The Fermi Thread - Part 3

[GoldenTiger]

Well I have a 4870 and no issues, and golden saw the need to post the same list twice (thrice?). I can make a list of issues too, but I'm not that an*l.

I'm not being "an*l" I only posted it once here, at the end of the last thread in response to someone asking for it when I mentioned I had 5870 issues. The thread was locked so I posted it back in here so we could continue our discussion. The original quote was made at a different forum by me that I copied/pasted. These aren't minor issues that I am being "an*l" about and nitpicking: these are major visual glitches/errors that are persistent and constant, as well as other stability issues not listed. There's many many more that would be "nitpicking" to most people, believe me... I don't consider flickering bushes, being unable to force AA/AF in some games, etc. as "nitpicking" and I sincerely doubt many would.

In short, in my personal experiences, the 5870 drivers are mostly garbage: they get the card to run, but it looks like cr*p while doing so. As I said, I'm very much looking forward to being back on an nV card without all the rendering errors. In short, just because you disagree with my opinion and haven't personally had the issues doesn't mean many don't: no need to try to minimize me and claim "oh he posted it a buncha times, must be trolling!" when no such thing occurred .

Anyway... 26th is NDA drop, anyone know if any cards will be available then or is it the 6th as rumored for being on newegg/et. al ?

[Sn0wm@n]

no.

like i said i want a mature explanation... not only a no ...... :P

If Fermi doesn't actually have a dedicated tesselation unit, then yes in that situation the ATI card would win. But we're not sure either way, we'll see when the NDA is over.

Again if this is the case, ATI will push for more games with a more balanced load while nvidia will push for higher geometry.

thanks you

[Picao84]

like i said i want a mature explanation... not only a no ...... :P





thanks you

Ok Its enough time to end this nonsence. BOTH nVIDIA and ATI use SP to do Tesselation! Both!

Why?

Because Tesselation is a three stage process, where the middle stage can be done in fixed function, but the first and last stages are the Domain and Hull SHADERS. You surely know shader work is done on CUDA/Stream/Shader/put_fancy_name_here cores.

So, NO! The tesselator in GF100 is not a pure SP one, same as ATI is not pure fixed function! Those are myths created by Charlie and ATI FUD machine.

[Manicdan]

I wouldn't call them being lazy. If there is no comp right now, and your reference cards are selling well, and your getting positive reviews from sites and customers whats the point in dropping a refresh? Plus I think they are waiting to see how well Fermi does , as a lot of other people are theorizing. Around the time or after the release of Fermi then u'll probably start to see refreshes. Is that a word? Refreshes? lol

i like to think of the $ per mm2. yes the 4800s were cheaper when they were released, but they were also alot smaller than the 5800s

5870:
400$/334mm2 = 1.2$ per mm2

4870:
300$/256mm2 = 1.17$ per mm2

they are practically identical comparing their release prices. both were not the first they did on the node (55 and 40nm) and both are the biggest they did on the node. so theres alot of similarities between when trying to think of the simplest cost and revenue potential. based on this its pretty safe to say they were priced competitively, and there is room to drop prices now that this card is quite a few months old. if fermi came out at the very end of last year, then we would have seen prices drop then, slowly over time like normal markets. but since its a few months later, AMD could either drop it a massive ~30% to make up for the lost time, or kinda let time skip, and start a slow price drop from here on and enjoy their last few months ahead of the curve.

[takamishanoku]

What gets me is that people are saying fermi does not have a dedicated Tess unit. Can anyone explain why by specs it has approx 3 times as much triangles than the 5870? Is this assuming shaders are used and how would 3 times more triangles not be better with dx 11 tesselation?

[Teemax]

Alright, Zed_X came here exclusively to "leak" about Fermi, so I did a bit of a trace:

5-jan-2010
NVIDIA at CES: GeForce 3D VisionFinity and Dark Void PhysX
At CES will NVIDIA to demonstrate new "(nV)Eyefinity" technology (true)
Three displays with GeForce 3D Vision and Dark Void PhysX game - on ONE Fermi card! GeForce 3D Vision and Eyefinity together (false)

All of you will see what is Fermi in TWO days,...
All of you will see what is Fermi in TWO days, and someone will be hardly surprised Performance is BRUTAL! (false)

7-jan-2010
No one reviewer will get GeForce 300 this week, NVIDIA will send samples later in february. But they will have many infos about GeForce architecture, features, performance - but no CLOCKS! In few days NDA lifts (someone will leak it, for sure) and every server will bring article about it, be patient. (true enough)

9-jan-2010
GeForce 380 has huge power consumption (true), but performance is only few percent under HD 5970 and far away from HD 5870 perf (false). Trust me, will see next week on many servers, when NDA for info lifts!

NDA will lifts few days after 14th, today is main GeForce 300 Deep Dive meeting. I can say this: GeForce 300 has higher perf then GTX 295 with lowest power consuption. Look at GTX 295 vs HD5870/5970...(at this point, I say false)

11-jan
Every invited journalists has complete Fermi Geforce slides now, NDA will lift in few days, but i hope someone broke it and publish something on B3D or somewhere in Asia like always. If you want to know +/- perf. look at my previous post here in this thread. Dont worry about TDP, on CES are only first silicon unfinished products, remember first black press samples 4870X2 with TDP 400W? (true about NDA, likely false regarding TDP)

12-jan
Deep Dive meeting has over, and we already know how powerful GF100 is, and IT IS! Performance is better then expected, most peoples here will search in own words and feel to shy for craps in thread....(false)

that TGD news is bull (it seems they were not invited), i told you before a WEEK, there will be NO CLOCK, TDP info fo journalists (am not a journalist!). (I'm saying false, everybody is saying 270-280w for GTX 480 today)

You are wrong, Deep Dive was not "Fermi architecture white-papering", like in October was. Every info are brand new, every slide talking about perf in real numbers. I am sure, no one will be (false, Farcry 2 was the only performance number we had)

13-jan
his screen i found on TPU forum, represent GF100 power truly REAL!
http://www.czechgamer.com/pics/clank...6-19_fermi.jpg
Between 20-23 percent above HD5870 and only little under HD 5970(false)

24-jan
thats all is BS, For Surround are needed TWO VGAs in SLI, not only ONE GTX 285, this graps says "estimated" performance, not "measured". Absolutely nonsens
PS. For 3D surround you need TWO GF100 too, not only ONE (true, correcting himself on the 5-jan claim)

Hehe, you are wrong like always. You meant poor launch availability of Radeons HD5870 and no buyable HD5970 till today, dont ya? For launch, will be enough card for enthusiast! (remain to be seen, though I can't believe that GF100 will satisfy enthusiasts' demand)

15-feb
NV finally found final CLOCKS of GTX 470/480. Cards are ready to launch, will be in second half of March in reviews. There is one word: Gemini, when GF100 will be launched, remember for that word!(true)

22-feb
Tomorrow will be exciting date, for fans! (false, it was anything but "exciting")

launch is still in Q1/2010 (true)

23-feb
GTX 480 has 512SP, i thought it is a taken as a facts, where are you was last few weeks/months triniboy? (false)

7-mar
yeah FAKE, frequencies are higher ... more higher
PS. GF100 is fully supported in 196.75 (ID removed from inf), and final release of driver is 196.78-196.80 NOT 197.xx
PS2 - i am guarantee, last GPUz did not recognise GF100 cards
(we'll see about this)

12-mar
Yeah, GTX 480 is not competitor for 5970, but 5870 This is not NEWS. (again, contradicting his earlier post)
Against 5970 NV has planned dual GTX 470 (we'll see)



To me, the guy seems to have some real insider's info. But his view point is rather ... inflated. He hyped GF100 as least as much as giving out some info about it.

[hennyo]

Ok Its enough time to end this nonsence. BOTH nVIDIA and ATI use SP to do Tesselation! Both!

Why?

Because Tesselation is a three stage process, where the middle stage can be done in fixed function, but the first and last stages are the Domain and Hull SHADERS. You surely know shader work is done on CUDA/Stream/Shader/put_fancy_name_here cores.

So, NO! The tesselator in GF100 is not a pure SP one, same as ATI is not pure fixed function! Those are myths created by Charlie and ATI FUD machine.

Well, good job at pointing out some of the mechanics of how tesselation works, but you did fail to mention the singular reason why the 4xx Nvidia hardware will ALWAYS have a huge advantage over the ATI 5xxx series in tesselation. You did part of the work for me, mentioning that this was a 3 stage processes but left out the most important detail, in ATI 5xxx hardware, this is a multipass operation which requires it to go out to board memory and back the core at least once, and more depending on how much tesselation is used, this slows down the process considerably. Now on Nvidia 4xx hardware, it is also a multipass operation, but due to a unified cache architecture it doesn't require a trip out to main memory for each pass, only a trip to on chip cache making very significant gains over the ATI 5xxx series on tesselation, this also caries over to big gains on any multi pass shader such as some of them used in Crysis.

EDIT: The multi pass shader operation I thinking of in Crysis is called "parallax occlusion mapping" in case anyone was wondering.

[ajaidev]

I said fermi has both hardware based fixed tessilation unit as well as the ability to use SP's for tessellation. The scores you see on Unique are a result of that, one does not need huge tessellation processing in games "at least for now" so in this case Fermi will use its hardware based tessellation unit. But once the tessellation load gets heavy enough they may use SP's to do tessellation.

The end thing i want to point out is Fermi's BIG thing is tessellation power it carries and the flexibility it offers. Fermi owners maybe in luck if next gen games use tessellation more and require less raw SP's to render. If one looks at Unique one can notice that it slows down quite a lot on 5870's when a scene with heavy tessellation is rendered "dragon" but in such a scene there is lesser use of its SP's because the tessellation unit is loaded to the max.

[Zed_X]

Alright, Zed_X came here exclusively to "leak" about Fermi, so I did a bit of a trace:

5-jan-2010
NVIDIA at CES: GeForce 3D VisionFinity and Dark Void PhysX
At CES will NVIDIA to demonstrate new "(nV)Eyefinity" technology (true)
Three displays with GeForce 3D Vision and Dark Void PhysX game - on ONE Fermi card! GeForce 3D Vision and Eyefinity together (false)

ok, false you need two cards ... but only for driver limitation ...

All of you will see what is Fermi in TWO days,...
All of you will see what is Fermi in TWO days, and someone will be hardly surprised Performance is BRUTAL! (false)

this is true ..

7-jan-2010
No one reviewer will get GeForce 300 this week, NVIDIA will send samples later in february. But they will have many infos about GeForce architecture, features, performance - but no CLOCKS! In few days NDA lifts (someone will leak it, for sure) and every server will bring article about it, be patient. (true enough)

ok, true

9-jan-2010
GeForce 380 has huge power consumption (true), but performance is only few percent under HD 5970 and far away from HD 5870 perf (false). Trust me, will see next week on many servers, when NDA for info lifts!

this is true, 480 has im many games better perf then 5970!

NDA will lifts few days after 14th, today is main GeForce 300 Deep Dive meeting. I can say this: GeForce 300 has higher perf then GTX 295 with lowest power consuption. Look at GTX 295 vs HD5870/5970...(at this point, I say false)

ok, this is true!

11-jan
Every invited journalists has complete Fermi Geforce slides now, NDA will lift in few days, but i hope someone broke it and publish something on B3D or somewhere in Asia like always. If you want to know +/- perf. look at my previous post here in this thread. Dont worry about TDP, on CES are only first silicon unfinished products, remember first black press samples 4870X2 with TDP 400W? (true about NDA, likely false regarding TDP)

this is true

12-jan
Deep Dive meeting has over, and we already know how powerful GF100 is, and IT IS! Performance is better then expected, most peoples here will search in own words and feel to shy for craps in thread....(false)

this is true

that TGD news is bull (it seems they were not invited), i told you before a WEEK, there will be NO CLOCK, TDP info fo journalists (am not a journalist!). (I'm saying false, everybody is saying 270-280w for GTX 480 today)

that was a true!

You are wrong, Deep Dive was not "Fermi architecture white-papering", like in October was. Every info are brand new, every slide talking about perf in real numbers. I am sure, no one will be (false, Farcry 2 was the only performance number we had)

where is problem here?

13-jan
his screen i found on TPU forum, represent GF100 power truly REAL!
http://www.czechgamer.com/pics/clank...6-19_fermi.jpg
Between 20-23 percent above HD5870 and only little under HD 5970(false)

this is real! OMG kid, where youv been last month?

24-jan
thats all is BS, For Surround are needed TWO VGAs in SLI, not only ONE GTX 285, this graps says "estimated" performance, not "measured". Absolutely nonsens
PS. For 3D surround you need TWO GF100 too, not only ONE (true, correcting himself on the 5-jan claim)

true

Hehe, you are wrong like always. You meant poor launch availability of Radeons HD5870 and no buyable HD5970 till today, dont ya? For launch, will be enough card for enthusiast! (remain to be seen, though I can't believe that GF100 will satisfy enthusiasts' demand)

15-feb
NV finally found final CLOCKS of GTX 470/480. Cards are ready to launch, will be in second half of March in reviews. There is one word: Gemini, when GF100 will be launched, remember for that word!(true)

true

22-feb
Tomorrow will be exciting date, for fans! (false, it was anything but "exciting")

true, it was!

launch is still in Q1/2010 (true)¨

yeah

23-feb
GTX 480 has 512SP, i thought it is a taken as a facts, where are you was last few weeks/months triniboy? (false)

hehe, because in that time has ultra part 512SP

7-mar
yeah FAKE, frequencies are higher ... more higher
PS. GF100 is fully supported in 196.75 (ID removed from inf), and final release of driver is 196.78-196.80 NOT 197.xx
PS2 - i am guarantee, last GPUz did not recognise GF100 cards
(we'll see about this)

hehee

12-mar
Yeah, GTX 480 is not competitor for 5970, but 5870 This is not NEWS. (again, contradicting his earlier post)
Against 5970 NV has planned dual GTX 470 (we'll see)

hehhe

To me, the guy seems to have some real insider's info. But his view point is rather ... inflated. He hyped GF100 as least as much as giving out some info about it.

yeah, because this guy is insider

[Dimitriman]

I wouldn't call them being lazy. If there is no comp right now, and your reference cards are selling well, and your getting positive reviews from sites and customers whats the point in dropping a refresh? Plus I think they are waiting to see how well Fermi does , as a lot of other people are theorizing. Around the time or after the release of Fermi then u'll probably start to see refreshes. Is that a word? Refreshes? lol

Because releasing a refresh is capitalizing on "free" R&D and process maturity. As 40 nm scales better and operates with less heat and higher frequencies, AMD can release a "new product" without extra R&D costs and low yield problems.

This applies to both AMD and Nvidia, the refresh is a very very cheap way to release a new product on the market and boost revenue from higher prices (while lower production costs)

[Marios]

Zed_X
Do you know anything about GeForce GF100 GPU versions ready to ship or about to ship in the near future?
Is GF100-375-A3 a 480SP version?
Is GF100-275-A3 a 448SP version?
Is there a high end 512SP GPU version?

[thatdude90210]

To me, the guy seems to have some real insider's info. But his view point is rather ... inflated. He hyped GF100 as least as much as giving out some info about it.

"Fanboy with a source."

[Picao84]

Well, good job at pointing out some of the mechanics of how tesselation works, but you did fail to mention the singular reason why the 4xx Nvidia hardware will ALWAYS have a huge advantage over the ATI 5xxx series in tesselation. You did part of the work for me, mentioning that this was a 3 stage processes but left out the most important detail, in ATI 5xxx hardware, this is a multipass operation which requires it to go out to board memory and back the core at least once, and more depending on how much tesselation is used, this slows down the process considerably. Now on Nvidia 4xx hardware, it is also a multipass operation, but due to a unified cache architecture it doesn't require a trip out to main memory for each pass, only a trip to on chip cache making very significant gains over the ATI 5xxx series on tesselation, this also caries over to big gains on any multi pass shader such as some of them used in Crysis.

EDIT: The multi pass shader operation I thinking of in Crysis is called "parallax occlusion mapping" in case anyone was wondering.

Thanks for finishing the idea
But, I was not talking about performance. I was only talking about the silly fixed/SP tesselator discussion

[Marios]

Could Windows Vista/7 coarse context switching affect the multi GPU tessellation performance?

[Sn0wm@n]

"Fanboy with a source."

direct link to jjh ??? :O

[Picao84]

According to our dear Charlie, nVIDIA has tapped out GF108:

GF108 is what Nvidia taped out, so its triumphant vaporware GF100 is indeed going to be followed up by a huge wave of GPUs, all of them low end.

Even with somewhat good news, he cant resist to make it look bad:

At this rate, Nvidia will have a full line of parts, in quantity, but not until Q4. Q4 2011, that is.

http://www.semiaccurate.com/2010/03/...pes-out-gf108/

[hennyo]

Thanks for finishing the idea
But, I was not talking about performance. I was only talking about the silly fixed/SP tesselator discussion

Ok, good point, the whole fixed function tesselator discussion is a bunch of BS spouted by people without the first clue how the process actually works.

[MTP04]

I was at my local supplier this past weekend and they told me the GTX 480 will cost around $700(canadian). Not sure how accurate that is but just thought I'd put it out there.

[Marios]

By Q4 2011, Nvidia will have the first 28nm GeForce Fermi based cards at stores.

[flippin_waffles]

According to our dear Charlie, nVIDIA has tapped out GF108:



Even with somewhat good news, he cant resist to make it look bad:



http://www.semiaccurate.com/2010/03/...pes-out-gf108/

You need to make up your mind, does Charlie post news or not? It seems when it's news you like, you're fine calling it news. Selective reading 101.

[Sn0wm@n]

By Q4 2011, Nvidia will have the first 28nm GeForce Fermi based cards at stores.

he had to joke about it because everyone is tired of waiting for fermi .....

but Q4 2010 is horrible .. its 1 year after the official announcement right ?? :|

[Marios]

I'm afraid Mr. CD proves right even when he's joking.
So just in case I had to write it.

[Sn@ke:~]

Last time I checked a std 32bit os had 4gb total shared address space for all AIC's including video cards.

This is not pointed to you personally, but to the whole buggus in question.

"Windows 32bit" will only see and utilize no more than 4 gig (just a little under), but all the other programs installed will use eveything available. So when playing a game, it is not windows 32 managing the memory, but the game itself and the hardware in question (which is sort of a program on its own).

So basically you guys are half right and confuse.

Everything will be in use, but windows 32 will only use what it can, 2 different thing.

[Macadamia]

he had to joke about it because everyone is tired of waiting for fermi .....

but Q4 2010 is horrible .. its 1 year after the official announcement right ?? :|

Very scalable architecture indeed!

I remember when the whitepaper came out all the fantards were all over that word. Irony.

[Sn0wm@n]

I'm afraid Mr. CD proves right even when he's joking.
So just in case I had to write it.

its ok .. im sure hes right ..... hes been right on alot of things ... and thats what is counting in the end .....

Very scalable architecture indeed!

I remember when the whitepaper came out all the fantards were all over that word. Irony.

highly scaleable arch + top of the line fab process = epic gpu

cant wait for nvidia's die shrink on tsmc's next node

/sarcasm