-
Some Ballpark Rendering Benchmarks with Xeon, Decklink & Hyperthreading
I just installed a dual dual-core Intel Dempsey 5060 system and did some tests to see if HT had any effect. Got some interesting results. Although the Dempseys have HT but the Woodcrests don’t, I thought it would be interesting to see the results anyway . I used a 60sec clip on the PPro 2.0 timeline as the reference source. The machine specs are:
Supermicro X7DA8 V1.1a
2 x 5060 3.2 Dual-core Xeon
2gig Crucial 667 FB-DIMMs
4 x Seagate 400gb 7200.9 SATA II
Highpoint 2220 PCI-X 133MHz Raid Controller
Decklink HD Extreme (in a PCIe x4 slot)
BFG 7900 GT VGA
Dual 24″ Dell 2405’s
Shutle Pro 2
Wacom 9×11 Intuos 3Render times (all sources/exports use the same 60sec 10-bit BMD uncompressed clip)
The RAID (at half full) will do 209mbs read and 166mbs read.Straight 10-bit Export
HT On 1:07
HT Off 1:0510-bit Export with Magic Bullet WArm & Fuzzy Filter Applied
HT On: 2:47
HT Off: 2:48Abode Media Encoder to 1024k WMV PAL 720 x 576
HT On: 5:32
HT Off: 4:26Procoder 2.0 to 1024k WMV PAL 720 x 576
HT On: 1:22
HT Off: 1:28Cinemacraft to 6000kbs 1-pass CBR (System)
HT On: 1:00
HT Off: 0:38It would seem some apps like Procoder may like the extra threads in HT but Cinemacraft doesn’t. It looks like Premiere native export shows little difference whether HT is on or not. I suspect CC may be using the HT threads when it should be using the actial 3.2 cores in the processor.
Anyway, it wasn’t a scientific test but may be of interest to those who need to know…
Peter Corbett
Powerhouse Productions
Australia
http://www.php.com.au -
4 Replies
-
Thanks, Peter! That’s interesting info.
– MythProd
(John David Hutton)Build Your Own VelocityQ. See how:
https://www.customflix.com/Store/ShowTtl.jsp?id=205835_______________________________________
VelocityX, Version 1.0.01
Kansas City, Kansas – United States
_______________________________________ -
As you have noticed, HyperThreading does not increase the speed for graphics intensive applications.
When an algorithm is correctly optimized for P4, your CPU at the subsystem-level should be in use for 100%, even more, HyperThreading makes sure that some realtime events are not so realtime anymore: an instruction can stall when another instruction is running on the other side of the processor. Remember hyperthreading makes you think there are 2 processors where in reality there is only 1.
Also, very often the limit on a CPU for graphics is not really the speed it can calculate things, it is the speed in which it can read/write (= memory bandwidth) data to/from the memory. Hyperthreading does not help this.A nice thing to do is to compare a dual xeon and a dual opteron. Also compare the speed in which it can just copy memory-data from and to the MMX/SSE registers.
-Yves
-
Yes I suspected as much, but as for Opteron, the Supermicro with CPU’s only cost US$650 so the Opterons were just out of reach. I figure I’ll stick with this setup until the quad-cores appear in a couple of months.
Peter Corbett
Powerhouse Productions
Australia
http://www.php.com.au -
Some interesting results regarding HT vs. non-HT. Thanks for posting.
Aanarav Sareen
premiere@asvideoproductions.com
Reply to this Discussion! Login or Sign Up