[rico2001]

Quote:

Intel Larrabee finally hits 1TFLOPS - 2.7x faster than nVidia GT200! During the recently held SC09 conference in Portland, Oregon - Intel finally managed to reach its original performance goal for Larrabee. Back in 2006, when we first got the first details about Larrabee, the performance goal was "[email protected] 16 cores, 2.0 GHz clock, 150W TDP". During Justin Rattner's keynote, Intel demonstrated the performance of LRB as it stands today. At SGEMM Performance test [4K by 4K Matrix Multiply, QCD], Intel achieved 417 GFLOPS using half the cores on the prototype card, and reached 825 GFLOPS by enabling all the cores. While looking at the numbers alone, one might think that these scores are below the level of ATI Radeon 4850 and nVidia GeForce GTX 280/GTX 285. Of course, there is a "but" coming - unlike theoretical numbers that are usually disclosed by ATI and nVidia - this was an actual SGEMM benchmark calculation used in the HPC community....

Source

 

[Lord Xeb]

Interesting.

 

[dham]

Haven't we discussed multiple times that that tflops means nothing in game performance. Load up crysis and then we'll see what this means.

 

[Vlasov_581]

Quote:

Originally Posted by Lord Xeb Interesting.

^^^This

 

[sodaholic134]

Great read. Good to hear that Larrabee is performing quite well.

It should be relatively easy for them to make lower-end cards since they just have to disable some cores to give different performance levels to meet different market segments.

Hmm...I can't wait to overclock Larrabee. They got 1TFlop with 16 cores @ 2.0 ghz. I wonder how far we could push the cores?

 

[jrharvey]

Quote:

Originally Posted by dham Haven't we discussed multiple times that that tflops means nothing in game performance. Load up crysis and then we'll see what this means.

You make a good point. Has anyone ever figured out what a Flop calculates? If it is a graphics card then there should be graphics scores. Im tired of seeing how many GFlops a card can do, haha.

 

[Kirmie]

Quote:

Originally Posted by jrharvey You make a good point. Has anyone ever figured out what a Flop calculates? If it is a graphics card then there should be graphics scores. Im tired of seeing how many GFlops a card can do, haha.

Um, did you just ask the computer equivalent of the automobile world's "what does a MPH calculate"? Try Googling the word "flops". Different programs can come up with different FLOPS because of the type of operations being done. Like SGEMM compared to DGEMM (single vs dual precision).

Also, this article is referencing the old Tesla. The new one blows away the old one in double precision (probably going to be way more important to the people who the Tesla GPUs are for). Glad to see the Larrabee is going forward at least. I mean sure, we find out how fast the car can go and accelerate but what about the amount of braking required to grip throughout a turn or the ability of the driver (convenient term) to know how to properly drive the car to get the maximum speed\ime possible through the turn?

Edit: By the way, the graphics score would probably be more useful to over 90% of the people in the world as well, since I'm pretty sure they give a more accurate representation of the ability to produce graphical effects in games rather than just getting stuff done. Measuring the FLOPS to figure out gaming ability is like drag racing a car to see how it will do on a twisty race track.

 

[vdek]

Quote:

Originally Posted by jrharvey You make a good point. Has anyone ever figured out what a Flop calculates?

Are... You... Serious...???

http://science.howstuffworks.com/light.htm

http://en.wikipedia.org/wiki/3D_rendering

http://en.wikipedia.org/wiki/Matrix_theory

http://en.wikipedia.org/wiki/Calculus

http://en.wikipedia.org/wiki/Quarternion

http://en.wikipedia.org/wiki/Euclidean_vector

etc...

 

[003]

Quote:

Originally Posted by dham Haven't we discussed multiple times that that tflops means nothing in game performance. Load up crysis and then we'll see what this means.

This isn't theoretical FLOPS though. It is real world FLOPS. Theoretically, the GTX285 can operate at 1TFlop at peak efficiency, but this never happens. In this benchmark, the GTX285 puts out 425GFlops while Larrabee puts out an actual 1TFlop in a real word scenario.

 

[dham]

Quote:

Originally Posted by 003 This isn't theoretical FLOPS though. It is real world FLOPS. Theoretically, the GTX285 can operate at 1TFlop at peak efficiency, but this never happens. In this benchmark, the GTX285 puts out 425GFlops while Larrabee puts out an actual 1TFlop in a real word scenario.

Oh ok. Well thanks for clarifying this.

 

[corky dorkelson]

Quote:

Originally Posted by 003 This isn't theoretical FLOPS though. It is real world FLOPS. Theoretically, the GTX285 can operate at 1TFlop at peak efficiency, but this never happens. In this benchmark, the GTX285 puts out 425GFlops while Larrabee puts out an actual 1TFlop in a real word scenario.

Looking pretty good for Intel's first showing. (And yes I know they made a fail card some years back).

We are looking at having "the big three" in the discrete graphics market.

 

[003]

Quote:

Originally Posted by corky dorkelson We are looking at having "the big three" in the discrete graphics market.

This will be interesting because Larrabee and Fermi will likely perform extremely similarly.

 

[the_milk_man]

wait wait wait..... A TWO Ghz GPU!!!?

 

[vdek]

Quote:

Originally Posted by the_milk_man wait wait wait..... A TWO Ghz GPU!!!?

A GTX 285s shaders run at nearly 1.5ghz...

 

[DuckieHo]

Quote:

Originally Posted by corky dorkelson Looking pretty good for Intel's first showing. (And yes I know they made a fail card some years back). We are looking at having "the big three" in the discrete graphics market.

This is good show for general processing power... however, Intel does not have as much experience in graphics. Drivers and optimization make a large part of game performance.

 

[Dannythm]

Quote:

Originally Posted by dham Haven't we discussed multiple times that that tflops means nothing in game performance. Load up crysis and then we'll see what this means.

And then again, this is interesting for HPC, not for the gaming community, as this was on a HPC convention.

This is really interesting, but I'd love to see how will it perform with double precision tests, as it's principal competitor on HPC, Fermi, is a beast on double precision calculations.

This only shows that intel won't dissapoint.

 

[darksideleader]

Quote:

Originally Posted by DuckieHo This is good show for general processing power... however, Intel does not have as much experience in graphics. Drivers and optimization make a large part of game performance.

all in good time. Give it a couple months after release and a company like Intel will get the bugs out of everything.

 

[corky dorkelson]

Quote:

Originally Posted by DuckieHo This is good show for general processing power... however, Intel does not have as much experience in graphics. Drivers and optimization make a large part of game performance.

Very true. Was just having this convo with a co-worker. They need some strong drivers, competitive pricing, and pure gaming performance to be able to get the ball rolling quickly. If they don't hit on all 3 of these, they may have some problems for a while.

 

[vdek]

Quote:

Originally Posted by corky dorkelson Very true. Was just having this convo with a co-worker. They need some strong drivers, competitive pricing, and pure gaming performance to be able to get the ball rolling quickly. If they don't hit on all 3 of these, they may have some problems for a while.

They don't need anything to be honest, all they have to do is continue their vendor-lock in strategy and you will see Larrabee on at least 20-30% of all new computers sold.

Especially once the next CPU comes from Intel with an integrated Larrabee on die. Then we will be stuck at GTS250 performance forever.

 

[DuckieHo]

Quote:

Originally Posted by darksideleader all in good time. Give it a couple months after release and a company like Intel will get the bugs out of everything.

It takes Intel over a year to work out driver bugs with their IGPs...