[TechReport] Inside Fermi's graphics architecture

8273 Views 150 Replies 61 Participants Last post by NoDestiny, Nov 25, 2009

003

Discussion Starter · Nov 23, 2009

Quote:

If you've been following Fermi since it was announced, you'll know Nvidia didn't really talk about the specific graphics transistors in Fermi implementations. We're going to take a stab at that, though, using information gleaned from the whitepaper, bits teased from Nvidia engineers, and educated guesswork. Remember, however, that graphics transistor chatter does ultimately remain a guess until the real details are unveiled.

Quote:

The GF100's architecture means the SKU we've described (the GeForce GTX 380, possibly) comfortably outruns the GeForce GTX 285 in every way, to the point that (and we really generalize here, sorry) it should usually be at least twice as fast.

See less See more

Reactions: 4

1 - 20 of 151 Posts

dejanh

· #2 · Nov 23, 2009

Interesting...my current NV card is dying and will be needing a replacement in the near future. I'm not willing to buy however until I see what the top-dog card will be...so hurry up NV so I can make my decision...

Which will it be, HD58xx or NV3xx

On topic, are those not the numbers of the non-GeForce cards? That could mean higher out of the door numbers still...

See less See more

Monsk

· #3 · Nov 23, 2009

Cool if true

See less See more

sLowEnd

· #4 · Nov 23, 2009

What's a hot clock?

JKBenchmarks

· #5 · Nov 23, 2009

Quote:

Originally Posted by sLowEnd

View Post

What's a hot clock?

Other word for shader clock, I guess.

See less See more

Open1Your1Eyes0

· #6 · Nov 24, 2009

Quote:

Originally Posted by sLowEnd

View Post

What's a hot clock?

Random "techno-babble" term for shader clock.

See less See more

88EVGAFTW

· #7 · Nov 24, 2009

If this true, looks good. Can't wait for the new nVidias

See less See more

xlink

· #8 · Nov 24, 2009

hope it's true. I'll be out for a new card in early 2011, by then fermi should be putting up a good fight against ATi's 6 series or so I hope. While I'm expecting a single chip fermi to underperform the 5970 I do think it'll likely end up faster than the 5870... it better be, I don't want another FX5900

skunksmash

· #9 · Nov 24, 2009

not sure if im reading this right, but by the looks of it the HD5870 is still in front (Tflops) ??

See less See more

mechtech

· #10 · Nov 24, 2009

2

Quote:

Originally Posted by skunksmash

View Post

not sure if im reading this right, but by the looks of it the HD5870 is still in front (Tflops) ??

*not trying to be biased one way or the other with this response*

Teraflops don't transfer to 3d performance. Tflops are useful as a metric for the total computing power of the card, which only really shows itself in scientific processing or GPGPU type activities.

As an aside, most scientific processing and many GPGPU functions need dual precision math, which Fermi is pretty much designed from the ground up to steamroll. (even is Fermi turns out to be a total flop, it should still kill the GPGPU benchmarks with its MIMD architecture).

RV870 wins in SP TFlops, but DP TFlops are what really matters to get those juicy supercomputing contracts.

Keep in mind though that RV870 is just a doubled RV770, which is essentially a retooled R600. AKA, ATI has had a long time to develop a new architecture, you can bet that they have some MIMD stuff cooking up in their back rooms.

See less See more

Reactions: 1

Eastrider

· #11 · Nov 24, 2009

Nothing to be completely be sure about yet (I mean the 2xGTX285 supossition), but well... at least it's respladed by numbers.

skunksmash

· #12 · Nov 24, 2009

2

Quote:

Originally Posted by mechtech

View Post

*not trying to be biased one way or the other with this response*

Teraflops don't transfer to 3d performance. Tflops are useful as a metric for the total computing power of the card, which only really shows itself in scientific processing or GPGPU type activities.

As an aside, most scientific processing and many GPGPU functions need dual precision math, which Fermi is pretty much designed from the ground up to steamroll. (even is Fermi turns out to be a total flop, it should still kill the GPGPU benchmarks with its MIMD architecture).

RV870 wins in SP TFlops, but DP TFlops are what really matters to get those juicy supercomputing contracts.

Keep in mind though that RV870 is just a doubled RV770, which is essentially a retooled R600. AKA, ATI has had a long time to develop a new architecture, you can bet that they have some MIMD stuff cooking up in their back rooms.

thx......

See less See more

ZealotKi11er

· #13 · Nov 24, 2009

Most likely is going to end up same level as 4870 was with GTX280 but this time they are late in the game.

KarmaKiller

· #14 · Nov 24, 2009

Looks good. Can't wait for the new cards to hit!

See less See more

criminal

· #15 · Nov 24, 2009

Looks very good. Lets all hope it is true. Should be about 30-50% faster than the 5870 me thinks.

almighty15

· #16 · Nov 24, 2009

So by the looks of it, Fermi could end up being faster but the specs aren't as high as I was expecting, heck if ATI release a 5890 and it's clocked high enough it could match or even beat those specs.

Considering the wait, I'm disappointed.

BizzareRide

· #17 · Nov 24, 2009

Based on the specifications alone, we knew it would be nearly twice as fast as GT200. You can't expect to be take serious if you thought otherwise.

With that said, I'm tired of people talking about floating point operations(Flops) as if it is some benchmark of real world performance.

FLOPS is just a measurement in the most optimum(Re: Perfect) conditions and will in no way translate to real-world performance superiority. Its completely theoretical assuming 100% efficiency which will NEVER be the case.

Price point should be $399-449, subsequently forcing ATi to lower the HD 5870's price a bit.

JKBenchmarks

· #18 · Nov 24, 2009

Supercomputing clusters will love this.

Gamers will go to ATI.

tNok85

· #19 · Nov 24, 2009

I have proof that I can't give that's behind the NDA that could kill you. It's so powerful, just hearing the information could kill you.

Be glad there's an NDA.

ZealotKi11er

· #20 · Nov 24, 2009

Quote:

Originally Posted by BizzareRide

View Post

Based on the specifications alone, we knew it would be nearly twice as fast as GT200. You can't expect to be take serious if you thought otherwise.

With that said, I'm tired of people talking about floating point operations(Flops) as if it is some benchmark of real world performance.

FLOPS is just a measurement in the most optimum(Re: Perfect) conditions and will in no way translate to real-world performance superiority. Its completely theoretical assuming 100% efficiency which will NEVER be the case.

Price point should be $399-449, subsequently forcing ATi to lower the HD 5870's price a bit.

I have a felling that no ay this card is going to be under 450$. Look at GTX285.

See less See more

1 - 20 of 151 Posts

This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.