X1800XT up to 7X faster than 7800GTX

[Ugly n Grey]

Distributed computing programs are nothing new and neither are discussions about using the VPU of your video card to fold. A recently leaked document from the Stanford University Graphics Lab has the X1800XT leading it's rival the GTX by over 7x in certain types of calculations with performance equivelant to a 10.5 GHz P4 CPU.

Maybe the overall architecture lends itself better to scientific calculations or maybe the X1800 series is just everything ATI claims. In any case it looks like it might be THE graphics card for folding at home applications in terms of performance. This could mean a huge boost for Folding at Home teams all the world just like OURS here at XS that recently passed over it's one millionth work packet.


See the attached PDF for details.

[Vapor]

That's pretty freakin sweet--too bad D2OL isn't for GPUs yet, I'd definitely pick up an XT at a little over retail just to help the cause (and my benchmark scores, of course ).

[Cooper]

Impressive.
I wonder how much of that power is used in todays games if XT beats GTX for ~15-20% ???

[P_1]

Distributed computing programs are nothing new and neither are discussions about using the VPU of your video card to fold. A recently leaked document from the Stanford University Graphics Lab has the X1800XT leading it's rival the GTX by over 7x in certain types of calculations with performance equivelant to a 10.5 GHz P4 CPU.

Maybe the overall architecture lends itself better to scientific calculations or maybe the X1800 series is just everything ATI claims. In any case it looks like it might be THE graphics card for folding at home applications in terms of performance. This could mean a huge boost for Folding at Home teams all the world just like OURS here at XS that recently passed over it's one millionth work packet.


See the attached PDF for details.

I wonder how an Athlon 64 stacks up.

[JuanFlaiter]

I thought the X1800XT would be 7x faster than 7800GTX, when dropped from a tenth floor. xD

[twiggy]

Next we'll see folding farms with x1800xt's in X-fire!

[freecableguy]

I thought the X1800XT would be 7x faster than 7800GTX, when dropped from a tenth floor. xD

Hahahahahahhahaha.,........

[One_Hertz]

Impressive.
I wonder how much of that power is used in todays games if XT beats GTX for ~15-20% ???

More like 10 on average... It just shows how inefficient the architecture is maybe?

[Welz]

wait, so we can vold with our gpu's? lol i think i missed something here

[Ugly n Grey]

More like 10 on average... It just shows how inefficient the architecture is maybe?

Maybe how innefficiently we use graphics and drivers.....there's a lot going on when you have to format output for a screen, when you don't, you can get away with murder....

As far as the dropping comment, it was funny , but didn't make a lot of sense in this context really. More of a thread crap, please see the Xtreme Toilet for your needs.

To the question on how an AMD 64 stacks up, on this type of calculation it's pretty close to a RISC chip 1:1 so pretty much tied with a G5 at the same frequency for this type of thing, maybe within 90% of the G5. Not for general use of course (A64 is better) but for science PPC is hard to beat really. One of the reasons for that is they are usually running Unix 64 bit, so it's more about the architecture and the use than the actual chip.

[Mikesta]

I wonder how an Athlon 64 stacks up.

Athlon 64's stack up badly against P4's in F@H. This is because F@H uses Intel compiler that does not allow SSE, SSE2 & 3 to be used with non-intel processors.

So A64's have been shafted VERY badly. Even when competing on certain Gromacs where SSE is not that big an issue P4 still pulls ahead around 10-15% against equivalent a64's. (i.e. P4 3.2 vs a645 3200+).

F@H also loves cache so A64's are weaker compared to Intel proc's.

Stanford better get their asses in gear and get something that helps us AMD users produce more.

[P_1]

Athlon 64's stack up badly against P4's in F@H. This is because F@H uses Intel compiler that does not allow SSE, SSE2 & 3 to be used with non-intel processors.

So A64's have been shafted VERY badly. Even when competing on certain Gromacs where SSE is not that big an issue P4 still pulls ahead around 10-15% against equivalent a64's. (i.e. P4 3.2 vs a645 3200+).

F@H also loves cache so A64's are weaker compared to Intel proc's.

Stanford better get their asses in gear and get something that helps us AMD users produce more.

I see, but aren't they making a custom version to run on the x1800xt? If they can do that, why not make a custom AMD version?

[STEvil]

Athlon 64's stack up badly against P4's in F@H. This is because F@H uses Intel compiler that does not allow SSE, SSE2 & 3 to be used with non-intel processors.

So A64's have been shafted VERY badly. Even when competing on certain Gromacs where SSE is not that big an issue P4 still pulls ahead around 10-15% against equivalent a64's. (i.e. P4 3.2 vs a645 3200+).

F@H also loves cache so A64's are weaker compared to Intel proc's.

Stanford better get their asses in gear and get something that helps us AMD users produce more.

would this be a good application to test the geniune intel check remover then (bypasses this check which allows intel cpus to use sse/2/3 but not other cpus)?

[agenda2005]

DANG DANG DANG!!! GOOD FIND. WAO I didn't know that these X1800's will pull this kind of stunt! I may have to pick up two X1800XT and pair them with a dual core-dual CPU(4 CPU) Opteron. Offloading all the force calculations to the GPU will definitely do some damage in molecular dynamics. Thumbs up to the Stanford team. I've been following them, but previous results were not as encouraging as this new ones. ATI get my money this time.

[Lithan]

That's one of the wierdest things I've seen lately. Might be some insight into the architecture of ati's chips.

One hertz, cooper, where are you getting these numbers from? I haven't seen any benches yet and really want to see some.

[Mikesta]

would this be a good application to test the geniune intel check remover then (bypasses this check which allows intel cpus to use sse/2/3 but not other cpus)?

HELL YES!

Plz do test it. It would help me, and others immensely!!!!

[MaxxxRacer]

pretty freeking intersting....

Just wish the benches relfected this.. but if they did, they could charge 900 dollars for an X1800XT and still sell them faster t han they could build them.

[Cooper]

I agree with UnG on inefficient hardware use.
Lithan, actually I pulled this number out of my finger - I`m really confused with all those reviews. IMO non of todays alpha games have the ability to fully use R520 hardware.
But I think with the right software/drivers X1800 could be more than 20% efficient against GTX. It`s just for the software developers to use R520 architecture fully. This one will take time to complete. Anyway this will be achieved to succesfully use WGF 2.0 compatible hardware - and R520 architecture is closer to that than G70.

[slider99]

I agree with UnG on inefficient hardware use.
Lithan, actually I pulled this number out of my finger - I`m really confused with all those reviews. IMO non of todays alpha games have the ability to fully use R520 hardware.
But I think with the right software/drivers X1800 could be more than 20% efficient against GTX. It`s just for the software developers to use R520 architecture fully. This one will take time to complete. Anyway this will be achieved to succesfully use WGF 2.0 compatible hardware - and R520 architecture is closer to that than G70.

You mean like not implementing full SM3.0? Leaving out essentials such as "Vertex Texture Fetching"? Perhaps, Epic ought to recode their whole U3 engine to make it "more efficient" on ATI's hardware?

It will be truly interesting to see what nVidia does next (81.xx drivers + 512MB 7800 + ?). I can agree that fear was really a fright.

[n00b 0f l337]

Ha! Yes! Truth be revealed finnaly! I think this is were its paying off for intel. r520 may not be the best but there technology, like 90nm and maybe 65nm coming up and there other architechtures are amazing.

Also look at the graph, the 7800gtx is the only thing that works better under the full application over just kernel....

[One_Hertz]

One hertz, cooper, where are you getting these numbers from? I haven't seen any benches yet and really want to see some.

Yeah I just kinda looked at the reviews and made my own assumption according to the games that I play. Will be different for every person.

[Ugly n Grey]

You mean like not implementing full SM3.0? Leaving out essentials such as "Vertex Texture Fetching"? Perhaps, Epic ought to recode their whole U3 engine to make it "more efficient" on ATI's hardware?

It will be truly interesting to see what nVidia does next (81.xx drivers + 512MB 7800 + ?). I can agree that fear was really a fright.

The vertex texture fetching issue is a non starter, it's already been explained in other threads how ATI has built the work around for that. Until we see tests of it there no way to see if IQ is impacted. To me the issue is a nothing event, the ATI card will run all those those games just fine and I'm willing to bet they will run fast.

A company that released a game that didn't run on ATI cards or Nvidia cards would be a stupid company indeed.

Almost makes me wish I played games so I could join in the fanaticism over subtleties that make no real difference. This goes for the IQ complaints against Nvidia and the whining about the SM 3.0 ATI functions. In the end they get fixed and the games work. Then somone benches them and everyone is happy all over again and the bandwagon gets full again... sigh

In any case, the real question is how fast will the different architectures calculate these science packets. I certainly hope we can use GPU's from both companies to fold for team 36362 in the near future. Think about the numbers I could put up on a system with 4 875 Opterons and a pair of XL or XT's!!!

[perkam]

Think about the numbers I could put up on a system with 4 875 Opterons and a pair of XL or XT's!!!

Yea, but Nvidia's GPU was never designed for this type of folding and I'm sure there are other types of calculations it can do faster.

Exhibit A: 7800GTX doesnt lag behind as much in every test




Exhibit B: This has more to do with ATI's Architecture and less to do with whether the X1800 is faster than the 7800 (even the x800xtpe ends up faster in these calculations)




Exhibit C: In ATI's defense, the X1800 could show even more performance seeing as how it isnt optimized for the X1800's new "scatter" feature



I just wish this defeats the need for a ppu in the future if companies continue to keep making hardware with 3 to 4 times the processing power of today's cpus.

Perkam

[Shadowmage]

You mean like not implementing full SM3.0? Leaving out essentials such as "Vertex Texture Fetching"? Perhaps, Epic ought to recode their whole U3 engine to make it "more efficient" on ATI's hardware?

It will be truly interesting to see what nVidia does next (81.xx drivers + 512MB 7800 + ?). I can agree that fear was really a fright.

lol


Go read my post in that thread you're talking about.

[Ugly n Grey]

Or imagine equipping your folding rig with a few of THESE Clearspeed devices...

http://www.clearspeed.com/news/pr.php?pr=28

But I am hung on using video cards, as most of the enthusiast community HAS video cards they do not use 24/7