![]() |
|
|
#1 |
|
Aug 2002
2·32·13·37 Posts |
How much of a penalty is onboard graphics?
Back in the bad old days, it was significant, but I theorize that with today's increased memory bandwidth and improved graphics core it no longer is an issue... Here is a sad benchmark from my old 2.4GHz P4... (133MHz FSB, DDR266 and onboard graphics) Code:
Intel(R) Pentium(R) 4 CPU 2.40GHz CPU speed: 2392.22 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2 L1 cache size: 8 KB L2 cache size: 512 KB L1 cache line size: 64 bytes L2 cache line size: 64 bytes TLBS: 64 Prime95 version 22.9, RdtscTiming=1 Best time for 256K FFT length: 12.611 ms. Best time for 320K FFT length: 16.211 ms. Best time for 384K FFT length: 19.425 ms. Best time for 448K FFT length: 22.982 ms. Best time for 512K FFT length: 26.083 ms. Best time for 640K FFT length: 33.028 ms. Best time for 768K FFT length: 40.097 ms. Best time for 896K FFT length: 47.672 ms. Best time for 1024K FFT length: 52.699 ms. Best time for 1280K FFT length: 71.222 ms. Best time for 1536K FFT length: 92.015 ms. Best time for 1792K FFT length: 112.400 ms. I happen to like onboard video, even though I rarely use it... It just comes in real handy for changing crap in the BIOS and stuff like that... So what do I have for you today? How about Intel's newest onboard video... This version has a "console" option that uses only 1MB of memory from the system... This particular motherboard/chipset also allows for DDR400 memory in either single or dual channel mode... So the question is... Has an improved graphics core and increased memory performance negated onboard graphic's bad reputation for stealing bandwidth? You tell me... Code:
AGP/DC AGP/SC PCI/DC PCI/SC 384 14.883 15.125 14.858 15.049 448 17.659 17.942 17.637 17.876 512 20.097 20.410 20.076 20.327 640 24.111 24.395 24.084 24.286 768 29.336 29.705 29.255 29.579 896 34.679 35.055 34.632 34.937 1024 38.863 39.300 38.799 39.172 1280 51.195 51.840 51.006 51.587 1536 62.837 63.554 62.681 63.464 1792 74.619 75.414 74.543 75.171 2048 84.674 86.000 84.595 85.212 PCI = PCI video card with onboard video disabled... SC = Single channel (1x128MB) DDR400 memory... DC = Dual channel (2x128MB) DDR400 memory... In all cases, having Hyperthreading enabled was slower, so I didn't even bother to type in the numbers for it... What is amazing here is not only are the onboard graphics numbers good, the single channel numbers are too! I think that at least in the case of the 2.4C, there is adequate bandwidth in a single stick of DDR400 to run work efficiently, even with onboard graphics turned on... A year ago this was not the case... Yes, there is a slight improvement when a real video card or dual channel memory mode is used, but offset that with a $30 savings per node for sticking with single channel memory and figure in the savings from not having to buy video cards or the time savings from not having to deal with a cardless system... Everything is a compromise and I think this is the best scenario if cost is a concern... For these benchmarks I used an Intel 2.4C P4 200MHz FSB processor ($170), an Intel D865GLC motherboard ($109), one or two sticks of Crucial DDR400 memory ($30 or $60) and a 200W Sparkle power supply ($20)... The memory on this board was run at the SPD setting of CAS3... Not counting the storage solution, the actual node cost is $329 per node, delivered... Lessons learned... 1) Onboard graphics (Intel Extreme Graphics 2) are now viable in 1MB console mode... 2) For a 2.4GHz "C" processor, one stick of DDR400 memory in single channel mode is adequate... 3) 128MB is more than enough memory for a 33M test, especially if you use Linux... 4) Hyperthreading is not worth enabling... Links... Chipset - http://www.intel.com/design/chipsets/865g/ Graphics - http://www.intel.com/products/deskto...ets/graphics2/ Memory - http://www.crucial.com/store/PartSpe...0B&cat=RAM Power Supply - http://www.mersenneforum.org/attachm..._s2850_100.pdf Processor - http://www.intel.com/products/deskto...sors/pentium4/ Motherboard - http://www.intel.com/design/motherbd/lc/ |
|
|
|
![]() |
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| Memory Bandwidth | Fred | Hardware | 12 | 2016-02-01 18:29 |
| High Bandwidth Memory | tha | GPU Computing | 4 | 2015-07-31 00:21 |
| P-1 memory bandwidth | TheMawn | Hardware | 1 | 2013-06-15 23:15 |
| Bandwidth and site hosting. | Xyzzy | Forum Feedback | 12 | 2007-08-18 23:28 |
| Interesting observation re 2kp+1 numbers | Fusion_power | Math | 18 | 2004-01-09 16:05 |