Hi guys. I have a real problem and quite worried. I have just spent a fair bit building a new system and the performance of vray using distrubuted rendering (so i have 12 cores all together) is slower than using just my quad core system.
A simple scene I made takes 112 seconds on the quad core 6700, when using dr it takes 238 seconds. The xeons seem to sit for ages on the last 2 buckets.
I have 2x quad core 5300 2.66 Ghz xeons. 4gb in the server and a supermicro x7dwa-n motherboard.
Any ideas ? I thought the xeons would be at least as fast as my quad core…why are they running so slow ??
Sorry guys, i thought this might be something to do with with DR, but it seems that when i just use the xeon machine it is slower…what the hell. I’ll post another topic in off/topic as this doesnt have anything to do with vray (i think anyway)
mmm so you have twice the power but you are twice as slow … damn, did you load the cpu drivers? .. correct ram? correct voltage going to the cpu’s ? (should alreay be correct but hey)… did you build the system yourself? … well for sure something is a miss.
Although, saying that. When i installed what I thought was the chipset drivers, it installed a load of inf files somewhere. So im not sure if it using the correct ones or not.
For rendering, not all CPU are equal, I think. I think the 5300 is considered a transactional power horse, which is not necessarily what is needed when rendering.
We find that the 6600 and 6700 (as well as our new 9300) represent the best bang for our bucks.
Every other test I performed on these cpus proved that they were twice as fast at processing as the qx6700. Maths calculations, which is what we are talking about here.
Theres no way a quad core 6700 would be slower than dual xeons rated at the same ghz and with a higher level 2 cache…im pretty sure anyway.
have you tried a range of scenes to see if they all get the same result?. the only thing I can think of is if you use lightcache and still have the number of passes set to 4 instead of 8 (2x quad core xeons right?) but even then, it should be faster.
Maybe check the temperature of the CPU’s because don’t they slow down if overheating?. I remember ages ago reading that intel does this instead of allowing the CPU to have a meltdown
its funny though that this only occurs with vray rendering but every other test you have run indicates the correct speed difference. So now i’m thinking that it may be worth trying out some other versions of vray - or at least contact Vlado to see if he can help figure it out. Hopefully its not a problem specific to that hardware setup in vray!
The xeons were not overheating, i made sure to keep a close eye on that. So there was no steppig down or throttling invovled.
The tests I did included vray and the scanline renderer. They both proved to be a lot slower. Maybe this is just a max issue. I didnt have time to do any other tests as Ive now sent the items back for testing. Still waiting to see if they come up with anything or not, and if I can exchange for a different set of components.
In general, for such issues, it is best to test a few different scenes with varying complexity. Different components of V-Ray will stress different parts of the system (e.g. smaller scenes with no textures and with brute force GI will stress raw CPU power; irradiance map/light cache/textures/proxies depend a lot on memory access too).
There are enough benchmarks out there to show that the type of software being run on different types of CPU plays a role.
A lot of “server” CPU’s are geared towards multiple, short transactions. Typically, that would be a web server or a SQL server. These would be required to perform thousands of transactions, each independent of one another, in a very short time. The entire architecture is optimized for this transactional environment and such CPU’s may not be ideal for long running batch processes like a 3ds/Vray render.
Case in point would be that 3ds Max is often used as one of the benchmarking categories to test a CPU’s performance. That is clearly because 3ds places different requirements on a CPU/Motherboard than Excel, even if both programs spend their time doing math calculations. As such, in princple, I disagree with your basic premise.
All the same, when I said “bang for bucks” what I meant was that for the price of a single Dual Xeon machine, I can buy 4-5 Intel Q9300 (at least here, in Costa Rica). Those 4-5 machines, jointly, can run rendering circles around a single Dual Xeon - at least with 3ds.
I agree that if you get a CPU that’s fast enough, you’ll be able to make up for whatever architectural shortcomings it may have. You may be able to hold your own but you wouldn’t necessarily be getting more bang for your bucks.
See http://www.tomshardware.com for interesting CPU/Software benchmarks and note that 3ds doesn’t perform like other software on any given CPU- sometimes much better and sometimes not.
(i think anyway)