Best bang for the buck for stingy renderers?

[notareal]

I would not take Intel DX58SO. So if that only what available for i7 920, then that route seem closed. I don't think i7 870, is a particular good deal, it's faily expensive ($579.99) when you compare it to i7 860 ($299.99) and not that much faster.

[Ecuadorian]

Ok, guys.

Stinginess, together with a strong will to prove that you can get more for your buck, made my decision. I've just returned from the PC store with a Core i5 750 processor, an Intel DP55WB motherboard, and two 2GB modules of Kingston DDR3 RAM = 4GB. The total was $446.64 + 12% tax = $500.24. Now I'm waiting for a friend of mine to come and put these parts inside my computer. I think I'll sell the old processor, motherboard and RAM for $100, so then I'll be able to say the upgrade was only $400.

I can upgrade to whatever comes to the 1156 socket later. It seems to have a bright future. A nice perk I noticed in this motherboard is that it has a firewire port, so I'll be able to connect my brother's MiniDV camera without any need to install an expansion card. I also like the fact that this mobo got rid of all the useless legacy stuff, such as serial/parallel/PS2 ports and internal IDE connectors. None of my devices use any of those anymore. Oh, and I got two sets of dual-channel memory slots as I wanted , out of which I'll only use one set (two slots) right now. Anyway, XP+SU+Twilight together seldom use more than 1.3GB, even in my "thousands of trees" scenes, so 4GB should be plenty for a long time (and I can't use it all in XP 32-bit anyway).

Thank you again for expressing your opinions. I'll post a "worm" render test as soon as I have this thing working.

[notareal]

Waiting for your speed testing results. It should be somewhere between quad and 920.

[Ecuadorian]

My friend could not come tonight, so I ended up making the upgrade myself... He he he, most of the time I had no idea what I was doing, so this required a LOT of trial and error. I finally learned what those little connectors are...

upgrade.jpg (72.73 KiB) Viewed 10513 times

I didn't understand the CPU retention mechanism, so I just unscrewed that odd-looking screw and put it back after placing the CPU. Yeah, it basically jumped when I unscrew it, it almost caused me a heart attack! I also played for a long time with the little metal "flaps" in the perforated panel that comes with the motherboard. I just could not fit it right in the openings, until I had an idea that worked. I didn't know if I should choose the black or the blue memory slots, so I figured since blue is more eye-catching perhaps I should use those... I also had a hard time making the last of the CPU fan... tentacles... "snap" to the motherboard, until I figured that you can rotate it a little bit and then snap it.

The biggest surprise to me was that the CPU has no holes! I always thought that since the motherboard has pins, the CPU must have holes, but it doesn't! I was also surprised by the small size of the CPU fan. It's about half the height of the fan in my old Core 2 Duo E6300. Looking on the bright side, this means with a bigger fan I might do some overclocking... No, I've never overclocked anything.

The good news is that despite being a total noob to motherbard and CPU installations I managed not to break anything in the process. When I started Windows afterwards it froze during the new hardware detection (I should have figured, with such a monumental change :dho: ), so I had to use the Windows repair function. I'll do the tests tomorrow. Besides the worm test, I did two timed MLT and PM tests with the old parts, so I have a reference point.

[notareal]

Miguel: "When I started Windows afterwards it froze during the new hardware detection (I should have figured, with such a monumental change :dho: ), so I had to use the Windows repair function."

I think many would had given up at that point, I suppose you had a working backup.

[Phil Meadows]

I wish I could make this sign a lot bigger:

[Ecuadorian]

Well, everything is working fine now.

The quick "worm" test showed a 3x difference in speed, but longer tests rendering much bigger images show less of a difference. Let's say the worm test is a "sprint" race, while other tests are "endurance" ones.

I did an MLT 3000x2000 test before and now; the results are:
Core 2 Duo E6300: 17m43s
Core i5 750: 8m02s
Relative speed: 2.2x

In another 3000x2000 test with the same scene, but with "Low" preset:
Core 2 Duo E6300: 5m26s
Core i5 750: 2m57s
Relative speed: 1.84x

I'll re-do the "tatami room" test and post the results, too.

[Ecuadorian]

The tatami room now renders 100 passes in 1h 38m, whereas with the Core 2 Duo it took more than 4 hours (4h 6m, if I remember correctly) for 100 passes at the same size of 800x532. This means it now rendered at 2.51x the previous speed. However, I think the original was done with Twilight 1.0.6, so the comparison is not precise .

Interior-1m38s.jpg (90.38 KiB) Viewed 10457 times

Original render here:
http://twilightrender.com/phpBB3/viewto ... 63&start=0

On the other hand, I've seen a speed increase of 3.5x with the new processor in LightUp, using a model of a stadium and an HDR as an illuminant. I wonder why so disparate speed increases with different render apps and even with the same app (Twilight) when testing different scenes. I expected to have a single increase factor across all scenes in Twilight. What might be causing this? It can't be lack of RAM, as total memory usage has never gone above 1.3GB in any of them. As an added curiosity, I tested the "worm" with different numbers of threads in the Core i5:

1 thread - 103s
2 threads - 56s
3 threads - 45s
4 threads - 40s

You would expect a 4x difference between using 1 or 4 threads, and a 2x difference between 2 and 4 threads, but that's not the case. It just doesn't follow a simple inverse function. I guess that turbo boost has come to complicate testing and show us that processors are now much more complex than we think. Perhaps the specific model of RAM and motherboard, and even temperature influence tests. (My ethernet router moves files faster between computers after I let it cool down, for example). Now you can't simply say "processor A is 2x faster than processor B"... You can only say: "processor A is 2x faster than processor B under these specific test conditions". Also, I remember that I once created a test that gave way faster results with only one thread instead of two because of the processing overhead imposed by multi-threading.

I guess this means three things:

- It is possible to create a new speed test scene that produces results with less marked differences across different CPUs.
- Test results should always be taken with a grain of salt.
- I should stop testing and go back to work.

[Fletch]

remember that portions of the rendering process are not multithreaded... only ray tracing/anti-aliasing are multithreaded (I could be wrong, but I'm certain scene processing is only single-thread no matter how many you have.) So, the shorter the render time, the lower the time diff. will be. To get accurate render time diff. subtract the processing time. Of course, if you processor on 'old' machine was 2mhz, and the new processor is 3.2mhz you will definitely see a speed increase even there, but it will not be proportional to the number of processors you have... and it will of course be effected by turboboost if you have that.

[Ecuadorian]

Thanks for the detail, Shane. That means a "cleaner" test should be done this way to bypass the scene processing time:
- Render once
- Set update method to "none"
- Render again, and this is the render time that counts.