Holomark 2 Released!

The GPU16 result looks for me like a typical synthetic test result, since the bike model test show me, if a model is slow, than it stay slow, nearly independent from the hardware. During the bike test the GPU/CPU usage is extreme low - the bottle neck is the software. And … AA doesn’t matter too. I tested it, the AA has no big influence on the graphic speed of models like the bike. If the GPU get something to do from Rhino, than it doe’s the job so quick, that it’s no problem to to do the job in high quality too.

Hi nick,
All help would be appreciated so PLEASE make a spreadsheet and share, you can do this just as fast as I can and Holomark is made on my spare time, and unfortunately I have no spare time to gather and process the data right now.

It will happen eventually thoug, but probably after the next SR.

-Jørgen

GPU_16 is designed to show the true potential of the graphiccard in Rhino. It uses ONLY MESHES and uses the Holomark RenderSpeed display mode that has all eye candy turned off, so it handles the mesh data as fast as possible.

Xeon X5650 (2.66GHz Nehalem), 48GB DDR3, Quadro 5000
So, the bottleneck is the CPU.

As you say that, I made three runs with my card. One all maxed out cpu and gpu clock, then one with the gpu values minimized to about 60% of whats possible, and the last one with the cpu throttled down to only 2.5 GHz, which
is nearer to your and Micha’s cpu speed.
And, it’s true, quite a lot of the gpu routines in Holomark2 depend heavily on the cpu, some don’t at all.

I took the liberty to complete the shown values of the equipment within the pictures. So it shows the actual cpu and
gpu clocks. Otherwise comparing these diagrams is not very helpful without showing the actual clocks.
So, Micha, you are right being suspicious wether a replacement of your GTX is worthwhile, as one can see, that my card in your computer would only yield to a betterment of under 20% for GPU_04…
If that would be the only parameter important.
There is a last thing: Initially I thought, Holomark2 sets up it’s own options for all the runs, but then I found out, it makes a difference to turn off antialiasing all together under options view open gl.
I set this graph at the very beginning. The other three runs were with antilaliasing set to two none none.



Your standard CPU spped is quite high, great. Thank you for testing the relation CPU-graphic.
Sidenote: my machine is running at 3.1GHz and if not all CPU are used at 3.4GHZ. So, your card could bring 50s for GPU4 at my machine.

But we are miles away to use the hardware power of our machines. Multithreading and no software bottle neck and we could get an incredible speed enhancement.

Yup. I understand Jørgen does all this stuff in his sparetime, so what can one say… I say that, because Holomark2 should pick up the real clocks somewhere, and not just the branded defaults. For example, cpu-z probably has somehow a way to pipe the data to another program. I am not a coder, so no clue about the details. But I think, I am not the only one to crank up cpu output. Sorry to mix your numbers with some others I had seen. 3.1 isn’t so low at all. Anyway, to get somehow ‘normalized’ results in Holomark, the real picture, the actual clocks must be included or be seen.
Also, because someone asked for a table of all the results so far. Could that be an excel spreadsheet or something, and the Holomark results list could be directly inputted, so one wouldn’t have to write it in by hand?..
Could be made sticky up on this thread, and someone wanting to add his result just uploads the list generated by Holomark…
As for my cpu, I did not want to go to extremes, so it’s only an aircooler on top, and no finetuning done yet. Still, it runs stable whatsoever at 4.2 GHz. Running troublefree is most important. I have two IBM x3650 servers sitting here in the background, 16 cores at 2.5 GHz. But loud as a jet, and blowing hundreds of Watts, so thats only for hardcore rendering. CPU rendering that is, outdated almost, as it stands against modern GPUs…

Unfortunately I don’t know enough about computer components nor the tests to be even able to set up the spreadsheet. e.g. total score or individual tests, which components (or combinations of components), etc?
I realize that you are doing this in your spare time and totally appreciate the work you are doing. But the problem is that you are also the person who understands the test!
Nick

1 Like

here is my score …

not too good :frowning:

Unload the Vray plugin and try again.
I presume that is what causes the low CPU_02 score.
And then upgrade V-ray to the latest SR. It should fix it.

And it seems like Rhino uses the GeForce and not the Quadro, so I would change that and set the Quadro driver setting to Workstation-dynamic streaming.

1 Like

Hello Jorgen,

thank you for your suggestions, indeed I had dynamic streaming enabled at my old workstation but obviously I forgot to turn it on on my new one … did it and the gpu score rose to 17500, gaining one one unit in the important gpu 08 score :smile:

but now comes a big surprise…
I unloaded the vray plugin via the plugin manager and the gpu score reached 34610!
Sadly gpu 08 remained at 9 units… nevertheless i am as amazed as I am clueless what causes this behaviour???

thanks again

Andreas

Hello again … I have another question, why do you think that rhino uses the geforce gtx 580? When I look under rhino options / open gl, it lists the quadro 4000 as the current card …

best

Andreas

…carefully overclocked the hole thing from 3.2 GHz to 3.9GHz slowly we are getting somewhere…

its a shame though that gpu 08 still is the weakest part, because like Micha and Steff said I consider this to be one of the most relevant tests…

I have ordered a used quadro k4000, It should arrive in a few days. I will try if it further improves the scores, I am not expecting much especially on gpu 08, i bought it more as an option for possible future pipeline optimizations…

Andreas

Hi Holo,

what is better, the lower or the higher number of the Total Score?
Where can we find some charts list?

My old computer (built in 2009) score:

Petr, higher numbers are better. There are no chart lists.

Thanks, i though so :slight_smile:
It would be great if the test results will be sent directly (by hitting button send) to a proper place for further analyse or charts. If it is only in pictures, you are not able to search according manufacturer name or other criteria.

Happy New Year!

So, I ditched the laptop in favour of a cooler running desktop, sad to say that the desktop doesn’t perform any better but at least it’s running a hell of a lot cooler.

I need to look into the CPU settings though as it’s supposed to be OCd to 4.2GHz but that might just be what the test is returning rather than the actual clock speed.

Test on a virgin system:

1st tweak:

2nd tweak:

@2DCube: interesting result, top GPU 4 value under moderate CPU speed of 3 GHz. Looks like GTX 970 could be my favorite and I should sell my Quadro 6000.

@Micha you could sell it and go for a 2 way sli at around £600 if you’re using cuda cores for rendering and then go on holiday with the rest :smile:

Otherwise you could go for one card and take someone else on holiday with you!

The price of a Quadro 6000 is around 450€ only (at ebay).