Re: celeron vs k6-2
Hi It's most likely due to the current celerons having better memory bandwith than the K6-2's. The more data pr. time unit that can get through the memory system the more time will be spend by the CPU doing calculations instead of sitting idle waiting for data. This is one good reason for using an athlon with a good motherboard for serious software RAID machines. --- http://www.elof.dk -- Kristian Elof Soerensen [EMAIL PROTECTED] (+45) 45 93 92 02 On Mon, 24 Apr 2000, Seth Vidal wrote: Hi folks, I did some tests comparing a k6-2 500 vs a celeron 400 - on a raid5 system - found some interesting results Raid5 write performance of the celeron is almost 50% better than the k6-2. Is this b/c of mmx (as james manning suggested) or b/c of the FPU? I used tiobench in sizes of than 3X my memory size on both systems - memory and drives of both systems were identical. Thanks -sv
Re: celeron vs k6-2
[Seth Vidal] I did some tests comparing a k6-2 500 vs a celeron 400 - on a raid5 system - found some interesting results Raid5 write performance of the celeron is almost 50% better than the k6-2. Can you report the xor calibration results when booting them? Is this b/c of mmx or b/c of the FPU? FPU should never get involved (except the FPU registers getting used during MMX operations). As per Greg's report of the K6-2 having MMX instructions, remember that a chip having instructions doesn't mean they get used. Again, this is something that the xor calibrations should help show, though. MTRR could certainly be another source of additional performance, but I haven't dealt with the K6-2 in any capacity so I don't even know whether it has that capability (although I haven't personally heard of anything not based on the P6 core using MTRR) I used tiobench in sizes of than 3X my memory size on both systems - memory and drives of both systems were identical. If possible, let the resync's finish before testing... this can cause a huge amount of variance (that I've seen in my testing). speed-limit down to 0 doesn't appear to help, either (although the additional seeks to get back to the "data" area from the currently resyncing stripes could be the base cause) When looking from a certain realistic POV, it'd be hard to believe that even a P5 couldn't keep up with the necessary XOR operations... is there anything else on the system(s) fighting for CPU time? James
Re: celeron vs k6-2
Raid5 write performance of the celeron is almost 50% better than the k6-2. Can you report the xor calibration results when booting them? sure I should be able to pull that out of somewhere from the k6-2: raid5: MMX detected, trying high-speed MMX checksum routines pII_mmx : 1121.664 MB/sec p5_mmx: 1059.561 MB/sec 8regs : 718.185 MB/sec 32regs: 501.777 MB/sec using fastest function: pII_mmx (1121.664 MB/sec) If possible, let the resync's finish before testing... this can cause a huge amount of variance (that I've seen in my testing). speed-limit down to 0 doesn't appear to help, either (although the additional seeks to get back to the "data" area from the currently resyncing stripes could be the base cause) I did both tests just about identically. When looking from a certain realistic POV, it'd be hard to believe that even a P5 couldn't keep up with the necessary XOR operations... is there anything else on the system(s) fighting for CPU time? no. they were blanked - i didn't put them into runlevel 1 but I did shut down everything I could. they were pretty low load. -sv
Re: celeron vs k6-2
early stepping K6-2s did not have an MTRR. later steppings do (i believe stepping 8 was the first one to have an MTRR... but i can't say for certain): my cpu: processor : 0 vendor_id : AuthenticAMD cpu family : 5 model : 8 model name : AMD-K6(tm) 3D processor stepping: 0 cpu MHz : 300.689223 fdiv_bug: no hlt_bug : no sep_bug : no f00f_bug: no coma_bug: no fpu : yes fpu_exception : yes cpuid level : 1 wp : yes flags : fpu vme de pse tsc msr mce cx8 sep mmx 3dnow bogomips: 599.65 this guy's http://www.tux.org/hypermail/linux-kernel/1999week21/0052.html cpu: processor : 0 vendor_id : AuthenticAMD cpu family : 5 model : 8 model name : AMD-K6(tm) 3D processor stepping : 12 cpu MHz : 350.810582 fdiv_bug : no hlt_bug : no sep_bug : no f00f_bug : no coma_bug : no fpu : yes fpu_exception : yes cpuid level : 1 wp : yes flags : fpu vme de pse tsc msr mce cx8 sep mtrr pge mmx 3dnow bogomips : 699.60 James Manning wrote: MTRR could certainly be another source of additional performance, but I haven't dealt with the K6-2 in any capacity so I don't even know whether it has that capability (although I haven't personally heard of anything not based on the P6 core using MTRR.
Re: celeron vs k6-2
early stepping K6-2s did not have an MTRR. later steppings do (i believe stepping 8 was the first one to have an MTRR... but i can't say for certain): my cpu: processor : 0 vendor_id : AuthenticAMD cpu family : 5 model : 8 model name : AMD-K6(tm) 3D processor stepping: 0 cpu MHz : 300.689223 fdiv_bug: no hlt_bug : no sep_bug : no f00f_bug: no coma_bug: no fpu : yes fpu_exception : yes cpuid level : 1 wp : yes flags : fpu vme de pse tsc msr mce cx8 sep mmx 3dnow bogomips: 599.65 important flags from my cpu: flags : fpu vme de pse tsc msr mce cx8 sep mtrr pge mmx 3dnow interesting. mtrr is there so maybe its motherboard quality. -sv
RE: celeron vs k6-2
-Original Message- From: Seth Vidal [mailto:[EMAIL PROTECTED]] Sent: Monday, April 24, 2000 2:39 PM To: [EMAIL PROTECTED] Subject: celeron vs k6-2 Hi folks, I did some tests comparing a k6-2 500 vs a celeron 400 - on a raid5 system - found some interesting results Raid5 write performance of the celeron is almost 50% better than the k6-2. Is this b/c of mmx (as james manning suggested) or b/c of the FPU? NOT because of MMX, as the K6-2 has MMX instructions. It could be because of the parity calculations, but you'd need to do a test on a single disk to make sure that it doesn't have anything to do with the CPU/memory chipset or disk controller. Can you try with a single drive to determine where things should be? Greg
RE: celeron vs k6-2
NOT because of MMX, as the K6-2 has MMX instructions. It could be because of the parity calculations, but you'd need to do a test on a single disk to make sure that it doesn't have anything to do with the CPU/memory chipset or disk controller. Can you try with a single drive to determine where things should be? I can probably do that test tomorrow. -sv
RE: celeron vs k6-2
At 02:45 PM 24/04/00 -0700, Gregory Leblanc wrote: -Original Message- From: Seth Vidal [mailto:[EMAIL PROTECTED]] Sent: Monday, April 24, 2000 2:39 PM To: [EMAIL PROTECTED] Subject: celeron vs k6-2 Hi folks, I did some tests comparing a k6-2 500 vs a celeron 400 - on a raid5 system - found some interesting results Raid5 write performance of the celeron is almost 50% better than the k6-2. Is this b/c of mmx (as james manning suggested) or b/c of the FPU? NOT because of MMX, as the K6-2 has MMX instructions. It could be because of the parity calculations, but you'd need to do a test on a single disk to make sure that it doesn't have anything to do with the CPU/memory chipset or disk controller. Can you try with a single drive to determine where things should be? This may indeed be because of the MMX instructions in the K6-2. The MMX instructions in the K6-2 are SLOWER than the Intel counterparts. The FPU is also slower. If you were able to use proper 3DNow! instructions instead of MMX, you would notice a marked improvement in the K6-2's speed. The K6-III improved this somewhat, but the performance of MMX instructions didn't equal that of 3DNow! until the Athlon came along. You'll see this sort of difference if you run something like rc5 key cracking on your machine, as it has modes for using MMX or 3DNow! for its calculations. -- -=[ Stuart Young (Aka Cefiar) ]=--- | http://amarok.glasswings.com.au/ | [EMAIL PROTECTED] | ---