The GT300/Fermi Thread - Part 2!

[trinibwoy]

3. Repeated mention of "CUDA" and "PhysX" instead of DX11. I guess nVidia will continue promoting proprietary path. So, future TWIMTBP games like Batman Arkham Asylum2 will continue to require nVidia card for all the features. AMD users get shafted?

They made a big deal about Tessellation in some other videos but honestly, DX11 has a much more subtle effect than something like PhysX does which is in your face and easy to explain to the masses. Besides, what company wants to focus on stuff that everyone else can do as well? It's all about highlighting the distinguishing features of your product.

[Olivon]

Official in the sense that even the bottom PR people are commenting on a Q1 '10 release???

I personally thought that the latest they could be was end of '09. While it might have been mentioned, it was nothing that could be traced back to Nvidia. We have video and tech media that have direct quotes from Nvidia PR on a Q1 '10 launch. If they miss that, they are in a world of more hurt than even Charlie could imagine.

Hope will be out for Crysis II ...

I joke but nVIDIA is really really late. If they have no GTX 4X0 cards ( with several tech specs etc ...) to introduce in CeBIT that will begin to stink really hard IMHO ...

[zalbard]

Its all bad though, fermi being late and costing a lot when it does come out just hurts PC gaming, Good gaming PC's need to be under 1,000 USD in order for PC gaming to make it back into the mainstream, and right now you can build a good Direct X 10 PC for around that, but throw in a 5850 or 5870 and you break the budget. Fermi needs to come out now, cost the right amount, and work at least haft as well as the fan-boys expect or PC gaming is going to be in trouble.

Very true. What worries me is the scalability of Fermi architecture.
Being the top GPU is nice, being a folding monster is great as well, but if mainstream gamers aren't going to be able to enjoy some of these features within 3-6 months after the launch then not many gaming devs will feel like rushing with the new features...

[Piotrsama]

Good gaming PC's need to be under 1,000 USD in order for PC gaming to make it back into the mainstream, and right now you can build a good Direct X 10 PC for around that, but throw in a 5850 or 5870 and you break the budget.

That's what the 5770 and 5750 are for.
They are good and cheap DX11 cards.

[eric66]

ever since the 600$ 3dfx voodoo2 sli decade+ ago.. thats where the fastest video cards have been hoovering around

one fermi beats 1ghz 58xx @ lower price
one fermi beats 1ghz 59xx @ lower price
one fermi beats 1ghz 6000 @ this is the mofo around 600$

are you smoking something when posting replies ?

[informal]

No doubt that Fermi (top model) will be a great performer,but don't get your hopes too high. It will beat single Cypress ,but 5970 will be more than enough to take on it. Even a 1Ghz single Cypress will be competitive in many cases.
Fermi design policy was primarily GPGPU computing capability ,after that gaming. It doesn't help that it is huge and power hungry and that it had design issues(Charlie talked about issues NV had early on with Fermi),and it doesn't help that TSMC's 40nm node is not top notch yields-wise.

[dengyong]

No doubt that Fermi (top model) will be a great performer,but don't get your hopes too high. It will beat single Cypress ,but 5970 will be more than enough to take on it. Even a 1Ghz single Cypress will be competitive in many cases.

Is this a fact or just your speculation ?

[informal]

This can be clearly seen from the leaked numbers so far(can be found on the previous pages of this thread). BTW,nothing is "fact" until the product is launched and tested,so speculation fits perfectly.

[ajaidev]

Is this a fact or just your speculation ?

I think its speculated fact based on the performance shown by fermi in farcry 2 bench.

[Your_Boss]

YES the whiteout demo was amazing, here a link for a bit of a refresh for everyone http://www.youtube.com/watch?v=FL2hnBe6tiE

Based on http://www.youtube.com/watch?v=3-1GMbqzGX0, I am guessing its 3 GF100 cards, one for rendering, one for PhysX/CUDA, and one for Direct X11 Compute and Tessellation, or maybe its all 3 cards working together on everything, there is no way to tell I guess, Thus I agree with hurleybird, I am not sure if that demo is really anything to get that excited about, its cool tech, but if it takes 3 GF100 cards to run its really not going to impact most peoples gaming. Then again do demos like this ever really relate to what we will play this and next year?

http://www.nvidia.com/object/gf100.html is a fun read though to see how Nvidia sees GF100 at least.

That Ruby demo looks good still..and has falling ice, kicking up snow, etc, and guess what? NO PISSX, isn't that crazy?

Also, yes I think I read somewhere that the Sled demo was running 3 Fermis SLI'd. So, no, not too impressed.

[NapalmV5]

is that what you're hoping for? or is that what you know?

anything less and they can go pack up and go home.. do i make myself clear ?!?

6 months later: 5870 ~ fermi < 5970

whats the point of/in that^ ?!?

[NapalmV5]

No doubt that Fermi (top model) will be a great performer,but don't get your hopes too high. It will beat single Cypress ,but 5970 will be more than enough to take on it. Even a 1Ghz single Cypress will be competitive in many cases.
Fermi design policy was primarily GPGPU computing capability ,after that gaming. It doesn't help that it is huge and power hungry and that it had design issues(Charlie talked about issues NV had early on with Fermi),and it doesn't help that TSMC's 40nm node is not top notch yields-wise.

dont get your hopes too high on 1ghz 58xx/59xx for they shall soon be six feet deep

[***Deimos***]

Very true. What worries me is the scalability of Fermi architecture.
Being the top GPU is nice, being a folding monster is great as well, but if mainstream gamers aren't going to be able to enjoy some of these features within 3-6 months after the launch then not many gaming devs will feel like rushing with the new features...

True True

A must read:
Inside look at how RV770 got made
http://www.anandtech.com/video/showdoc.aspx?i=3469&p=5
Inside look at how RV870 got made
http://www.anandtech.com/video/showdoc.aspx?i=3740&p=3

Look at the die size comparisons, and read the part about power budgets, and yields (harvesting). In the ultimate quest to be #1, nVidia seems to be making same mistake as last time.

[dengyong]

This can be clearly seen from the leaked numbers so far(can be found on the previous pages of this thread). BTW,nothing is "fact" until the product is launched and tested,so speculation fits perfectly.

Fair enough.

[weston]

No doubt that Fermi (top model) will be a great performer,but don't get your hopes too high. It will beat single Cypress ,but 5970 will be more than enough to take on it. Even a 1Ghz single Cypress will be competitive in many cases.
Fermi design policy was primarily GPGPU computing capability ,after that gaming. It doesn't help that it is huge and power hungry and that it had design issues(Charlie talked about issues NV had early on with Fermi),and it doesn't help that TSMC's 40nm node is not top notch yields-wise.

Agreed, I'd put fermi just above the 5870 in games. And as always, nVidia will do better in some games, ATI will do better in a few others. tesselation will give a big advantage to nVidia but games that don't use it, or only use it a little bit will be very close.

/speculation, but based on evidence

[***Deimos***]

Sacrificing performance to meet the 16mm x 16mm die size targets wasn’t going to happen, but what ultimately convinced Carrell to go with a larger die this time around was something that ATI didn’t get nearly enough praise for: the ability to launch 4 different 40nm DirectX 11 GPUs in less than 6 months.

Remember that Carrell’s beef with building the biggest GPU possible is that it takes too long for the majority of customers to get access to derivatives of that GPU. Look at how long it took G80 or GT200 to scale down. And who knows when we’ll see $150 Fermi/GF100 derivatives.

Engineering eventually figured a way to fix most of the leakage problem through some changes to the RV740 design. The performance was still a problem and the RV740 was mostly lost as a product because of the length of time it took to fix all of this stuff. But it served a much larger role within ATI. It was the pipe cleaner product that paved the way for Cypress and the rest of the Evergreen line.

As for how all of this applies to NVIDIA, it’s impossible to say for sure. But the rumors all seem to support that NVIDIA simply didn’t have the 40nm experience that ATI did. Last December NVIDIA spoke out against TSMC and called for nearly zero via defects.

imagine the sly smiles on ATI engineers faces when they hear nVidia is having same problems that they did - which the ATI guys already fixed months ago.

[Chumbucket843]

Chen's speech included a call to "my friends at TSMC to give me more 40 nm parts," and a plea for improved via defectivity. D(i)ck James, a technology analyst at Chipworks (Ottowa, Canada), said via defects have shown up on ICs manufactured by TSMC. Chipworks has inspected products from graphics vendor ATI, now part of Advanced Micro Devices (AMD, Sunnyvale, Calif.). "The problem appears to be that when they cut a via, a residue of photoresist gets on the edge of the via, which creates a ring-shaped discontinuity in the metal," James said. "The discontinuity could create electromigration issues. We've seen the same problem on the upper metal levels on the ATI chips we've studied. It creates a reliability failure mode."

engineers dont spend their time laughing at the competition, they do their jobs. chipworks is a company that analyzes competitors products.

[eric66]

dont get your hopes too high on 1ghz 58xx/59xx for they shall soon be six feet deep

must be some strong stuff

[LordEC911]

This is a couple weeks old but a good article from Damien.
http://www.behardware.com/articles/7...evolution.html

[NaMcO]

must be some strong stuff

Yeah... will they work? As these don't yet

[FischOderAal]

anything less and they can go pack up and go home.. do i make myself clear ?!?

6 months later: 5870 ~ fermi < 5970

whats the point of/in that^ ?!?

So NVIDIA should just say "sorry dudes and dudettes, but Fermi won't beat HD 5970 so we ain't gonna relase it"? What would be the point of that?

Fermi might not be the new messiah for gamers, but as quite few already pointed out Fermi has unique HPC-capabilities.

A product that might not sell well to gamers but do "quite" good in HPC-applications is better than no product at all, don't you think?

Sometimes you've got to release what you got, no matter what. Don't you think ATI didn't like what R600 turned out to be? RV670 was basically R600 done right but wasn't really faster. Should they not have released RV670 because it was late and not fast enough?

Wake up NapalmV5! WTH happened to you?!

[Otis11]

So... very rough estimate 40% (1 - (500000 / (60000*6)) = .39) faster than GTX285 while needing to render a scene and do physics as well? Sounds pretty good. The whole demo looks sweet, but so did whiteout. In fact, I'd say whiteout still looks better than any game on the market and came out with the HD2900. Just shows that demos aren't always the best way to gauge power.

Yeah, but you also have to consider that these particles may be rendered differently. They talk about all the different forces they are taking into account...

Idk how much that effects it, but just wanted to throw that out there.

[saaya]

I gotcha, it really does seem like they fell asleep at the wheel in regards to getting any dx11 hardware out the door. Both ATI and Nvidia seem to release the latest on the highend part first before releasing the lower parts so with gt300 delay's you would have thought some midrange to lower end parts would have been scheduled to release by now.

i dont think they fell asleep, they didnt think dx11 was that important, or, most likely, didnt think dx11 matters all that much since the gaming industry has slowed down. they probably checked the plans of major game devs and most of them said they will focus on dx10.1 at most, if not even dx9 cause thats what consoles are stuck at for now, so they decided not to focus on dx11 yet...

and i think they are right... dx11 doesnt matter all that much and isnt necessary yet... BUT that doesnt mean they dont need it... its very important for marketing... who will buy a new 10.1 or 10.0 vga 2 years after he already bought a 10.0 vga? it either has to be cheaper or notably faster than the old card... for most people perf isnt all that important since games arent as demanding right now, so dx11 does matter as a selling point.

and i think nvidia underestimated how many game devs ati can convince to actually adopt dx11, even if its only patched on...

ever since the 600$ 3dfx voodoo2 sli decade+ ago.. thats where the fastest video cards have been hoovering around

one fermi beats 1ghz 58xx @ lower price
one fermi beats 1ghz 59xx @ lower price
one fermi beats 1ghz 6000 @ this is the mofo around 600$

itll take you a long time to regain some credibility in the eyes of everybody who read this post :P

We have video and tech media that have direct quotes from Nvidia PR on a Q1 '10 launch. If they miss that, they are in a world of more hurt than even Charlie could imagine.

im not sure... i think they have until Q2 to come out with fermi... if they still havent launched fermi properly (shipping in volume) in the middle of this year, then LOTS of investors will jump ship... cause thats what nvidia told all the investors really matters, and most fell for it... they all believe in the fairy tales of gpgpu and the huge margins and profits nvidia supposedly will cash in on with fermi. and nvidia promised fermi tesla cards in mid 2010, they never changed that date, they said mid 2010 even at the end of 2009 already, and they stuck to it... but if they miss THAT date, then their stock price will take a major hit...

dont get your hopes too high on 1ghz 58xx/59xx for they shall soon be six feet deep

riiiight
just like G92 has been 6feet deep since gt200 launched...
oh wait, gt200 was shrunk to 55nm, its price was cut in half, it was shrunk to 40nm, had dx10.1 glued on, cut in price again and G92 is still a kick4ss gaming card and unrivaled in price perf... :P

engineers dont spend their time laughing at the competition, they do their jobs. chipworks is a company that analyzes competitors products.

if they have a sense of humor, they sure do

FYI, i asked around regarding fermis @ cebit and didnt get a single reply...
Looks like there wont be any fermis at cebit, at least not in public...
they will probably do another deep dive like event or something like that... i just wonder if it will be a 100% repeat of the event in vegas or if there is anything new... at all...

[OutSider]

they better show something new...

[saaya]

http://www.semiaccurate.com/2010/02/...ding-examined/

looks like the financial markets know something we dont?
they seem to have lost most of their faith in nvidia and expect "ok" to bad results for 2010...

nvidia really cut their rnd budget? didnt they just brag about how much they spent on rnd over the past years and that they are investing more and more?

if they cut rnd it sounds like they are definately not working on a new architecture but are refining fermi... they are probably preparing a 32nm or 28nm fermi?