View Single Post
Old 25 March 2016, 17:36   #13
Registered User
Join Date: Jan 2010
Location: Kansas
Posts: 875
Originally Posted by eXeler0 View Post
AFAIK very few real time engines use Double Precision Float (and only very recently). Don't think a lot of ppl are going to render or do CAD on their Vampires, to double precision seems overkill tbh. For compatibility reasons may be it can be solved with some clever libs. Idk
The default fp precision in C is double precision. Using single precision fp can cause programs to fail which expect double precision. Most Programs and algorithms could be created to use single precision fp or fixed point math but it is more work. Most existing 3D software and engines use some double precision fp. Without double precision fp, these do not get ported. OpenGL ES is one of the few exceptions (used in embedded systems, smart phones, pads, etc.) which supports integers and single precision fp. Lack of at least double precision would cause almost all Amiga programs using fp to not work properly (the double precision parts could be done in software but very slowly). Even double precision fp could cause problems on the Amiga where extended precision fp is using the extended precision to simplify algorithms.

Most FPUs support double precision. The 68k and x86 FPU supported extended precision. FPUs are becoming less popular as SIMD units and GPUs are doing more of the fp work. Many SIMDs and GPUs are single precision fp only but supporting double precision fp is becoming more common. Most of the data being processed remains single precision fp.

Originally Posted by eXeler0 View Post
First time I hear they are running out of space in the FPGA, but then again, the rumors about the expansion port on the V1200 allowing bigger FPGAs maybe makes a bit more sense now :-)
(Not that I'm complaining, I wouldn't mind paying $100 more to get the more advanced FPGA right away on the 1200 version, though I'm pretty sure it's not gonna happen).
I wouldn't say they are "running out of space" in the FPGA. They want to fill the FPGA with useful logic which may make it look like they are "running out of space" before they optimize. Majsta is using a pretty good sized FPGA now. Implementing an FPU and/or SIMD would eat up space quickly though. A full extended precision 68060 compatible FPU (a good standard) would be expensive. Even a single precision fp SIMD would be very expensive. Right now, they are doing SIMD like instructions with integer only (probably to save logic). Eventually, a full modern processor design would need a bigger FPGA but that may only make sense if planning for an ASIC.

The higher end accelerators (1200 and big box Amiga) and stand alone FPGA Amiga boards probably could sell for a higher price. I don't think you are the only one who would opt for a larger faster more expensive FPGA (perhaps with SerDes for SATA), more memory (up to 1GB), USB and ethernet. The accelerator for 68000 systems is more price sensitive though.
matthey is offline  
Page generated in 0.05379 seconds with 9 queries