ATI 5870/4850/4870/4890 Users for Benchmark WANTED

[Boombeng]

HD4850 Gainward Golden Sample - 10.6.3

Benchmark Highend GPU

 

 

Note that the score is lightly better than Hangten's one that I guess has a stock HD4850 ; the only difference between Gainward HD4850GS and stock HD4850 is GPU and Memory clock (700/1100MHz for the GS and 625/923MHz for the stock), they have the same quantity of shaders.

[mitch_de]

Thanks !

 

Have you tried the new (in 1.5.4) sponza scene, which puts very much load on OpenCL ?

I get in GPU only mode (3 threads on NV 8800GTX, needs in High Benchmark Mode 59 sec) around 16 kSamples/sec (avg.) and 453 kRays/Sec (avg.).

Let it run some time - avg. values needs some time (at least 2 minutes of fast gpu, 5 min on slower gpus)

Would be interesting if ATI 4870 is also 3,9 times faster or even more faster (4+) / less faster (3+).

 

EDIT: iMac 27" ATI 4850M ist incredible fast for an mobile GPU !

Benchmark high mode in 21 sec , RESPECT !

[Boombeng]

Here we go,

 

still gainward HD4850 GS and 10.6.3

 

Sponza Scene 640x480 Gpu only 3 threads :

 

91 kSamples/sec

1876 kRays/sec

 

[mitch_de]

Thanks !

91 kSamples/sec ATI 4850 vs 16 kSamples/sec with Nvidia 8800GTX - RESPEKT

Even more difference than in the benchmark mode, which does less complex scene luxxball as benchmark.

 

My 8800GTX got only 16 kSamples/Sec = 5.7 times slower !

8800GTX =59 sec in High Benchmark Mode vs 17 sec ATI 4850 = "only" 3.5 times slower

[mitch_de]

Updated to 1.5.5.

Added Ultra highend Benchmarkmode (diff to High: uses much more complex sponza scene than luxxball ) !

8800 GTX = 101 sec

GTX 285 (Mac) = 44,7 sec

As before, any 48xx gpu will outperform that (pain for my 8800GTX )

[Boombeng]

As expected...

1.5.5.

Ultra high Benchmark

Gainward HD4850 GS

[mitch_de]

Thanks !

You can see how much the gpu boosts with slg 1.5.6.

Added Highend Hybrid Benchmark + Highend CPU only Benchmark for that.

Time used for Hybrid (CPU+GPU) is much less than CPU only when GPU is fast -like ATI 4850 or GTX 285.

The faster the GPU and slower the CPU the more time diff= boost.

ATI 4850 should give on C2D or i5/i7 big time safing running Highend Hybid vs Highend CPU only.

I get "only" around 30% boost by my 8800GTX

 

The also added UltraHigh Hybrid needs an CPU with 4 real Cores to get speedboost by gpu, because

the cpu load is so much (extra for CPUs >= 4 real cores like MacPro XEON) that the OpenCL driver task cant be "feeded" fast enough.

You may get, in diff to HighEnd Hybrid, near zero or even negative(slower) effekt running UltraHigh Hybrid vs UltraHigh CPU only. In this cases your cpu is overloaded with work- XEON or i7 Desktop CPU needed for good gpu benefit running UltraHigh Hybrid

[Boombeng]

1.5.6 Hybrid vs Cpu only

Q6600@3GHz + HD4850 Gainward GS

 

Highend Cpu only = 30.1 sec

Highend Hybrid = 13.1 sec

[mitch_de]

1.5.6 Hybrid vs Cpu only

Q6600@3GHz + HD4850 Gainward GS

 

Highend Cpu only = 30.1 sec

Highend Hybrid = 13.1 sec

 

WOW !

I had not guessed that you get using an Quad CPU also so much speed boost by the ATI 4850 !

You can imagine what happend if there where an ATI 5850 or Fermi GTX470 build in

At least for very fast MacPro 2009 (2*4 real CPU Cores) such an ATI 5850 may helpfull to get also more than 60% Speedboost by the GPU. My NV 8800GT would give such an Mac less than 5% speed boost

 

Could you run UltraHighend hybrid vs Ultrahighend CPU also ?

I get in this Ultra hybrid an boost down - my 2 cores CPU is so overloaded with cpu tasks that the gpu cant get the data fast enough = cpu only same time as hybrid in UltraHigh. So i guess 4 cores+ CPUs needed sometimes.

[mrheat]

My Benchmarks:

 

GPU Only:

 

Midrange: 9.3 sec

Highend: 15,7 sec

Ultrahigh: 24,0 sec

 

ATI 4870 1GB Powercolor PCS+

Intel Core i5 750 2,66 Ghz

Asus P7H55-M

4 GB DDR3

 

All @ Stock Clocks for now.

[mitch_de]

Thanks !

Could you try the new Highend /UltraHigh Hybrid (CPU+GPU) vs Highend /UltraHigh Hybrid CPU only Modes ?

(They are available since Version 1.5.6 - no other changes)

They show better the difference of real life , how much your GPU boosts your CPU (+ some depends of Systembus Speed).

CPU Speed does matter - the faster the CPU the more speed an GPU must have to get an BIG boost.

I get 16,6 sec in Highend Hybrid vs. 29,9 sec in CPU only (C2D 3.0 Ghz, 8800GTX)

Good boost, even on slow 8800GTX.

Same GPU (8800GT) in an MacPro 2009 (2*XEON, 8 real cores) will boost (=time saving in % compared to CPU only) much less.

[Blaap]

core i7 920 on a intel dx58so

and a club3d 4890 1gb (which is somehow recognized as being a 4870 in smallux)

 

bechmark hybrid:

Highend: 11.7

Ultrahigh: 23.8

 

CPU only:

Highend: 30.1

Ultrahigh: 30.9

 

I'm not sure if this is good but i guess its worth noting that the cpu only tests are getting almost the same score. I'm running 10.6.3 with the newest vga drivers which took me forever to get installed right

[mitch_de]

Good boosts !

30.1 vs 11.7 ! Near 3 times faster using gpu as booster than i7 CPI alone.

The UltraHigh 30.9 vs 23.8 cant boost so much , because the CPU itself has much cpu Load.

[mitch_de]

SLG updated to 1.5.8 !

[Pencs]

Not sure what's going on, because seeing everyone else's results, these seem a little far fetched....

 

I'm running a i7 920 @ 2.7ghz, with 12gb RAM and a Sapphire 4890.

 

[N00bishMacer]

Here is what l got with my Hackintosh Q6600, 4G Gskill, P35-DS3P and ASUS 4870:

 

 

 

 

[nicko_029]

Hi,

 

sorry for a little bit off topic, but has anyone manage to get dual monitors working in 10.6.2/10.6.3 with 4830/4850/4870?

 

Thx

[mitch_de]

Not sure what's going on, because seeing everyone else's results, these seem a little far fetched....

 

I'm running a i7 920 @ 2.7ghz, with 12gb RAM and a Sapphire 4890.

 

You cant compare different versions of the slg benchmarks !

At least with 1.5.8 some major changes (ny slg itself) and changed settings the results vary to older versions.

The next 3-4 versions (like 1.5.3... 1.5.7) will not changed in this case.

[@ROBASEFR]

Smallux 1.5.8 with OSX 10.6.3

Luxbalscene 640x480

Corei7 920 3GHz

ATI 4850 GS Gainward 512Mb

 

Benchmark Modes CPU only and Hybrid

 

High CPU only : 62,4 s

Ultra High CPU only: 62,1s

 

Highend hybrid: 23,5s

Ultrahigh hybrid: 30,1s

 

Benchmark Modes GPU

 

Midrange GPU: 6,5s

Highend GPU: 27,3s

Ultrahigh GPU : 1,6s

[mitch_de]

Ultra High GPU only was an Bug.

Now 1.6.0 available !

I added OpenCL Pixel Filter benches and cleanded up the gui.

Now all gpu only benches ware beside cpu only and hybrid and use same settings. Before the gpu only benches

had own settings compared to hybrid + cpu only.

Now its more clear and should be bugfree.

Ready to collect references again (will hold next versions).

 

Att pixelfilter Mega Samples/Sec of 8800GTX and Ultra GPU only (4870 will perform much faster, but not anymore 1,6 sec )

[kerr]

Smallux 1.6.2 with OSX 10.6.3

Luxbals 800x600

 

 

 

 

 

[mitch_de]

Thanks !

Really huge MegaSamples/Sec in the new pixelfilter bench !

I will use that 4870 values as reference for that model.

 

4850 is a bit slower, but as before much faster than 8800GTX or even GTX 260.

 

PS: The scene or resolution settings havent any effect in all benchmarks. They all use fixed settings.

[MaLd0n]

smallluxGPU 1.6.2

 

 

 

 

[mitch_de]

Thanks for detailed results.

Compared to the fastest Mac Nvidia GTX 280 comuting on GPU (by OpenCL) is much faster.

But ATI OpenCL driver part of OpenCL pciture support isnt fast as the "normal" OpenCL driver part.

GTX280 with a Core i5-750 2.66ghz

CPU Only Ultra: 74.4sec

Hybrid Ultra: 40.4sec

GPU Only Ultra: 39.7sec (ATI 4870 near twice as fast)

 

OpenCL pic filtering benchmark: (ATI is weak here, compared to normal OpenCL code speed)

FILTER NONE: 875.99M

FILTER PREVIEW: 272.68M

FILTER GAUSSIAN: 142.42M (twice as fast ATI 4870! in complex picture OprnCL filtering)

[sdfsdfsdf]

1.6.2

HD4890, 1GB, weird Japanese brand.