[Bubbaloo]

Yeah, something's not right.

[tom]

I think the following entries should be excluded from the list, because they are using more than physically available threads and not reflecting the same conditions with others on the list so it makes no sense sorting them. I also have a linux test with normal threads and it's far behind windows on same hardware.

LU // Intel Core 2 Quad Q6600 // 2.40 // 2 // 1.6 (64bit) // Ubuntu // 23m 14s // 462 (16 threads)
LU // Intel Core 2 Quad Q6600 // 2.40 // 2 // 1.6 (64bit) // Ubuntu // 22m 29s // 477.62 (32 threads)
LU // Intel Core 2 Quad Q6600 // 2.40 // 2 // 1.6 (64bit) // Ubuntu // 22m 1s // 487.7 (128 threads)

And not mentioned in the entry but the following test uses excessive threads and not a valid entry.

rmw // Intel Core 2 Quad Q6600 // 2.40 // 2 // 1.6 (32bit) // Debian // 24m 13s // 443.23


Some tweaks? If LU is reading me I'd like to know more...

LU // Intel Core 2 Quad Q6600 // 3.20 (2.40) // 2 // 1.6 (64bit) // Ubuntu // 17m 2s // 630.46 (some tweaks to ubuntu)

[P@]

I thought those processors were build to do this kind of thing: using many threads to eliminate any process latency? You have to recompile the kernel. Update your compiler too, do some reading on the compiler settings. Consider Gentoo Linux.

[-Adrian]

Yeah thanks tom, the 0 thread setting looks like the fair way to do it. I removed those entries, though they obviously were good spirited

For now i've stopped the ability to add new entries as there were a lot of bogus entries in the last days, and i don't feel like cleaning them over and over.

I'll have to come up with a better system, probably seperating Computer-Name, CPU-Type and CPU-Amount to allow for more flexible combinations. And i think it's better if the new entries get send to me before i give my ok. My original plan was to have the site running all on its own without my doing, but that was a naive and things don't work that way. Going over a couple entry requests is a thing of a minute, so this isn't a problem at all, it just won't be as instant as before.

What do you think about demanding a screenshot to be uploaded? Too much?

[codygo]

The 0 thread standard makes sense if your aim is to benchmark how any processor handles that particular setting, but isn't the point of this benchmark to compare setups? If so, then any tweaks, so long as they are noted and accurate serve help to improve other's setups.

For example, the first time I ran this test I ran it at either 4 threads or 0, not sure anymore, but running it again at 1024 threads resulted in a time to completion that went from around 35 minutes down to 22 minutes - something that might be of value to others trying to optimize their systems.

[-Adrian]

In a way yes, but the problem is that the more factors come into play, the harder it is to get something out of the list. If you want to get a good idea on what cpu to buy (the original cause), the 0 thread setting is the best option. I still don't get why having more threads than cores gives a speed boost at all (admittedly don't know the background of threading). In earlier Maxwell version, the time measurement was inaccurate for the PPC architecture, could that be what happens?

Now i see why a list full of optimized custom setups would be very interesting, but that's an e-penis comparison then, not a liable price/performance guide. The red pill or the blue one, we have to stick to one...

[arch4d]

wow, think i am the first with a dual quadcore xeon pc on vista 64 bit...

Intel Xeon X5365 3.0Ghz 8Gb Ram Vista 64bit

Benchmark 1056.49
Rendertime 10m01s


finally got my new machine !!!

[numerobis]

...I still don't get why having more threads than cores gives a speed boost at all (admittedly don't know the background of threading). In earlier Maxwell version, the time measurement was inaccurate for the PPC architecture, could that be what happens?

yes, i asked that too some time ago... but still no answer.

I think every tweak that helps in "real life" and not only for the benchmark should be possible. I don't know how it works with high thread numbers for v1.6, but some time ago i did the test with v1.5 and it turned out that higher thead settings (32, 64...) resulted in a VERY much higher RAM consumption (3-4x). So i think it's nice for a benchmark but not usable for real scenes... unless you have too much memory.

[sandykoufax]

I agree that a screenshot is good idea.

And how about a password to adjust one's own result?

In current table, we cannot adjust our entry.

So there are too many entries are added by same person.

Thanks for the your effort, Adrian.

[P@]

When you have a cpu with 1 core and set it to one thread, the cpu either renders (100% cpu usage) or writes the result to disk (say 10% cpu usage). And while it writes to disk it does not render. Only, the cpu can do what is called multitasking (it can render and write to disk at the same time) but you have to tell the cpu that it should do this. so you set threads to 2 and the 1 core performs the same thing twice. when one thread writes to disk the other one takes over the cpu usage (more or less in simple words). Now, it could happen, that both cpu's write to disk and cpu usage is low again. so you set more threads. it just happens that for the maxwell scene the optimum is 128 on a quad core. sure it uses more memory but maxwell doesn't use that much in the first place- so you can put your 8 GB back to work. most likely, for long running scenes, two or four times the number of cores as a setting for threads should be fine.

I don't not how maxwell runs on windows but you can watch above behavior in the activity monitor on a mac (10.4) or under linux.

besides, i found the benchwell website to be of huge value when setting up my system as it gives you an idea about where you should be (or could be).

[Mr Whippy]

20m 34s

Benchmark 521.926

Intel Core 2 Quad @ 2.66Ghz (not sure what the correct designation is but it's stock)

Win XP

2 x 2gig dimms (but only 2.93gig seen by XP 32)



Is this enough info?

Dave

[-Adrian]

I won't be manually updating any at the moment, chances are you can add it yourself some time. But it's stored in this thread here if i happen to need it, so that's good

[ivox3]

Adrian, .. how long ago did we dump the old entries ? .. I'm out of the loop .... lol... Are these 1.6 entries only?

[-Adrian]

Check the menu, there are two lists - one for 1.6 (default) and the 1.5 to the left.
I'm piecing a site update together, but it's not progressing quickly. In the future all entries will first go to me for approval. Slower, but in the long run less chaos.

I'm also gonna remove the feature to add CPUs/OSes, instead i'll make a far more extensive list by default and add a 'request' form, should there be one missing. But putting all that in a single drop-down menu would be terrible to navigate through, so i'm thinking about a kind of live-search to quickly narrow down your pick.


Works with Manufacturer+CPU or CPU alone

How to properly include Apple is another story... Apple MacPro Quad (Xeon E5160)?

[ivox3]

Thanks Adrian,.. I asked because my entry is apparently in the recycle bin ... lol ..and another test is in order.

I think that you can drop Apple and sublimate it to Mac Pro Xeon[E5160], ..I suppose you could add the word 'quad', but my opinion is that the nomenclature is enough for distinguishing the processor speed and cores. Visually, ..it's an easier assessment.