The Fermi Thread - Part 3

[***Deimos***]

4 AMD GPU = 3 NVIDIA GPU

Maybe this is why reviews have such different results.. depends on AA level.

[570091D]

So many times I prediced that GTX 480 would barely OC. That nVidia would have an impossible time reaching 750Mhz.

If so many people are getting 800Mhz+, perhaps 480SP and 700Mhz had more to do with limiting power consumption than yields.

If you had chips easily able to do 800Mhz, why else would you sell them at 700Mhz?

By using LN2 cooling on a GeForce GTX 480 video card he was able to take the core clock from 700MHz up to 1165MHz, which is an amazing 66% overclock! He then used an Intel Xeon E5640 processor at 4GHz and was able to score just shy of 30,000 3DMarks in the Performance test of 3DMark Vantage

Holy Smokes Batman. Can somebody do the math on how much current must be going through those 6 voltage regulation circuits!!?

Me thinks at whatever ungodly voltage Shamino used for 1165, "normal" 1200W-1500W PSU could be not enough for the system.

At stock voltage it seems LN2 alone gets 900Mhz on the core - ie cooling seems to be the main issue

http://hwbot.org/hardware/videocard/geforce_gtx_480

i understand this is actually because tsmc isn't producing the chips properly the channels are too wide, see my post above yours. Bx cards will probably have higher stock clocks, but won't overclock as far.

[fiskov]

I just noticed you know.. None of these cards come with any of those awesome free gifts (mouse mat, mouse.. any games?) which where shown..
All they come with is that car/rocket sledge thing.. I would have thought Metro would be bundled in.. Hmm

I'm considering cancelling my pre-order now..

[570091D]

I just noticed you know.. None of these cards come with any of those awesome free gifts (mouse mat, mouse.. any games?) which where shown..
All they come with is that car/rocket sledge thing.. I would have thought Metro would be bundled in.. Hmm

I'm considering cancelling my pre-order now..

really? you're willing to pay $499 for a mouse pad?

[flippin_waffles]

4 AMD GPU = 3 NVIDIA GPU

Maybe this is why reviews have such different results.. depends on AA level.

Lot's of people going on about how 4xx supposed scaling as groundbreaking and revolutionary. I wondered how that was possible when Evergreen scales at %100 or very close in many situations.

Scaling looks pretty sad here for fermi.

[Chumbucket843]

Lot's of people going on about how 4xx supposed scaling as groundbreaking and revolutionary. I wondered how that was possible when Evergreen scales at %100 or very close in many situations.

Scaling looks pretty sad here for fermi.

straw man? i dont remember a huge discussion on multi-gpu ever. 5870 crossfire scales similarly in far cry 2. fyi 100% scaling is impossible.

[Your_Boss]

Just wondering...

Do you think the two different cards HD5870 & GTX480 are equal in performance design?

Meaning, if you cranked up the 5870's voltage for memory and core to it's stable max, but also used (let's say some equal cooling, let's use the 480's grill HS&F as an example) the same cooling for both cards, would the two cards be about equal?

Because the 480 is a little faster than the 5870(in general) but at was cost? Extra volts and heat, power, etc. So if you could crank the same voltage, etc into a 5870..would they be about the same performance?? I don't mean the MHz would have to be equal, just what the cards could take in terms of power/heat. Using same cooler, when card A is 90c (check performance) and card B is 90c (check performance), who will win? Or anyone think they will be equal?

Just askin. (Puts on flame retardant suit)

[ElSel10]

Just wondering...

Do you think the two different cards HD5870 & GTX480 are equal in performance design?

Meaning, if you cranked up the 5870's voltage for memory and core to it's stable max, but also used (let's say some equal cooling, let's use the 480's grill HS&F as an example) the same cooling for both cards, would the two cards be about equal?

Because the 480 is a little faster than the 5870(in general) but at was cost? Extra volts and heat, power, etc. So if you could crank the same voltage, etc into a 5870..would they be about the same performance?? I don't mean the MHz would have to be equal, just what the cards could take in terms of power/heat. Using same cooler, when card A is 90c (check performance) and card B is 90c (check performance), who will win? Or anyone think they will be equal?

Just askin. (Puts on flame retardant suit)

Increasing voltage alone wouldn't do anything for performance.

[Serpentarius]

while you guys are talking about gtx 480 vs hd 5870

aren't you guys forgetting 5890 and 5950? i remember it's Q1 or Q2 2010. But nevermind about that .. not be relevant anyway

[aussie-revhead]

Assuming there will be a 5890 then that should be interesting, hurry up ATI ....

[zalbard]

If so many people are getting 800Mhz+, perhaps 480SP and 700Mhz had more to do with limiting power consumption than yields.

No.
They would just go for all SPs and slightly lower clock then.

[***Deimos***]

Just wondering...

Do you think the two different cards HD5870 & GTX480 are equal in performance design?

Meaning, if you cranked up the 5870's voltage for memory and core to it's stable max, but also used (let's say some equal cooling, let's use the 480's grill HS&F as an example) the same cooling for both cards, would the two cards be about equal?

Because the 480 is a little faster than the 5870(in general) but at was cost? Extra volts and heat, power, etc. So if you could crank the same voltage, etc into a 5870..would they be about the same performance?? I don't mean the MHz would have to be equal, just what the cards could take in terms of power/heat. Using same cooler, when card A is 90c (check performance) and card B is 90c (check performance), who will win? Or anyone think they will be equal?

Just askin. (Puts on flame retardant suit)

5870 is 2B transistors and spec is 189W (actually very close - like 200W)
GTX480 is 3B transistors and spec is 250W (in a dream world - in reality more like 300W)

50% more transistors, 50% higher power. Don't know, but Fermi probably uses between 0.95V and 1.05V. Similar to 5870. Also, 5870 doesn't have any "special" transistors - both use the same ones... but the design determines the power profile. Dynamic logic typically uses more power than say DRAM.. that is why CPU chip is 100W, and DRAM chip is only 1W.

If, as you say, you increased voltage on 5870, you would need a lot to get to 300W. The 2x 6pin and power regulators might not withstand it.

But, what would that prove? You can bring a Phenom up to 300W. Same with even Pentium. Or a small piece of wire inside an argon filled enclosure.

[LordEC911]

Increasing voltage alone wouldn't do anything for performance.

He was asking, if you crank the voltage and clocks to the max for both cards and have the same cooling, where would the performance be.

[ElSel10]

He was asking, if you crank the voltage and clocks to the max for both cards and have the same cooling, where would the performance be.

Hmm well I think the 5870 already has the highest clock/voltage combination that gives AMD yields they are most satisfied with.

Although, obviously the 5870 would be faster. Rather than boosting the 5870, imagine lowering the 480 clock speeds and voltage to match the power envelope of the 5870. It's quite obvious it would be slower.

That is pretty irrelevant though.

[flippin_waffles]

straw man? i dont remember a huge discussion on multi-gpu ever. 5870 crossfire scales similarly in far cry 2. fyi 100% scaling is impossible.

Strawman? And because you don't remember one means what? There have been plenty of posts from the green team claiming exactly that. And there have also been a multitude of reviews showing crossfire scaling >100% at times.

[Luka_Aveiro]

Strawman? And because you don't remember one means what? There have been plenty of posts from the green team claiming exactly that. And there have also been a multitude of reviews showing crossfire scaling >100% at times.

Could you post where you got those concepts from? Thanks in advance.

[annihilat0r]

Strawman? And because you don't remember one means what? There have been plenty of posts from the green team claiming exactly that. And there have also been a multitude of reviews showing crossfire scaling >100% at times.

Personally I've never heard anything about SLI scaling from anyone before the Fermi launch. And after it, several sites noted scaling better than Crossfire. I think the most interesting one was from HardOCP review - that review put Fermis in the worst light possible but strongly praised their SLI performance.

[saaya]

ive read up the last 30 pages since i last posted here, pheew
very amusing read though...

lol@:
1.) chiphell benchmarks from a few days ago...
2.) the perf advantage of gtx480 over 5870 people claimed in this thread steadily decreasing from 50% to 13% in the last 30 pages
3.) high res super low fps numbers that aimed to show %tage advantages
4.) paper launch and no cards in retail for 2-3 weeks at least
5.) no 3d vision soround support in review drivers
6.) 50% perf hit on gtx480 with physix set to high in baa
7.) still no angle independant filtering, same as gt200
8.) bad scaling to 2560x1600 for the 480 (wtf?)
9.) identical perf of 480 and 5870 in vantage (wtf????)
10.) 55W idle and 85W video playback
11.) 60W higher power consumption than 5970 (295W tdp) in unigine


my impression of gtx480 vs 5870
the good:
very fast in tesselation
10-15% faster than 5870
higher perf boost from ocing on air than 5870

the bad:
+100$
+125W load
+10-15C load

the ugly:
+7dba load
identical perf to 5870 in vantage
not available

if your gaming on a display below 1920x1080 or 1920x1200 res, current cards are enough for you in 95% of all games, and you should really upgrade your screen before upgrading your vga

if your gaming on a 1920x1080 or 1920x1200 screen there isnt a big difference between a 5850, 5870, 470 or 480... they cost about the same and perform about the same, only thermals are better for ati, which means less noise, less heat, less power consumed... but i doubt many will care about this.

if your gaming on a 2560x1600 screen, your either not a real gamer or your an idiot and should downgrade your display instead of spending tons of cash on multi gpu solutions to actually be able to play the latest and greatest games without multi gpu drawbacks such as stutter, huge power consumption, noise and bugs.

the only reason i can see for buying a 480 is if you either dont like ati or catalys drivers or like nvidia or their drivers... slightly more perf you wont notice in games for a higher price and more heat... id call it a draw at best...
if your into folding at home, then a 480 or 470 sounds great too...

i totally agree with sof! its a shame that there are no water results for 480s and 470s... i could imagine that a 470 on water would get close to a 5970 if not higher, and that at almost half the cost and its a single gpu...

The driver gains are going to be insane with fermi. I have never seen a card with such inconsistent performance. Forget 10% performance gains, we are going to be seeing 20-30 percent performance gains. Especially with titles like Quake and crysis, which are typically NV favoured, the heaviest gains are going to be seen here. I have guessed that when making the latest drivers NV focus on their weakness first and will later focus on their strengths. With strong performances in games like dirt 2, hawx, battleforge and unreal, which are usually AMD favoured, this card should be killing in games like crysis and quake wars. Its not those and at this point one has to suspect drivers(assuming the recent leaked charts are true).

so if i like playing crysis and quake i should actually buy a card that is not performing well in those games because in the future it might perform very well there... lol what?

They signed NDA... I really dont know why they broke it...Maybe for some website traffic....

and how exactly would you know who signed an nda with nvidia and who didnt? 0_o

Honestly it sounds like almost everything Charlie said about Fermi has been wrong, with the exception of silicon revisions and release timeframe. He has said 5% faster than the 5870, castrated to 448 SPs for the top bin GPU with 600/1200 (core/sp) clocks, 300W TDP, 70C at idle while running at 70% fan speed, etc... All of that seems to be nowhere near reality.

448sps was for tesla cards, nvidia purposely mislead everybody about final specs and he was still very close, gtx480 consumes MORE than 300W in furmark and 260W in games, and the 70C idle at 70% fanspeed isnt too far off either, 65C at 60% iirc, and thats with low ambient temps...
charlie was right about heat, lower clocks than anticipated, lower perf than anticipated, lower mem clocks than anticipated, very late q1 launch and no 512core part. he claimed lower clocks and at some point less cores... but while he wasnt spot on, he was pretty close considering how uptight nvidia was about fermi and at the same time spread misleading info on purpose...
who was more accurate than him? who gave nearly as many hints about fermi before it came out as him?

Regardless PhysX games are superior, atleast experience wise on NV hardware, so those games will be a given regardless.

this is highly subjective... i wont argue with what some people think of physix, but can we agree that this is subjective? physix does NOT necessarily improve the game play experience... not at all...

yeah but completely disregarding the new architecture argument, even ATI has been able to bring 5800 cards' performances up by an avg of 10% just with drivers. At least the same should happen to Fermi's.

well, ati needed 6 months for this 10% boost, right? and how long as nvidia been tweaking fermi drivers? about the same 6 months... rv870 didnt tape out that much before gf100... how much was it? 2 or 3 months?

I'd like to see Shattered Horizon. A good looking game, and performs horribly.

i wouldnt call it a game... it looks somewhat nice, yes... but its more of an interactive benchmark than a game ^^

BTW, Charlie "i'm an idiot" dedknefuhrefuohreofmerijian was wrong in most of the hate he posted. The product has one flaw and that is power draw but all in all it's high clocked and a real great performer. Eat that C.

so who do you think was closer to reality? nvidia claiming 2x-2.5x perf gain over 285 and the same power consumption as a 285, or charlie claiming a minor perf bump over 5870 bundled with a lot of heat?

charlie wasnt spot on, but the overall picture he painted was pretty close to what im seeing now... Q1 is pretty much over and there are still no cards for sale, nvidia claimed NOVEMBER 2009 availability, then december 2009 availability, then january, february, march, and now MID april... once they come out they will be hot, power hungry and prices will spiral up and up... and there is no notable perf difference when actually playing games.

how about making one of jensen?
HE WAS RIGHT!
512SP/750MHz "same thermals as gtx285"
>GTX285*2 performance/november 2009 launch, "this is what you want for christmas"
"yields are perfectly fine" "gf100 taped out in march 2009"

its easy to rip on charlie for being wrong, but was there anybody remotely close to him when it came to giving us a headsup on where things were moving and what to expect? for most of the time it was charlie vs nvidia pr, and charlie was way more accurate than nvidia... so say what you want and nitpick on some details all you want, i believed charlies overall picture and while he painted things a bit darker than it turned out to be, it was still very valuable info and very interesting and entertaining to follow his stories

http://img.techpowerup.org/100326/Capture598.jpg

ill have a veal filet please, medium

Charlie never said 480SP. He first said 512, then said it was "castrated" to 448, then said it would be 512 again. He was never accurate on SP count

no, he said it was castrated to 448 on tesla cards and that there would be close to no 512 cards if at all. there he changed from none at all to few 512core parts... and most likely cause nvidia couldnt make up their mind about it.

and he was never accurate on clocks.

i think he reported what he was told, and es cards WERE clocked that low, and were running way hot... while he was wrong about final clocks, he wasnt wrong about the clocks per se...

His "no tessellator, it's going to be emulated, will suck" claim was 180 degrees wrong. Maybe tess. performance isn't as good as Nvidia claimed but it's definitely no worse than ATI.

who knows how much is really done in specialized tesselation hardware... i think not much, but thats actually smart... the trend is general purpose so having lots of dedicated hardware is stupid... i think his claim of emulating tesselation on the sps is partially right... he was completely wrong about tesselation performance though, yes.

nVidia surprised me:
- 700Mhz, and yet can OC a good 10% more!
- Surely nobody was naive enough to believe +60% on everything. But amazing performance in some games (FC2, BFDX11)tesselation, folding@home and few demos. Fearing only +10%, performance better than expected (depending on which review you read)
- *only* 250W.. better than 280W, but something around 220W would be better still.
- it actually works in all the games. Some 2560 and TRAA driver issues to fix, but overall pretty well polished. Nothing broken (ie Phenom,R600)

1 true
2 no, a lot of people believed nvidias huge %tage gains, and huge fps in fc2... so what? an old game people can now play again at 120fps instead of 80... whats the point? where it matters, ie 2560x1600 the perf is very close to 5870...
3 official 250W tdp, actual max power draw is above 300W...
4 nothing broken? how about 60W idle and 85W video playback? power management def doesnt work as planned if you ask me... same as r600...

the card crashes at 875Mhz almost instantly in '06, at 850Mhz it crashes in the second test but it should do 835/840Mhz core. I haven't pushed the memory yet. I'll do that now.

hey hey! long time no see QD!
nice! thanks for sharing!

Personally I've never heard anything about SLI scaling from anyone before the Fermi launch. And after it, several sites noted scaling better than Crossfire. I think the most interesting one was from HardOCP review - that review put Fermis in the worst light possible but strongly praised their SLI performance.

what? check anandtech, they did some great articles a while back on multi gpu scaling, same as xbitlabs... they compared single vs dual vs tri vs quad in lots of benchmarks and resolutions... anandtech concluded that xfire and sli scale about the same, with sli having a slight advantage. xbitlabs preferred the ati cards cause they were cheaper than comparable nvidia multi gpu solutions at that time.

another thing id reaaaallly like to see is somebody mounting a 480 heatsink on a 5870 and then comparing both cards maxed out on air clock and volts wise

[Jowy Atreides]

4 AMD GPU = 3 NVIDIA GPU

Maybe this is why reviews have such different results.. depends on AA level.

*cough* Framebuffer, not processing power *cough*

[Behemot]

Whenever I take a look on the cooler, it reminds me the "Hot do not touch" sticks which were supposed to end on GTX 285. Where are they on this? Do NVIDIA even realize than if somebody mistakenly touch the metal, he could be terribly burned?!

[jmke]

the ugly:
identical perf to 5870 in vantage

that would go under the "doesn't matter" category, just proof that Vantage is no way to gauge actual gaming performance.

[SoF]

that would go under the "doesn't matter" category

definitly not for me and some benching-folks...I think most of them were expecting some "baaam" numbers for some benchies but there ain't any - and seeing what shamino could do in vantage I am still not convinced that this card can win more than 2 benches if any against the upcoming pimped 5870s.

[jmke]

definitly not for me

ok; it's definitely not in "the ugly" department, that is for sure.
I pretty much knew some OCer was going to reply to my statement, and you are correct that it is sad to see a GPU which is , on paper, much faster than HD 5870, not being in first place in a synthetic benchie.
doesnt make the Fermi any less interesting to OC though ;-)

[pentium777]

if your gaming on a 2560x1600 screen, your either not a real gamer or your an idiot and should downgrade your display instead of spending tons of cash on multi gpu solutions to actually be able to play the latest and greatest games without multi gpu drawbacks such as stutter, huge power consumption, noise and bugs

The only "IDIOT" and "not a real gamer" here is you. Downgrade 30" 2560x1600? You should tell that to people also that have 120"-144" screens for home theater that have 1080P they should downgrade to 50-60" screens so that 1080P looks sharper....

Ever think that maybe people with 30" 2560x1600 can afford multi-gpu solutions, BTW please explain and show me when 7900 GTX SLI, 8800 GTX SLI, GTX 280 SLI or GTX 480 SLI has "stutter" problems? Huge power consumption, I can afford $50 extra a year for power, noise? with home theater sound or headphones people aren't listening to their fans spinning up while gaming.

It is clear you don't have a 30" 2560x1600 display, but calling people idiots or not real gamers really shows a lack of intelligence.

PS: I find it funny that you actually spend the time to create all those quotes and discuss / refute all these points. Do you get out much?

[annihilat0r]

PS: I find it funny that you actually spend the time to create all those quotes and discuss / refute all these points. Do you get out much?

the good ol' LOL YOU ARE NOT SOCIAL GET A LIFE counterstrike at full force!