Slashdot Mirror
Tom's Hardware End of Year CPU Roundup
Wister285 writes "Tom's Hardware has just posted one of their now famous CPU comparisons. Aside from looking at all of the nice graphs, they also compare the speeds of overclocked processors with their factory rated counterparts. It looks like the AMD chips just don't overclock as well as the Intel ones do, but when run at their specified level AMD almost always has the best price/performance ratio. Hopefully the upcoming year will be as promising in the processor sector as 2003 was!"
Sure, a Mac is a Mac but there should be a G5 performance comparison with there. After all, not too many Tom's Hardware readers have Itaniums in their home PCs. And with the PowerPC970 (G5) climbing to 3Ghz by March 2004, it should really be included in the article.
If at the very least, they could do speed comparisons on the AMD64, the P4, and the G5 all running various Linux distributions to make it fair. (I'm heavily assuming the Yellow Dog distribution supports the G5)...
"Right now, somewhere in this world, Scott Baio is plowing a woman he doesn't love," - Peter Griffin, *Family Guy*
Overclockability reviews are pointless for a couple of reasons. The first, of course, is that there are never any guarantees - not every one of the famed 300MHz celerons would run at 450MHz, and just because the few samples a reviewer tests overclock well (or poorly) does not mean that all chips will be similar.
The other major problem is that review parts are often hand-picked, nullifying their value as indicators of overclockability completely.
My server
The "thermal problems" with the AMD Athlons is a PERFECT example of why you should NOT read Tom's Hardware Guide! At the very least do not take the articles read at face value without verifying the facts first!
1.) Their P4 was shown to run at a constant 29C. Thermal throttling on the P4 doesn't even start until ~65 or 70C. If the chips were running at 29C, they wouldn't be throttling at all.
2.) The P4 can throttle down to an absolute minimum of 1/8th of it's clock speed, though it's set to 30-50% by default (factory setting) according to Intel's thermal design guidelines. At 30% of it's clock speed, a P4 will still consume easily 20-30W of power, which is WAY more than you can disapate with no heatsink. Yanking the heatsink off a P4 WILL cause it to crash in a very short period of time.
3.) The comment that was made that AMD's thermal sensor could only react to 1C/sec temperature changes was absolutely ridiculous and CLEARLY showed that the author was completely clueless! Such terrible performance couldn't be accomplished by incompetance along, you would really have to TRY and make it that bad!
The whole deal about the instabililties of the PIII 1.13GHz wasn't so much technically incorrect for the simple reason that there was next to no technical info provided, it was almost all just self-congradulation.
I DO judge the articles by themselves, and the articles on Tom's site generally leave a LOT to be desired. The article linked from this story seems to be mostly fluff with a few benchmarks requiring the standard (ie very large) grain of salt.
" so what exactly then is the fastest solution?"
If you have infinite money to spend? Go with an AthlonFX-51. It's the single fastest solution available, but it's at a premium price. The boards are around $200, and 1GB of memory will cost you around $350 because you have to buy Registered ECC memory. The upside of all this is that you're buying components rated for server operation, so you're looking at very high stability. I just built a $4300 computer system for a customer based on the FX-51. I was expecting some problems here and there because it's all such brand new technology, but was pleasantly surprised at the unbelievable stability. Word to the wise: if you're going high end on everything else, go with a high end power supply. A True Power 380 or 430 from Antec is a smart choice. For reference, I went with an Asus SK8N for the mainboard in this case. Also, make sure you get the recommended memory from Asus (listed at the bottom of their website's page for the board). It'll cost you more money, but it's worth it to not have to worry about stability.
If you don't want to spend quite that much, an Athlon64 3200+ is also a good value. Intel has confirmed, accidentally, that it's got a 64-bit desktop CPU in the works in case the AMD64 platform takes off, so you can bet your bottom dollar that we'll probably see a bunch of 64-bit applications available in the next year and a half.
-- "Government is the great fiction through which everybody endeavors to live at the expense of everybody else."
I'm surprised no-one else is bringing this up ....
The review takes pains to point out that AMD-64 binaries are as rare as hens teeth, and for the reviewer's primary audience who are gamers on Windows, and who have to run whatever P4-optimised or Athlon-optimised binaries the games vendors supply, that's pretty much true.
However, for many readers of this august forum, things are a bit more flexible - the only app I run at home that works the CPUs at all hard is digital video processing (transcode / mplayer / mpegenc on Linux), all the binaries for which are of course built from source, thus could potentially be 64-bit if one had AMD-64 hardware and suitable compilers.
Likewise, for the scientific community using Beowulf clusters, who generally run home grown code, this surely has a lot of potential.
Can someone post a summary of the state of the art in terms of AMD-64 binary output from gcc/egcs, and some info on how well it runs with CPU-intensive number crunching like this?
Professionally speaking, all our stuff at work is Java based, and we are looking for price/performance and space/performance ratios - our latest batch of servers (1U pizza boxes with desktop 2 CPU chipsets are the best price/perf compromise) have dual P4's because of the better memory bandwidth of the i7500 dual channel setup compared the dual Athlon chipsets which were stuck at single DDR-266 for the longest time, but if there was a byte compiler which targeted AMD-64 I could see potential for really nice price/performance with the Socket 940 systems, and even just using 32-bit code the higher memory bandwidth would help a lot with Java apps.