L2 is turned on in the BIOS. The kernel can use it, but it doesn't. I don't believe that Celeron/northwood chip, in fact/ can be as slow @3000, as Athlon XP at 1533 (1800+). I can believe in 20% performance drop, not 100%, sorry.
Especially when L2 size is not reported properly by kernel...
I tried to follow your suggestion of directly measuring performance drop for memory access. I run some old benchmark, but results are not easy to interpret. Performance drops at 32-64K for Cel, and 64-128K chunks for Athlon XP.....;}
It's not so easy, as it seems, maybe you know good cache benchmark for linux ?
I don't like to be called 'lame':( - especially when nobody posts the answer to my question. I posted the problem on comp.os.linux.hardware/or kernel, don't remember/, and got no answer. It seems that suggestion of writing to kernel mailing list is the best solution.
I made google search, of course, found question you mentioned, but I have different problem. POST on my PC doesn't show L2 cache size, (or it disappears too fast;] - I'll check), but in Win2K everything is OK in Sandra L2 cache benchmarks. Cache works VERY fast and performance drops above 128KB blocks.
Probably you are right suggesting to directly contact people working on kernel, thx
It's not the hardware problem! - in win2k with drivers from the MB's CD, Sandra cache benchmark shows great results for L2 cache, below 128KB block size beating at 3GHz everything except Xeon 2800. So the cache is definitely 128KB, running with full CPU clock. Only linux kernels are not able to switch it on, because I can't believe that the CPU itself can be twice as slow as Athlon XP at the same clock frequency. (compilation times for Cel@3000 similar to AthlonXP@1533).
Slashdot Mirror
L2 is turned on in the BIOS. The kernel can use it, but it doesn't. I don't believe that Celeron /northwood chip, in fact/ can be as slow @3000, as Athlon XP at 1533 (1800+). I can believe in 20% performance drop, not 100%, sorry.
Especially when L2 size is not reported properly by kernel...
I tried to follow your suggestion of directly measuring performance drop for memory access. I run some old benchmark, but results are not easy to interpret. Performance drops at 32-64K for Cel, and 64-128K chunks for Athlon XP..... ;}
It's not so easy, as it seems, maybe you know good cache benchmark for linux ?
I don't like to be called 'lame' :( - especially when nobody posts the answer to my question. I posted the problem on comp.os.linux.hardware /or kernel, don't remember/, and got no answer. It seems that suggestion of writing to kernel mailing list is the best solution.
;) I thought the same, until I installed win2k and Sandra benchmarks which proved that L2 feels good and goes like crazy at 3000...
I made google search, of course, found question you mentioned, but I have different problem. POST on my PC doesn't show L2 cache size, (or it disappears too fast ;] - I'll check), but in Win2K everything is OK in Sandra L2 cache benchmarks. Cache works VERY fast and performance drops above 128KB blocks.
Probably you are right suggesting to directly contact people working on kernel, thx
It's not the hardware problem! - in win2k with drivers from the MB's CD, Sandra cache benchmark shows great results for L2 cache, below 128KB block size beating at 3GHz everything except Xeon 2800. So the cache is definitely 128KB, running with full CPU clock. Only linux kernels are not able to switch it on, because I can't believe that the CPU itself can be twice as slow as Athlon XP at the same clock frequency. (compilation times for Cel@3000 similar to AthlonXP@1533).