Risc PC x86 Card Performance and Benchmarks


Improving Performance - Overview

Different software will benefit from different improvements in your system.

Memory (RAM)

Windows benefits from additional RAM on virtually any PC system; 8MB allocated to the PC card is a good amount, but 16MB or even 32MB can improve speed still further. Most games are happy with 4MB or 8MB allocated to the PC card, but some now benefit from 16MB or more. With Windows 95 games, it is likely that the trend of requiring more and more memory will continue.

Hard disks

A faster hard disk can be very beneficial; the Conner hard disk drives shipped in standard Risc PC systems have access times of around 14ms and sustained transfer rates of between 1MB/second and 2MB/second, whereas more expensive drives can have access times of 8ms or less and sustained transfer rates of up to 8MB/second (with a suitable interface, such as RapIDE or SCSI-2). These faster speeds can make a lot of difference when using Windows applications due to their heavy usage of virtual memory and massively inefficient code.

CD-ROM

A faster CD-ROM drive can also improve things considerably in some circumstances; many PC games run from CD, and will therefore perform better with a faster drive; in addition many PC games now require a quad speed drive as a minimum for such things as video sequences.

Processors

Many people will seek to improve the speed of their PC applications by upgrading their PC card; however for some games, such as Doom, this will have a relatively limited effect, and using even a slightly faster ARM processor will make a larger difference - a StrongARM will give huge speed increases on video-intensive games like Doom. On the other hand, for some games, a higher specification PC card is almost essential; and of course for software that requires floating point calculation the original 486SX Acorn card will be much slower than the later 486DX and 5x86 cards.

Windows or Windows 95?

Windows 95 will run with only a 486SX processor, and it is usable, but it won't be terribly fast. Similarly, you should be able to get Windows 95 to run if you only allocate 8MB RAM to the PC card, but if you intend to stick with that much memory you would be better off using Windows 3.11

Summary

For the best performance you should aim to have both a fast ARM processor and a fast PC card; for those who can't afford both, the games compatibility list is now just beginning to carry details of the extent to which some games benefit from faster PC cards. Serious games players will probably want to purchase a StrongARM upgrade. Of course, more recent versions of the PC card software are faster for some things than the earlier versions; PC Pro is much faster for Windows and Windows 95, as well as for CD-ROM and hard disk access generally. You should also ensure you have at least 1MB VRAM in your Risc PC, as a system without VRAM will be very slow both for RISC OS and PC applications.

Additional Speedups


x86 Cards Comparative Windows Performance

David Coronel, at Acorn dealer The Data Store in Bromley (e-mail [email protected]), released the following performance information for some of the older x86 cards, so that people could roughly compare the differences in speed for Windows.

There seem to be some people who want some form of comparative benchmark timings for the various PC Cards now available. Whilst I would not wish to imply that the following details are definitive, they should at least give an idea of how the various cards compare to one another when running Windows.

These ratings were created using the Windows User benchmark program. They were run on a 34MB Risc PC with 16MB allocated to the PC card. Both PC partitions (drive C @ 200MB, drive D @ 400MB) exist on SCSI drives attached to the computer via an Acorn MkIII SCSI card. Windows 95 is being booted. The figures are averages of five runs through the program.

Benchmark        Acorn     Aleph1     Acorn      Aleph1
Description      SX/33     DX2/80     DX4/100    5x86/100
----------------------------------------------------------
Processor         4.6       15.5       21.5       36.5
Memory            4.1        8.2        9.3       12.2
Graphics          4.3        6.2        6.4        6.5
Disk              5.9       12.4       13.0       18.3
Windows Perf.     1.7        3.6        4.0        5.9
Overall           3.8        8.1        9.2       12.5
If one recalibrates these figures as percentage improvements over the lower rated cards, we get:
From               To DX2/80  To DX4/100  To 5x86/100
SX/33              113%       142%        229%
DX2/80               -         14%         54%
DX4/100              -          -          36%
(used with permission)

Note that Acorn and Aleph One now both supply 100MHz 5x86 cards (the Aleph One version with greatly increased secondary cache), while Aleph One no longer produce their 486DX2/80 card; see card availability details.


PC Pro versus !PC v1.92 (Winstone benchmark)

This from Aleph One's original PC Pro announcement:
We tabulate the Ziff-Davis Winstone(r) 96 benchmark, version 1, for the most recent general release of software, version 1.92l, and for PCPro. The significant variables from the Risc PC point of view are:
------------------------------------------------------------
              |   Windows 3.11        |     Windows 95
------------------------------------------------------------
Card    Speed | 1.92    PCPro   %up   |   1.92    PCPro  %up
Type    MHz   |                       |
------------------------------------------------------------
DX2     66    | 9.1     11.2    23    |
DX4     100   | 9.6     12.3    28    |   7.7     11.4    48
5x86    100   | 10.4    14.2    37    |   8.3     12.8    54
------------------------------------------------------------
Further details are available on request from Aleph One Limited, by fax or e-mail.

Wintune 95 highlights for 5x86/100 with 128KB cache

Dave Cooper wrote in comp.sys.acorn.extra-cpu:
Here are the results I got with WinTune '95 on my system. I have a Risc PC upgraded to StrongArm. A 586 PC Card. A 2.1 Gig. SCSI harddrive (Castle Tech SCSI card) using a 990Mb PC partition. I have 32+2V Ram and use 24MB for the PC Card. The display for the card is set at 800 by 600 in 256 colours (Palette). Suggested screen memory is 585Kb but I allocate 1510Kb which leaves enough (472Kb needed) for the Risc OS side to run 800 by 600 in 256 colours as well.

Wintune '95 Highlights are:

CPU type                CyrixDX
Clock rate              84 MHz
Dhrystone               109 MIPS
Whetstone               36 MFLOPS
Resolution              800 by 600
Colo(u)r depth          8 bpp
Video speed             2.7 MP/s
C:\Cached speed         5.6 MB/s
C:\Uncached speed       1 MB/s
Installed RAM           24 MB
RAM Read avg.           50 MB/s
RAM Write avg.          11 MB/s
RAM Copy avg.           7.8 MB/s

WinTach benchmarks for 5x86/100 with 512K and 128K cache

From the PC Card beta-testers mailing list:
Readout from my card (an Acorn 586/100 upgraded to 512K cache) using WinTach shows (in different screen modes):
       512-C16     128-C16      128-C8       128-B32     512-B32     512-C8
 WP      19.45       15.25       11.92        16.75        22.25      13.56
 CAD    104.85       60.10       44.28        80.50       117.25      54.73
 Spread  57.85       50.38       34.02        45.75        51.75      37.58
 Paint   28.77       26.41       19.09        32.00        34.12      20.32
 Total   52.73       38.04       27.37        43.75        56.34      31.55
Where 512 and 128 indicate the cache selected in the "Advanced Configure" setting. C8 = 256 colours, C16 = 32 thousand colours and B32 = 16 Million colours.

Screen mode is 1024x768 for C8 and C16 and 800x600 for B32.

The PC card is running with 32MB DRAM and using the maximum required (up to 2MB) VRAM, plus a StrongARM upgrade.


5x86 Cards with 512KB cache

Figures from CJE indicate that on a StrongARM Risc PC running PC Pro and with 24MB allocated to the PC card, the 512KB-cache card ran the "Windows redraw" section of the Windows 95 and Windows User benchmark test 30% faster than a 100MHz 5x86 card and 26% faster than a 133MHz 5x86 card, both of the latter being old 128KB-cache cards. We anticipate that the speedup will not be this great for the majority of applications, but these figures do make the cards look very attractive in terms of performance.

Other Benchmark Results on the WWW


5x86-100 Card Read and Write Speeds

Despite giving an excellent improvement in overall performance over the lower specification cards, the 5x86 card apparently has rather poor speeds for writing to RAM. When using !PCx86 v1.995l, Wintune 95 reports 64MB/s average speed for RAM reads, but only 12MB/s for RAM writes, unusually low for this class of system. This is probably caused by the PC card having to share the relatively slow system bus with the ARM processor, a problem shared by all the PC cards. However this does not appear to limit the performance of the 5x86 card unacceptably; for some applications, such as compiling large programs, the 5x86 card can be up to eight times faster than the original 33MHz Acorn 486SX card.

The Verdict of Rebel Assault II

The test program in the demo version of the LucasArts game Rebel Assault II gave the following results on three Risc PC setups we tested (using plenty of RAM and the Connor hard disk drives fitted as standard in Risc PC systems).
--- Actual Hardware ----   Processor   CD KB/sec at   ---- Video -----
ARM  PC card    CD-ROM    reported as  40/50/60% CPU  KB/sec  frames/s
======================================================================
610  5x86/100  Indigo 2x  198MHz 486    186/208/232     320      5
710  5x86/133  Gldstr 6x  133MHz 486    278/291/348     384      6
SA   5x86/100  Toshib 8x  196MHz 486    334/349/394    1152     18
These results showed the PC cards in a relatively good light, particularly in terms of processor performance. It is not entirely unreasonable for the card to be detected as a 198MHz 486, since the 5x86 chips, like the Pentium P24T, have a 32-bit external data bus to allow connection to 32-bit systems, but deliver substantially greater processor performance for both integer and floating point operations than any real 486 processor. Similarly it is likely that the test program is detecting the actual clock speed of the 5x86/133 card, rather than guessing its clock speed from its actual performance, as it appears to have done with the 5x86/100 card.

The CD-ROM access speeds were also acceptable, with even the slower system just about meeting MPCII minimum standards, although a quad speed or faster CD-ROM drive is useful for many recent games.

The video speeds show the importance of having a StrongARM for DOS games. The game was still playable with ARM610 and ARM710, but a more than threefold speed increase is rather impressive, and does make a substantial difference to the quality of the game and the video clips (the video clips were acceptable in SVGA modes with a StrongARM and PC Pro).


Back to RPC x86 Info Pages