[ Aleph One home page | Company Information | Security Privacy and Terms of sale | Search ]

Bench tests

> PC Card Home > Support > benchmark.html

512K cache 5x86 cards announced

Last change 24 June 1997

Aleph One have done an extensive set of Winstone 96 Benchmarks using 5x86 PC cards and PCPro to determine the factors that control overall speed in typical Windows applications.

These tests were all done using the following configuration: Risc-PC host using Risc OS 3.7, Seagate ST3491 IDE HD, !PCv2.03, Sharedmemory enabled, Windows 95, 800x600x256p video mode using Alpeh One ARMDRV video driver. Click here for full disclosure

The variables were:

ARM610 vs. StrongARM,
Cyrix/IBM 5x86/100 (8K L1 cache) vs. AMD 5x86/133 (16K L1 cache),
128K vs. 512K L2 cache,
8Mb vs. 16 Mb main PC RAM.

Summary of results

By far the most important factor in speed is PC RAM. Going from 8Mb to 16Mb gives 26% - 35% improvement in all cases. Going up to 32Mb only gives another 4% in the case tested.
StrongARM is the next biggest variable, being between 13% & 18% faster than ARM610
Level 2 cache makes between 5% & 14% difference (512K faster than 128K). Surprisingly it makes slightly more difference for larger main RAM.
The difference between the Cyrix & AMD 5x86s is small (AMD 2% slower to 5% faster than Cyrix/IBM)
Upgrading everything from the slowest to the fastest gives a 77% improvement overall.

Upgrades for existing cards from 128K to 512K cache are carried out on our behalf by First SMT Ltd in Letchworth, at a price of £49, including VAT and postage.

To take advantage of this send your Card with a cheque for £49 (not credit card) to First SMT Ltd, Spirella Building, Bridge Road, LETCHWORTH SG6 4ET. You are likely to get it back in less than a week.

Results

The results are shown in the table below, both in absolute overall Winstone 96 values, and in relative values (where the baseline 5x86/128K/ARM610/8Mb is 1.0).

Overall Results

ARM610
CPU	Cyrix 5x86/100		AMD 5x86/133
RAM\Cache	128K	512K	128K	512K
8Mb	15.5 (1.00)	16.3 (1.05)	15.8 (1.02)	17.1 (1.10)
16Mb	20.5 (1.32)	22.3 (1.43)	20.4 (1.32)	23.0 (1.48)
StrongARM 200
CPU	Cyrix 5x86/100		AMD 5x86/133
RAM\Cache	128K	512K	128K	512K
8Mb	18.4 (1.18)	19.5 (1.25)	18.1 (1.17)	20.3 (1.31)
16Mb	23.1 (1.49)	26.4 (1.70)	24.1 (1.55)	27.5 (1.77)

Percentage improvements

This analysis shows the percentage improvement gained from changing the given variable for each case.

RAM 16 over 8

	Cyrix 5x86/100		AMD 5x86/133
	128K	512K	128K	512K
ARM610	32%	36%	29%	35%
StrongARM	26%	35%	33%	35%

StrongARM over ARM610

	Cyrix 5x86/100		AMD 5x86/133
	128K	512K	128K	512K
8Mb RAM	18%	20%	15%	19%
16Mb RAM	13%	18%	18%	20%

512K cache over 128K cache

	Cyrix 5x86/100	AMD 5x86/133
ARM610, 8Mb	5%	8%
ARM610, 16Mb	8%	12%
StrongARM, 8Mb	6%	12%
StrongARM, 16Mb	14%	14%

AMD 5x86/133 over Cyrix 5x86/100

	128K	512K
ARM610, 8Mb	2%	5%
ARM610, 16Mb	0%	3%
StrongARM, 8Mb	-2%	4%
StrongARM, 16Mb	4%	4%

And for those of you without table support:

                                      ARM610
--------------------------------------------------------------------
              |     Cyrix 5x86/100       |     AMD 5x86/133
--------------------------------------------------------------------
RAM   L2 cache     128K         512K     |    128K         512K
--------------------------------------------------------------------
 8Mb          | 15.5 (1.00)  16.3 (1.05) | 15.8 (1.02)  17.1 (1.10)
16Mb          | 20.5 (1.32)  22.3 (1.43) | 20.4 (1.32)  23.0 (1.48)

                                StrongARM 200Mhz
--------------------------------------------------------------------
              |     Cyrix 5x86/100       |     AMD 5x86/133
--------------------------------------------------------------------
RAM   L2 cache     128K         512K     |    128K         512K
--------------------------------------------------------------------
 8Mb          | 18.4 (1.18)  19.5 (1.25) | 18.1 (1.17)  20.3 (1.31)
16Mb          | 23.1 (1.49)  26.4 (1.70) | 24.1 (1.55)  27.5 (1.77)