SPONSORED LINKS
On Sat, 18 Oct 1997, Paul Khoury wrote: > Actually, believe it or not, the DX4-75 is based on a > 25MHz clock, and the 100 a 33MHz clock. Go figure, > since 3*33=99, close to 100. Yes, Intels IntelDX4 aka '486DX4 runs clock tripled. We know that. What PEP is doing is putting an AMD '5x86-133 CPU [which runs clock tripled or quadrupled] onto a 25MHz bus and getting 100MHz out of it. It is obvious that they are running in 4x25MHz mode. If it were easy to get a 33MHz bus rate out of a TP701 I'm sure IBM would have done it if only to help with the very lackluster CPU performance of an otherwise very nice product. > That rating should only be relative, since it seems a lot of 701 users have > also used Linux, and OS/2, which use the processor more efficiently, the same > for resources. The effect of the L1 cache [or any cache for that matter] size is very much up to the code being run. For big things that overwhelm the cache, cache size doesn't seem to matter much at all. On one P133 system I built Quake was less than 2% faster after upgrading L2 cache from 256K to 512K. Quake uses huge sets of data and overwhelms the pitiful 512K L2 cache. I suppose a few MB worth of L2 cache might have helped. For small stuff like DOS apps 512K L2 cache can almost completely cache the first 640K and the code nearly always runs from L2 cache. This would give you a fantastic speed increase. > I thought that the DX4 had a 16K cache, > with write back also. It has 16K L1 cache with optional write back. I believe the magazine folks just didn't pay attention to which CPU was in which machine. Nearly everything I read said the 75MHz TP701 had the DX4 CPU in it. The difference between 8K L1 cache and 16K L1 cache is minimal for most things. Especially with no L2 cache to back it up. And re: Winstone.. Winstone is known for being a pretty darn inaccurate measure of system performance. Winbench can tell you a good deal and is about as a reliable "typical windows application" benchmark as you can get. I personally tend to use Quake and other goodies like ctcm and threadmark to make measurements more in tune with what I do. After all, how a system performs running the apps you use is what mattters. =) ============================================================================ ==== Steve Parker ==== San Luis Obispo, CA ==== Multi-OS & Multitasking ==== ============================================================================