The official GT300/Fermi Thread

[570091D]

the whitepapers said there will be future versions with less double precision for gaming. that probably wont happen this gen though. no one is expecting 3x performance in games. 2x faster could be possible.

[rumor]i am hearing 2.4x increase over gtx280 performance[/rumor]. and while nvidia have added a lot of fp performance, why would that hurt gaming performance? it might take up extra room on the die and consume a bit more power, but i don't understand how it would hurt gaming performance. it seems to me that there are A LOT of people on here that want the chip to fail hard, why? why does a percieved lack of competition in the market give you joy? do you wish to pay more for gfx cards? it also seems to me that some here are forgetting that yesterday's show and tell was all about tesla. everything nvidia is talking about in terms of fermi now is related to telsa, they have said that they will not talk about gaming performance because they don't want to tip their hand.

[003]

[rumor]i am hearing 2.4x increase over gtx280 performance[/rumor]. and while nvidia have added a lot of fp performance, why would that hurt gaming performance? it might take up extra room on the die and consume a bit more power, but i don't understand how it would hurt gaming performance.

Exactly. People are being very thick with regards to the GT300. For example, consider this excerpt from Ars Technica:

But Fermi marks the point at which NVIDIA has officially begin making its discrete GPU tradeoffs favor the HPC market at the expense of gamers. ... and quite possibly leaving the single-chip gaming GPU crown in the hands of AMD's more specialized Evergreen this time around.

Seriously? Who is writing this garbage? GT300 will be at least as fast as the GT200 architecture with the SPs increased by 2.13x in games. However, in reality, it will be a bit faster due to the efficiency of the shaders being increased.

While it is true a number of the HPC tailored features won't necessarily benefit game performance very much, they also will not hurt performance in any way.

Regardless, when the GTX380 is finally released, all this garbage information will be laid to rest.

[Solus Corvus]

Seriously? Who is writing this garbage? GT300 will be at least as fast as the GT200 architecture with the SPs increased by 2.13x in games. However, in reality, it will be a bit faster due to the efficiency of the shaders being increased.

Scaling isn't exactly linear. Some games will see around a 2.13x speedup over GT200, some will see less. In cases where the new arch removes bottlenecks, there may be a few games with more then 2.13x increase. But of course we will need to see benchmarks to know how it works out in reality.

While it is true a number of the HPC tailored features won't necessarily benefit game performance very much, they also will not hurt performance in any way.

They might not hurt performance, but they do cost die space and increase power consumption. Only time will tell if these end up being useful features for most customers, or just wasted space/electricity. It really depends on how the GPGPU market evolves.

[003]

Scaling isn't exactly linear. Some games will see around a 2.13x speedup over GT200, some will see less. In cases where the new arch removes bottlenecks, there may be a few games with more then 2.13x increase. But of course we will need to see benchmarks to know how it works out in reality.

True. I'm referring to the people who run around like a chicken with its head cut off screaming that the GT300 is going to suck for games and it will be beaten by RV870. Honestly, in a WORST case scenario, it will be roughly twice as fast as the GTX285, which will trump a 5870 easily.

[Solus Corvus]

True. I'm referring to the people who run around like a chicken with its head cut off screaming that the GT300 is going to suck for games and it will be beaten by RV870. Honestly, in a WORST case scenario, it will be roughly twice as fast as the GTX285, which will trump a 5870 easily.

The worst scenario could be way worse then that. Roughly twice as fast as a GTX285 is probably the REALISTIC scenario.

IMO, it sounds like this will be a repeat of the last round. NV will have the fastest single chip and ATI will have a somewhat slower, but price/performance competitive offering. I don't know about the dual cards. Obviously ATI will have an x2. I imagine NV will want to release a dual GPU card to counter. But how do you cool 6B full speed transistors in 2 slots of space? They'll have to cut down and/or reduce the speed of the chips or wait for a shrink. The fastest single card halo could go either way, imo.

[003]

Roughly twice as fast as a GTX285 is probably the REALISTIC scenario.

I'm not convinced. We'll all find out when it's released though.

I imagine NV will want to release a dual GPU card to counter.

Nvidia has already indicated that there will be a dual GPU version of the GT300.

But how do you cool 6B full speed transistors in 2 slots of space?

Number of transistors is not what determines heat output, that would be the TDP, which will be similar to the GTX285, so that really won't be much of an issue.

The fastest single card halo could go either way, imo.

Nvidia won't let that happen, and based on performance of the 5870, and knowing the specs of GT300, I believe it is pretty clear the 380 will be faster, which nvidia has already confirmed.

[LordEC911]

Nvidia has already indicated that there will be a dual GPU version of the GT300.

Eventually, yes.
Full spec'ed and full speed? No.
6months after release? Maybe, most likely longer.
32/28nm shrink needed? Possibly.

[RealTelstar]

The worst scenario could be way worse then that. Roughly twice as fast as a GTX285 is probably the REALISTIC scenario.

I expect the performance to be MORE than two 285 in high-res 8xAA.

[Xoulz]

True. I'm referring to the people who run around like a chicken with its head cut off screaming that the GT300 is going to suck for games and it will be beaten by RV870. Honestly, in a WORST case scenario, it will be roughly twice as fast as the GTX285, which will trump a 5870 easily.

That^^ is still to be determined! Most of whats new architecturally, is indeed for the CUDA end of Nvidia's expansion into the Scientific Community. So, given Anand's comments, the GT300 could certainly debut @ only 2X the 285's performance!


Meaning it on par with a HD5870 1GB ~ then, think price?

[DilTech]

That^^ is still to be determined! Most of whats new architecturally, is indeed for the CUDA end of Nvidia's expansion into the Scientific Community. So, given Anand's comments, the GT300 could certainly debut @ only 2X the 285's performance!


Meaning it on par with a HD5870 1GB ~ then, think price?

2x GTX 285 performance(not sli, but DOUBLE) would be better than "par" with the 5870...

Of course, no one knows the full performance on these cards, as been stated, but I heavily doubt it'll be less than double the GTX285(and again, NOT sli, but total double).

[zanzabar]

2x GTX 285 performance(not sli, but DOUBLE) would be better than "par" with the 5870...

Of course, no one knows the full performance on these cards, as been stated, but I heavily doubt it'll be less than double the GTX285(and again, NOT sli, but total double).

it all depends were its clocked, the 5870 is about 2x the 280, and we havnt seen a large review of the 5870 overclocked if it clocks and scales like the 4890 it will have a huge gain compared to the 300 if that clocks and scales like the 285. and price/watt will also be interesting it looks like 4 oced 5870 will go against 2-3 300's in wattage and price.

its going to be interesting to see performance and openCL now that khonos is finally validating drivers we should finally get some pro grade software (although im not sure what it will do for consumers)

[Farinorco]

2x GTX 285 performance(not sli, but DOUBLE) would be better than "par" with the 5870...

Of course, no one knows the full performance on these cards, as been stated, but I heavily doubt it'll be less than double the GTX285(and again, NOT sli, but total double).

Yeah but... why do you expect it to have 2x the performance (I'm suppose you're talking about real world performance) if it's going to have +113% CPs more but only +50% ROP more, +60% mem bandwidth more...

Consider that HD5870 is exactly double the HD4890 (+100% everything at the same clocks) except bandwidth (aprox. +30%) and it's far from double the real world performance (that's one of the most recent proves that doubling everything doesn't mean doubling real world performance), and NVIDIA is not even doubling processing units.

Can they improve the performance per unit and per clock? Sure. Maybe. But how much and why, I think is way soon with the info we have to say it's going to be 2x real world performance of a GTX285. I even would say I hugely doubt it, given that they are more focused in get the new (future?) HPC market before Intel has their Larrabee working (if it happens to be on this century).