ComputeMark v1.2 - CrossFire/SLI support, bench

[stinkas]

*sigh*

Another synthetic benchmark...

*sigh*

Another positive ATI benchmark...

[HelixPC]

yeah, ok, the 5770 is faster than the GTX 480? lol ok. Try again developer.

[Dezmen]

GTX480 score:


Stock speed, 197.44 drivers, Core i7 920 stock.

Seems like a load of total BS if you as me. So a 5850 is 3x faster? Something is definitely wrong.

Lol? Click Run not Run bench

[ElSel10]

Lol? Click Run not Run bench

That's the burn test, not the benchmark. If you look at the post directly above mine, someone ran the benchmark on their 5850 and scored 2007, ~3x faster than the 480.

That is total BS. Even if you consider the maximum theoretical SP performance of the cards, the 5850 is only 50% higher. Nowhere near 3 times higher.

[Dezmen]

That's the burn test, not the benchmark. If you look at the post directly above mine, someone ran the benchmark on their 5850 and scored 2007, ~3x faster than the 480.

That is total BS. Even if you consider the maximum theoretical SP performance of the cards, the 5850 is only 50% higher. Nowhere near 3 times higher.

Tell all that to soft developer, he will answer why that should be

[flippin_waffles]

Wow lol. Does nv have at least something positive about fermi that they can point to? Geez they're getting pounded left right and center!

[saaya]

I will only put in a comment on the F@H-client for ATI-cards here.

Over at B3D there is a thread discussing Folding at Home.

Mike Houston, System Architect in the Advanced Technology Development Group at AMD,
who has taken part in the development av F@H for ATI GPUs made a couple of post
which give some insight to what makes chrunching on a GPU not always straitforward.

His first post starts at #57. Worth the time to read IMHO.

/Kej/

thx!

its really easy to have your app running ok on one card and sh*t on another. he probably wrote the app on an ATi card and didnt have access to an nvidia card. yeah directcompute and opencl will run on both cards but porting performance is much harder.

hmmm really? then opencl and directcompute dont really work that well in unifying things? meh...

as long as its not possible to write general vendor unoptimized code that gets at least 70% of either cards performance this sucks... :/

[Olivon]

Wow lol. Does nv have at least something positive about fermi that they can point to?

Yeah, Just look at performance. Fermi is faster than HD58X0 but with high temperatures and high power consumption, that's the problem.

Don't act like a fanboy (I know it's difficult to you )and look at overclocked performances

[ajaidev]

Yep the fermi scales good with added mhz better than cypress. Specially GTX 470 OCed to 800Mhz performs good but the cost is watt consumption, when i oced my GTX275 it blew out my CM 600w PSU.

[Marios]

Fluid 3D: a Direct Compute 11 example from Jan Vlietinck simulating an incompressible fluid solving the Navier-Stokes differential equations via second order MacCormack including limiters, visualizing the amplitude of the speed vectors using a simple ray maximum projection to give color by linear interpolation. As Jan says, the result is some visually interesting turbulent behavior...

Fluid3D on the HD 5570 was slow to respond to mouse input, highlighting the HD 5570's reduced Stream Processor count vs. the HD 5870 the code was developed on...

Jan was kind enough to grant Rage3D permission to use his DirectCompute examples ...

Five years ago I began working on a new rendering engine, this time making use of real Voxels...

The 3d waves simulator was inspired by a 2d wave simulator I made in the past. It is an example of what can be done with data stream based computing as it involves volume processing and texture sampling...

Also the need of specifying a hardcoded thread block size with the kernel definitions is rather awkward and hardware dependent performance wise

Computemark is based on Jan Vlietinck's Fluid 3D. It looks like an HD5870 hardware dependent bench that involves volume processing and texture sampling. No polygon geometry calculations here. Just voxels.
It is only good to compare between 5 series radeons.

[fellix_bg]

Here is the original thread for the Fluid3D demo on B3D:
http://forum.beyond3d.com/showthread.php?t=55910

AFAIK, the author of the demo does not employ any particular architecture optimizations for his code (there are few more on his web page).

[Marios]

Jan Vlietinck created a CS5 program based on voxels. The code was developed on 5870. As Jan said "For DirectCompute ... the need of specifying a hardcoded thread block size with the kernel definitions is rather awkward and hardware dependent performance wise."
It's OK to program on a specific architecture but this doesn't make the program a universal bench.
There are many ways to accomplish the same results on different architectures.

[Chumbucket843]

ok cool, at least we have some new info rather than just compute mark. i thought it would be a package of different gpgpu tests. if its just fluid simulation then it should be inherently bandwidth bound.

[Helloworld_98]

when I did run bench on my 5770 @ 900MHz it gets 1181, wow the 480 sucks at this

[Vardant]

Wait, it was created by this guy?

Reading other forums, other websites... Definetly it seems like people want PhysX card. Thats not good :-/ AMD should do something. Nvidians PR sometimes just works. Damn.

It's not important if ECC is usable or unusable. It's important that it will be written on the box.

Nvda will sell it. More stripes more Adidas, rule #1.

Prepare for megahooorayy. Monster booom. Hot chick all over every1 with GF100 tatoos on their boobz.

They will demo GF100 as gods new child (Adams bigbugbrother).

You mean two fermis in SLI are less interesting than super-duper-hot-chickie with big ing gun kissing her p*ssy? Jesus than Nvidia is definetly ed up...

Here's the rest - http://www.semiaccurate.com/forums/s...earchid=156783

Do I need to say more?

[Oberon]

For anyone who is interested, NVIDIA has been in touch with the creator of ComputeMark and has been helping him to optimize the benchmark for NVIDIA GPUs.

The new version can be found here: http://www.friendsea.com/ComputeMark/ (direct download)

[Chumbucket843]

For anyone who is interested, NVIDIA has been in touch with the creator of ComputeMark and has been helping him to optimize the benchmark for NVIDIA GPUs.

The new version can be found here: http://www.friendsea.com/ComputeMark/ (direct download)

interesting. the 5870 is 6 times faster than the 5770. that correlates directly to bandwidth and fetch rates.