View Single Post
Old 06 August 2010, 19:37   #102
Thalion Webshrine
alexh's Avatar
Join Date: Jan 2004
Location: Oxford
Posts: 12,047
Originally Posted by modrobert View Post
I was checking the licensed soft cores for Spartan 6 when estimating the clock speed where MicroBlaze was tested around 120MHz.
The 120MHz is a quote from the Xilinx website? The cynic in me wonders how many weeks of tweaking the timing constraints it took to get that? I'd be surprised if there was anything else in the FPGA not that it matters. I wonder which device it was? The $30 part? From what I remember the Microblaze is a much more simple RISC CPU compared to 68060, particularly the CISC addressing vs RISC? (Although reading the specification of v7.30 I do wonder!) I wonder what speeds you get with the Microblaze + FPU (not at all a good comparison for 060 but Xilinx don't quote speeds for the two on their site).

I guess we need to gather practical numbers rather than pure guesswork by both of us. Someone needs to target something of comparable size and complexity at those $30 devices with minimal timing constraints and see what target speeds and LUT usage we can get.

Originally Posted by modrobert View Post
So how about 50Mhz then? It would still make an interesting product.
Whilst I stand by my suggestion that an MC68060RC50 running at high speeds would out perform what you could program into today's affordable FPGAs (based on my experience with them) I have to be scientific and say the following things are true :

We have no real idea of the true performance of the real 060, so no idea what the target performance of the FPGA core should be.
We have no real idea on the true complexity of the real 060 and hence the size of the FPGA required.
We have no real idea of the prices of FPGA's small enough but fast enough for it to fit.
We have no real idea (but we can guess) what else we need in the FPGA at the same time as the CPU other at least an SDRAM. for it to be a saleable product.

Without answers to those questions you cannot really start a project.

If the NatAmi people really do have the code for a more complex 68000 CPU than TG68 (I know they say they have) then now is the time to ask them for some timing reports for synthesis & routing aimed at these $30 devices using the free tools.

Come up with some international benchmarking tools which we could use to judge the speed of the real 060 and any FPGA equivalent (would specint & specfp do? Are they good enough to test Super Scalar operation, branch prediction, caching etc. etc.)

If you made something which was more expensive and performed worse than a 50MHz 060 then I think you would not have a saleable product. I think this is what happened to the Dragon Coldfire CPU board from Elbox. When running in 680x0 compatibility mode the performance was terrible. Hence no release.

The NatAmi people are thinking long term I believe, the price/performance of the technology will come to them just in time for them to be finished

Last edited by alexh; 06 August 2010 at 19:51.
alexh is offline  
Page generated in 0.03931 seconds with 10 queries