Slashdot Mirror
February 2008 Hardware Roundup
Tom's Hardware has a nice roundup of some of the new shiny hardware for February '08. Everything from a screaming fast 2 GHz DDR3 to liquid cooled cases and back again. "Unlike previous Zalman cases that used a heat pipe assembly, the LQ1000 has a traditional water pump and flexible hose for connecting the case's sinks to CPU and graphics coolers. A passively-cooled finned side panel and fan-assisted rear radiator remove heat, while a lighted flow indicator shows the bottom-mounted pump in action."
Please stop linking to "articles" on the page-o-ads tomshardware site and making them money. They have a hardware roundup every month; there's no need to link to them just to improve their ad revenue stream.
- a bunch of novelty cases
- overpriced power supplies
- and 6 new DDR3 modules at varying frequencies
Prediction: The real iPhone killer is going to be sex robots from Japan. Think about it.
It's a link to last measure. Don't click it, just a heads up.
I'm not sure who told you that DDR3 RAM was low latency, but the statement is dead wrong. If you RTFA, you'll see that the new 2+GHz DDR3 has a CAS 10 latency!! While it's true that it's clocked more than double a typical DDR2 module @ 800mhz, these DDR2 modules are typically CAS 4 or 5, and timings can sometimes be tightened even further. The throughput of DDR3 memory is certainly boosted greatly over DDR2, but no matter how you measure it, memory latency has not seen the same improvement.
It sounds like you should be looking at Silent PC Review instead. They focus more on how you can get a moderately powerful computer without it being obtrusive. Maybe a little bit extreme at times, but always good info.
EPIA 800 MhZ Nehemiah core on a nano_ITX mobo with 1 GB RAM, 500 GB 3-1/2" drive, and slot-loading DVD ROM in a Silverstone LC08 case.
Add an 802.11n USB dongle out the back (or traditional wired 100 Mb/s Ethernet), and you're golden.
Linux 2.6.23, alsa, xorg 7.1 with DRI, openchrome, xine --with-xvmc, and mythtv and you can render 1080i at anywhere from 40% (most streams) to (rarely) 95% (some particularly badly coded ones with lost of motion).
I just finished natively building (i.e. compiling from the sources themselves on the box itself) this starting from a stock Damn Small Linux install.
I'm trying to pare it down so it runs completely out of 512 MB RAM disk.
In Liberty, Rene
The only way you're going to see latency cause large effects on performance is if you turn off the cache on your processor, or run some kind of program that always results in cache misses (which isn't at all realistic). The whole point of cache is to mask the effects of latency. For example, if you have a cache with a 98% hit rate (meaning that only 2% of all memory accesses need to wait on memory, which is fairly realistic for today's processors), then if you doubled your memory latency, your performance (average latency seen by the CPU) would only increase by 17%. If you could cut the latency in half, then your performance would only increase by 9%. In reality, there isn't that much variance between different latencies on similar types of memory. (I've assumed that the cache is 10 times faster than memory, which is pretty conservative).
Applying Amdahl's law here, if you want to reduce memory latency, you really want to increase your cache's hit rate or speed, for example by getting a processor with a larger cache or lower cache latency.