I got a system with Athlon "Thunderbird" 1200C on Asus A7V133-C last week and, after some painstaking troubleshooting (turned out the SDRAM was faulty), I started testing the system with CPU-Z Vintage Edition and TOPBENCH. The BIOS has the option of turning L1 and L2 caches on and off; turning off L1 cache would drop the speed significantly to be equivalent to 486SX-25
However, unlike my Pentium-MMX build, turning L2 cache on or off on this T-bird build has no effect, with the difference of benchmark scores less than 1%, either with or without L1 cache.
Could it be
1. Norma: turning L2 off has little effect on T-bird builds
2. Normal: turning L2 off does slow down the system but CPU-Z and TOPBENCH simply can't reflect this; try another benchmark software
3. Abnormal: there's something wrong with CPU and/or MB
I know that companies like PCChips had the illegitimate practice of making fake L2 chips on their 486-era MB but I don't think Asus would do so in the era of P3/T-bird.
I got a system with Athlon "Thunderbird" 1200C on Asus A7V133-C last week and, after some painstaking troubleshooting (turned out the SDRAM was faulty), I started testing the system with CPU-Z Vintage Edition and TOPBENCH. The BIOS has the option of turning L1 and L2 caches on and off; turning off L1 cache would drop the speed significantly to be equivalent to 486SX-25
However, unlike my Pentium-MMX build, turning L2 cache on or off on this T-bird build has no effect, with the difference of benchmark scores less than 1%, either with or without L1 cache.
Could it be
1. Norma: turning L2 off has little effect on T-bird builds
2. Normal: turning L2 off does slow down the system but CPU-Z and TOPBENCH simply can't reflect this; try another benchmark software
3. Abnormal: there's something wrong with CPU and/or MB
I know that companies like PCChips had the illegitimate practice of making fake L2 chips on their 486-era MB but I don't think Asus would do so in the era of P3/T-bird.
You can check if the problem is 2 or 3 by using synthetic memory benchmarks that use different data block sizes.
The Tbird has rather big L1 cache (64K+64K) and not so big L2 cache (256K). So you should look at the results at 128K and 256K. If L2 cache is enabled then 128K-256K results should be slower than the L1 results (up to 64K) and faster than main memory results (bigger than 256K). If L2 cache is disabled then 128K and 256K results should be roughly the same as main memory results.
You can use e.g. cachemem under DOS or Sisoft Sandra under Windows to test this.
It's probably not turning off the L2 at all, because you would notice a difference.
I've had at least one system in the past outright ignore the L2 cache toggle in the BIOS. It was all or nothing with that particular system (turning off L1 in the BIOS would disable L2 as well). With both caches disabled, you could actually see the XP welcome screen fade in, using around 12 different shades of blue. Normally, you don't even see that fade-in effect because it happens so quickly. Who knew XP had speed-sensitive visual effects?
94 MHz NEC VR4300 | SGI Reality CoPro | 8MB RDRAM | Each game gets its own SSD - nooice!
You can check if the problem is 2 or 3 by using synthetic memory benchmarks that use different data block sizes.
The Tbird has rather big L1 cache (64K+64K) and not so big L2 cache (256K). So you should look at the results at 128K and 256K. If L2 cache is enabled then 128K-256K results should be slower than the L1 results (up to 64K) and faster than main memory results (bigger than 256K). If L2 cache is disabled then 128K and 256K results should be roughly the same as main memory results.
You can use e.g. cachemem under DOS or Sisoft Sandra under Windows to test this.
Thanks a lot! Turned out the L2 could be disabled with L1 alone enabled, but L2 always got disabled when L1 was disabled, just like Standard Def Steve said:
I've had at least one system in the past outright ignore the L2 cache toggle in the BIOS. It was all or nothing with that particular system (turning off L1 in the BIOS would disable L2 as well).
I ran cachemem twice just to make sure I had either enabled or disabled L2 back and forth, with L1 disabled.
With L1 and L2 both enabled (I'll just list Latency tests):
With L1 enabled and L2 disabled, the difference could only be visible when data block sizes fell within 128KB or 256KB.
Just as I said so. Compared to contemporary P4 the L2 cache was not so important to Athlon CPU's performance. That's why even the Duron with its anemic 64K L2 cache was more than competitive with P4 Celerons. That 64K+64K L1 cache is huge even compared to today's processor standards.
The biggest difference compared to your P1 MMX setup is that the Athlon Tbird has on-die L2 cache so the motherboard has nothing to do with it. Also that's why disabling L1 cache also disabled the whole cache hierarchy.
Cachemem has no specific cache level detection. It only infers the levels based on the measured bandwidth. You have not posted the bandwidth results but I assume that the relative difference between L2 bandwidth and main memory bandwidth was so small on your P1 MMX system that cachemem must have thought it is within margin of error.
Cachemem has no specific cache level detection. It only infers the levels based on the measured bandwidth. You have not posted the bandwidth results but I assume that the relative difference between L2 bandwidth and main memory bandwidth was so small on your P1 MMX system that cachemem must have thought it is within margin of error.
Indeed, the difference was visible (-33% read bandwidth) yet much less than other read bandwidth differences (3-4 times).
1Cache size/Memory speed info tool 2.65MMX - (c) 1999-2001, LRMS - DJGPP compiled 2CPUID support detected... 'GenuineIntel' with FPU TSC MMX 3Family=5 Model=4 Step=3 Type=0 Chipset (Vendor/Device ID(Rev)): Intel/7100(01) 4CPU clock: 233.9 MHz 5Using 16MB physical memory block (alignment = 32) 6Bandwidth - MMX linear access test... Read/Write/Copy (MB/s) 7Block of 1KB: 1234.5 / 175.0 / 337.9 8Block of 2KB: 1265.3 / 172.0 / 340.0 9Block of 4KB: 1281.3 / 170.5 / 340.0 10Block of 8KB: 1289.4 / 169.9 / 339.2 11Block of 16KB: 1230.3 / 169.6 / 332.6 12Block of 32KB: 239.8 / 169.4 / 226.4 13Block of 64KB: 239.8 / 169.3 / 226.3 14Block of 128KB: 239.9 / 169.3 / 224.0 15Block of 256KB: 239.9 / 169.3 / 226.4 16Block of 512KB: 239.9 / 169.2 / 226.4 17Block of 1024KB: 160.0 / 169.2 / 124.7 18Block of 2048KB: 160.0 / 169.2 / 124.6 19Block of 4096KB: 160.0 / 169.2 / 124.6 20Block of 8192KB: 160.0 / 169.2 / 124.6 21Block of 16384KB: 160.0 / 169.4