From: Alec Cawley Newsgroups: comp.sys.transputer Subject: Re: StrongArm as replacement for T805 - 25Mhz Date: Thu, 26 Nov 1998 18:09:24 +0000 Organization: Lost in the Wilderness Message-Id: References: <01be1919$f3cfa0e0$0600000a@peter.ross.IMSDOM> Mime-Version: 1.0 In article <01be1919$f3cfa0e0$0600000a@peter.ross.IMSDOM>, Peter Ross writes >I am currently looking for a relatively high power processor to replace a >25Mhz T805. For several reasons the StrongARM processor looks like an >attractive proposition. Since there have been previous discussions on this >newsgroup regarding replacing transputers, I thought I would pose some >queries. > >One area of concern however is the floating point performance offered >through software emulation on theStrongARM. I have seen a figure somewhere >(can't remember where unfortunately) that performance achievable through >emulation for the StrongARM will be 50-100 times worse than using a >dedicated fpu. Now, even allowing for the much higher clock speed of the >StrongARM over the transputer, on a niave level it would appear that I will >not be able to achieve the floating point performance of the T805. > >Can anyone offer any comment? 50-100 times seems a little pessimistic to me, but undoubtedly FP performance will be poor. The system I am using on the StrongArm (GNU toolkit) has (I believe) machine independent FP support written in C, which may be where such figures came from. (I don't need FP, so I haven't checked). A hand-coded assembler FP implementation would probably be 2-3 times faster. Also, in the pipeline is a new StrongArm CPU with a DSP-style co- processor on the same chip. The co-processor has an FPU which can be "borrowed" by the main CPU; whether this is interleaved with or exclusive with the use as a DSP, I don't know. The new CPU (SA1500) is still under NDA, as it has been for some time. I haven't heard anything further since the Intel takeover was implemented. -- Alec Cawley