The GT300/Fermi Thread

[Zed_X]

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!). But info about performance at all and in some games is OUT. In press pack is more info then on meeting place. Real gaming perf is great, but Nvidia has two months for even better results.

[neliz]

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!). But info about performance at all and in some games is OUT. In press pack is more info then on meeting place. Real gaming perf is great, but Nvidia has two months for even better results.

Dude.. that info has been spreading since October when nvidia started updating it's partners and very select press, that wasn't exclusive to the deep-dive event at all. Or do you think my 30% faster than HD5870 has been pulled from random sphincters?

[Zed_X]

Dude.. that info has been spreading since October when nvidia started updating it's partners and very select press, that wasn't exclusive to the deep-dive event at all. Or do you think my 30% faster than HD5870 has been pulled from random sphincters?

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 dissapointed in Fermis performance.

[neliz]

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 dissapointed in Fermis performance.

You're obtuse on Purpose? I'm not talking about the GTC conference. I'm talking about nvidia's top-to-bottom product presentation on Fermi based Geforce products.

If you knew anything from the slides, just give use ONE codename mentioned there.

[zerazax]

The R600 was just a crappy design by a graphics company hurting financially and only saved by AMD buying them up.

What's funny is that R600 was the basis for RV770 and RV870, and so far, they've done quite fine for ATI.

Maybe R600 wasn't the success ATI wanted, but the architecture has more than vindicated itself, seeing as how it's lasted 3 generations and has produced incredible scaling

And again, if the only performance on Fermi is what Nvidia PR is saying, I'd scale that back 10%

(it's my general rule regarding performance from PR sources... scale it back a bit and you see the truth )

[AuDioFreaK39]

Dual-GPU Fermi card to launch after single



First to launch are two single-GPU cards

Fermi GF100 is a single chip GT300-based 40nm DirectX 11 card. Nvidia plans to launch two cards that will aim to replace Geforce GTX 285 and Geforce GTX 260 in enthusiast and performance markets respectively, while the high-end dual-chip Fermi card will come after.

We don’t know how much later the dual-GPU should launch, but it should happen within one to two months. If all goes well and it doesn’t get delayed, we can hope to see it by May. Around that time, ATI might be ready with its refreshed 40nm cards that might have more juice and could compete better. However, this is only after Nvidia finally launches its GT300, Fermi-based products.

It is definitely not fun to be in Nvidia's shoes, especially when it comes to the desktop graphics card market.

Source

[neliz]

I hate it when people post in bold letters what I've been telling last week

but wait!

http://www.fudzilla.com/content/view/17198/1/

Fermi-based cards to end up hotter
Than previous Nvidia cards

We got some thin information on Nvidia Fermi and our sources have strong reasons to believe that Fermi will probably end up hotter than the previous single chip generation, including Geforce GTX 285.

Fermi is a big chip, so this doesn't come as much surprise. As long as it runs stable, people won’t complain about it too much. We’ve seen the card in person at least twice so far and we can tell that the first one we saw has a dual-slot air cooler and runs quite hot. However, this might not be the final revision that should ship later this quarter.

You can only imagine how hot the dual-GPU card will get, and yet Nvidia still plans to make it.

[AuDioFreaK39]

I hate it when people post in bold letters what I've been telling last week

lol...that's how I always post news, regardless of what it is.

[neliz]

lol...that's how I always post news, regardless of what it is.

Can you show us the EVGA OC/stress/benchmark tool news as well? :p

[AuDioFreaK39]

Can you show us the EVGA OC/stress/benchmark tool news as well? :p

Writing it at the moment actually. I'll post it in an hour or so.

[Aerwidh]

This just in!! Breaking News!!!

Attendees at PDXLAN will be amoung the FIRST LAN attendees to see GF100 in action!!! That's right, the NEXT, NOT EVEN RELEASED Video Card from NVIDIA!!!! Taking it a step further, CLUB SLI members can win one of the first ones AUTOGRAPHED by the CEO of NVIDIA!! ZOMG!!!!! To win one, it's all part of a contest taking place ONLY at PDXLAN 15!!!

All contest details at www.pdxlan.com - Last day to get tickets to PDXLAN is Tuesday night!! Register at pdxlan.com!!

Sounds like a really nice event!

[saaya]

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 dissapointed in Fermis performance.

how do you know it was new for everybody and not just you? you cant possibly know if some information is new or just new to you... and im sure nvidia has had, and most likely shared, perf estimations for fermi for quite some time...

What's funny is that R600 was the basis for RV770 and RV870, and so far, they've done quite fine for ATI.

Maybe R600 wasn't the success ATI wanted, but the architecture has more than vindicated itself, seeing as how it's lasted 3 generations and has produced incredible scaling

And again, if the only performance on Fermi is what Nvidia PR is saying, I'd scale that back 10%

(it's my general rule regarding performance from PR sources... scale it back a bit and you see the truth )

well idk... r600 was most likely revamped a lot... i wouldnt say that r600 in itself was such a great and scalable design... a truly great and scalable design doesnt first suck and needs to be modified greatly to perform well... thats against the whole definition of a very scalable design

[mao5]

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!). But info about performance at all and in some games is OUT. In press pack is more info then on meeting place. Real gaming perf is great, but Nvidia has two months for even better results.

are you PHK in NVnew forum ? so I'' take all your info with a big Spoon of salt

[mao5]

You're obtuse on Purpose? I'm not talking about the GTC conference. I'm talking about nvidia's top-to-bottom product presentation on Fermi based Geforce products.

If you knew anything from the slides, just give use ONE codename mentioned there.

he won't give us anything, he's going here just for fun.

[neliz]

well idk... r600 was most likely revamped a lot... i wouldnt say that r600 in itself was such a great and scalable design... a truly great and scalable design doesnt first suck and needs to be modified greatly to perform well... thats against the whole definition of a very scalable design

R600 was good shaders packed in a horrible design. R600 had the same design mistakes as NV30, thinking you can solve everything with Shader power.

[zalbard]

Fermi-based cards to end up hotter

Not that it is surprising, but still is a pity considering how cool and quiet 5xxx cards run.

[DeathReborn]

R600 was good shaders packed in a horrible design. R600 had the same design mistakes as NV30, thinking you can solve everything with Shader power.

Not forgetting being on that awful 80nm process which caused most if not all of the power problems. Too little texturing power didn't help either.

[Cybercat]

Whatr you talking about Saaya has agreed with me plenty times. We both agree that nvidias business model is doomed. You cant COMPETE in the gpu market, if you aint making gpu's period. Fermi is a trickle down product that has no business being in a desktop. Fermi is designed only for being a gpgpu cruncher. The only way nvidia can truly compete is if they split their designs and make dedicated ppu's and dedicated gpu's. You cant have one gpu that works both ways efficiently!

Yes, because we've seen how well dedicated PPUs sell. I'm not sure how you've come to the conclusion that Fermi is only meant for general purpose stuff. The DirectX model itself is moving towards more programmability, and more general purpose, and at the moment, the HD 5870 has more GPGPU-centric hardware than any other GPU out right now. Would you say it's unable to run graphics efficiently because of that?

If Fermi is fast in games, it won't matter what sort of GPGPU support is included in it.

[NickF]

The latest FUD:
http://www.fudzilla.com/content/view/17205/1/

Those in the known are claiming that Fermi despite the yield and thermal issues is only about 20% faster than Cypress, while Hemlock smokes it. The combination of low yields, high thermals, and marginally better performance than Cypress could be conspiring to place Nvida in the position of having to release the card, but have to sell it at a loss till they are able to address the issues in the next spin according to sources. Because of the situation a mole we know is suggesting that Nvidia may limit the sales of Fermi to consumers and instead use the chips for the Tesla and Quadro products where prices and margins are much better.

[Chumbucket843]

well idk... r600 was most likely revamped a lot... i wouldnt say that r600 in itself was such a great and scalable design... a truly great and scalable design doesnt first suck and needs to be modified greatly to perform well... thats against the whole definition of a very scalable design

yes, r600 had some bad choices. i have to disagree with how scalable the designs are. g80 was is simple as a gpu can get would require more work to make it dx11 compatible than r600 which was way more complex but its features allow it to last for multiple generations(just look at cypress). not that much has changed, they got rid of the ring bus, added LDS, gddr5, HW thread synchronization, double precision and some other small stuff. perhaps thinking too far ahead is why the arch performed poorly.

Not forgetting being on that awful 80nm process which caused most if not all of the power problems. Too little texturing power didn't help either.

80HS is not supposed to be efficient, it is designed for high clockspeeds which is why ln2 overclocks were very high. its good for its intended purpose.

[xlink]

shouldn't nVidia be banned from this tech site for trolling?

[zerazax]

The reason I said R600 was a pretty scalable design is based on how it's progressed

Yeah they got rid of the ringbus and had to change things over time, but it's core design has evolved very very quickly

80nm 320SPs -> 55nm 320SPs -> 55nm 800SPs -> 40nm 1600SPs

All of this progressed VERY quickly while keeping die size down after the 80nm debacle

Yeah things changed along the way... more texture units, AA moved from shaders back to RBE's etc., but the architecture scaled very quickly

In hindsight, G80 started off more powerful than R600, but Nvidia has had a harder time scaling it up

[highoctane]

Rumor...

[Rumour] Nvidia Fermi production in 3rd week February

http://vr-zone.com/articles/-rumour-...uary/8273.html

[DarthBeavis]

shouldn't nVidia be banned from this tech site for trolling?

guess you would have to ban me as part of that effort then. any site that bans such a major vendor is truly stupid. Even Kyle at [H], with his extreme harshness, was wise enough to elicit feedback from Nvidia instead of shutting them down in terms of communication

[Teemax]

The latest FUD:
http://www.fudzilla.com/content/view/17205/1/

Wow, a devastating article on Fudzilla about Fermi! I thought I would never see the day... (he was really insisting on a Dec 2009 paper launch for a while)

If the article has any truth in it, then NV30 might be repeating itself with Fermi. I hope that Fud gets it wrong this time as well!