SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Politics : Formerly About Advanced Micro Devices

 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
To: Maxwell who wrote (36987)9/14/1998 2:13:00 AM
From: Petz  Read Replies (4) of 1570476
 
Maxwell, good news on K6-3 performance.

I built a mathematical model which predicts performance of the K6 core
with various speed L2 cache and memory bus. The model very accurately
predicts performance of K6-2's running at various L2 cache speeds of
112MHz, 100 MHz, 95 MHz and 66 MHz, with either 66 MHz or 100 MHz
SDRAM as measured using Winbench 98 testing by tomshardware.com.

My assumptions for the K6-3 are

1)the K6-3 core will be marginally faster than the K6-2 core (5%).
Only about 1/3 of this improvement gets translated into higher
Winbench 98 scores because memory wait, I/O and other factors are
more significant.
2)the K6-3 L2 cache will operate at the clock speed of the chip, e.g.,
400 MHz for a 400 MHz chip.
3)the K6-3 L2 cache of 256K will experience 30% more L2 cache "misses"
than the 512K cache of the typical Super Socket 7 motherboard.
4)the penalty for an L2 cache "miss" will be 20% lower than with the
K6-2, because 50% of these L2 cache misses will be satisfied by the L3
cache on the Socket 7 motherboard.

Here is the table, all benchmark scores are Winbench 98 under Windows
98. There are two prediction columns, the first one is for a run of
the mill system used by "Patrick" at tomshardware.com. The "PredSuper"
column is for a system with ultra fast components (IBM Ultrawide SCSI,
128M SDRAM etc.) used by the big CPU comparison at tomshardware.com

CPU_Type CPUClk L2Clk MemClk Pred Actual PredSuper Actual NearestIntel
K6-2-300 300 100 100 22.12 22.1 25.66 25.7 P2 300/66 = 25.4
K6-2-333 333 95 95 22.68 ---- 26.36 26.2 CeleronA 333/66 = 26.3
K6-2-350 350 100 100 23.29 ---- 27.12 27.2 P2 350/100 = 28.3
K6-2-400 400 100 100 24.25 ---- 28.34 ---- P2 350/100 = 28.3
K6-3-350 350 350 100 26.13 ---- 30.75 ---- P2 400/100 = 30.8
""-variantA 350 350 100 26.23 ---- 30.87 ---- P2 400/100 = 30.8
""-variantB 350 350 100 25.70 ---- 30.19 ---- P2 400/100 = 30.8
K6-3-400 400 400 100 27.45 ---- 32.45 ---- P2 450/100 = 31.4
K6-3-450 450 450 100 28.57 ---- 33.91 ---- P2 500/100 est32.0

K6-3 variant A assumes no improvement in core CPU speed, instead of 5%
K6-3 variant B assumes a 5% smaller L2 cache miss penalty

The bottom line, a K6-2-400 will just equal a P2-350, but a
K6-3-400 will equal a P2-500 while a K6-3-450 will definitely
outclass it.
Report TOU ViolationShare This Post
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext