+mhz capability if not more. Once again that is a guess. The problems you are specifying relates more for cpu's in the Multi-gigahertz range and not sub gigahertz range of the NV-30. Heat is limiting the speed of the NV-30 in the GF FX Ultra. It is written all over it.

But that's the point, heat isn't the only thing limiting the speed of the card. Something to note, the original PIII was 0.25 micron process. It can suffer from *fewer* problems when trying to clock up than 0.13 micron would with the omission of the low-k dielectric. 0.13 microns and lower...practically go hand in hand with low-k dielectric. Separating the two nets pretty ugly results.