Slashdot Mirror
Energy Efficient Graphics Processors?
An anonymous reader asks: "The trends for graphics hardware these days seems to be to draw more power and create more heat to get faster processors and push more polygons. Yet in the CPU arena chips like the Via C3 and Epia, Transmeta Crusoe and Astro, Intel Pentium M, and IBM/Motorola PowerPC (G3-5) seem to favor more power per megahertz and cooler runnings without significant performance loss. Is this just because of the nature of the CPU versus GPU? I understand a GPU die is almost entirely reserved for calculation while the CPU is only 20% of so for calculation. Or are the graphics chip makers merely refusing to innovate and take routes that would reign in out of control energy consumption because of the race for more polygons? What kind of architectural changes could be implemented to alleviate graphics card power gluttony?"
The latest Pentiums are power hungry hogs too, if you want the latest and greatest it's going to be less effienct than it could be. Low power consumption, size of heat sinks, volume of fans are less of a design constraint that the raw power of the chip.
09F91102 no, 455FE104 nope, F190A1E8 uh-uh, 7A5F8A09 that's not it, C87294CE no. Ah! 452F6E403CDF10714E41DFAA257D313F.
Most versions of the Geforce FX 5200 (non-Ultra) run fanless, which should speak to its relative energy efficiency. It also runs about as fast as a Geforce 3, unfortunately.
For great justice.
You've compared high-end 3d desktop gamer cards which are excessive on heat and power to CPU chips which are designed for lower power low heat situations. The difference isn't nearly as pronounced with a more valid comparison on the CPU side, say a high end P4EE or Athlon64/Opteron. Also as you've stated, the GPUs are almost entirely dedicated to high-power processing, whereas the CPUs spend a lot of their silicon on other things. A high end GPU is generally superior to a high end CPU in terms of raw computing power. Therefore, it needs more power and makes more heat. If you forced intel or amd to build a CPU for you right now that had the raw compute potential of the latest high end cards, they'd have a hard time doing so without being just as hot and power hungry. All these things scale over time, but the demands of the user and his software scale up to meet it as well.
11*43+456^2
One of the problems is that PCI and AGP boards are "upside down" compared to ISA boards. Think component-side versus solder-side. In the case of ISA and PCI boards, it's important not to exceed a certain width because of adjacent slots, but since the AGP slot is always the first one, an AGP board could extend pretty far in the other direction.
Why don't they simply mount the GPU to the other side of the board to allow a much larger heatsink? I think this is either a design tradition or a limitation of the pick-and-place assembly machines, because there's no technical reason not to. I suppose if taken to an extreme, it could lead to physical fit problems in certain cases, but let's not go that far.
I tried a no-name GeForce 4 MX440 a couple of years back, but the display quality was awful. It was so poor I had to downgrade to 1280x1024 on my 19" Trinitron screen. After a few months the card broke and I went back to my TNT2 Ultra (Creative Labs) and back up to 1600x1200.
I was thinking about getting something fanless and by nVidia since their (binary-only) drivers are superb on Linux (I don't do the idealogical zealotry as much nowadays).
High-performance 3D is nice when you need it, but nowadays stuff is so powerful for under $100 that there's not much point to buying something really expensive. Some of us want a crisp, high-resolution display, flicker-free (70Hz+) without a great big noisy fan.
Stick Men