bathkey@netsys.UUCP (11/24/87)
OK folks, I just got a 68020, with a 68881 (20 Mhz) for my B2000
REV 3.9 . Its the CSA Turbo Board, and I must say FASTMAN is VERY FAST! But
thats the ONLY thing that was fast. I expected perhaps dos to speed up a bit
maybe even dare I say math functions should go much faster even if the
software doesnt directly address the 68020. Well since I only have the
very first version of Sculpt 3-d and its a slow dog, I decided to run that
to benchmark.. On a picture that normaly takes 12 hours to make with a 68000
it took sculpt 11.5 hours with the 68020 and 20 Mhz 68881. Mow I must admit
I tried using MANX 3.4b and the +2 and +f8 options on several c programs and
on one, I noticed instead of taking Perry's balls.c demo 80 seconds to creat
and run the balls, it took 30 seconds.. kind-of impresive.. but I dont
believe this HYPE about 60 Hour Scuplts taking 45 minutes .. Also if only
a handfull of Amiga applications have to address the 68020 and 68881 directly
what good is a 68020 in my amiga? If anyone has any Serious speed benchmarks
and would not mind sending (email) them to me, please do so. Id love to see
the difference
Scott Bethke, BATHKEY AT NETSYS.UUCP
{decvac, esi, ihnp4, killer}!netsys!bathkey
BBS Number 301-445-3777 Voice Number 703-866-9122grr@cbmvax.UUCP (George Robbins) (11/25/87)
In article <1718@netsys.UUCP> bathkey@netsys.UUCP (Scott Bethke) writes: > > OK folks, I just got a 68020, with a 68881 (20 Mhz) for my B2000 > REV 3.9 . Its the CSA Turbo Board, and I must say FASTMAN is VERY FAST! But > thats the ONLY thing that was fast. I expected perhaps dos to speed up a bit > maybe even dare I say math functions should go much faster even if the > software doesnt directly address the 68020. Well since I only have the > very first version of Sculpt 3-d and its a slow dog, I decided to run that > to benchmark.. On a picture that normaly takes 12 hours to make with a 68000 > it took sculpt 11.5 hours with the 68020 and 20 Mhz 68881. Did you invest in any of their "fast 32 bit memory"? A 68020 executing out of 16-bit memory, or worse 16-bit chip memory isn't going to blow away the world or anything... -- George Robbins - now working for, uucp: {uunet|ihnp4|rutgers}!cbmvax!grr but no way officially representing arpa: cbmvax!grr@uunet.uu.net Commodore, Engineering Department fone: 215-431-9255 (only by moonlite)
daveh@cbmvax.UUCP (11/25/87)
in article <1718@netsys.UUCP>, bathkey@netsys.UUCP (Scott Bethke) says: > Keywords: 68020 amiga B2000 68881 32 bit speed > > .. but I dont > believe this HYPE about 60 Hour Scuplts taking 45 minutes .. Also if only > a handfull of Amiga applications have to address the 68020 and 68881 directly > what good is a 68020 in my amiga? If anyone has any Serious speed benchmarks > and would not mind sending (email) them to me, please do so. Id love to see > the difference I'm not sure if Sculpts really go THAT fast, but they do go quite a bit faster. On the Commodore-Amiga 68020 board that I'm currently working on, we get maybe a 25x increase with Sculpt. But that's a Sculpt compiled for 68020 and coded with in-line 68881 instructions. Part of the lack of speedup on generic things is the state of the current math libraries. The best thing to do for better speedups in the floating point department is (A) hope that your turbo board supplier has a set of libraries written for the 68881. This will make everything that uses the libraries much faster. The next release of the Amiga OS will include math libraries that support the 68881 very nicely, but for now it's up to the third parties. The other part of the picture is CPU bound stuff. While the 68881 can boost floating point performance from 25x-100x depending on what you're doing, the 68020 itself out of fast, 32 bit RAM is more like a 4 times speed increase. That, of course, assumes that you have this RAM where the OS can get at it. How does CSA configure their 32 bit RAM? If it's not at the head of the FAST memory list, you may be getting other memory first. If it's autoconfigured, you definitely want to run SlowMemLast on a B2000. If not, that may or may not put the 32 bit RAM up front, depending on their AddMem command. Should be really noticable if it's wrong vs. right. And finally, you're not going to see much of a speedup on anything that's display bound, as you probably expected. We're supposed to have a mandelbrot program around that shows off the '881; if I get that, I'll send it along. Just a demo, but should at least give you an idea of the speed increase. > Scott Bethke, BATHKEY AT NETSYS.UUCP > > {decvac, esi, ihnp4, killer}!netsys!bathkey > BBS Number 301-445-3777 Voice Number 703-866-9122 Dave Haynie Commodore-Amiga Usenet: {ihnp4|caip|rutgers}!cbmvax!daveh "The B2000 Guy" PLINK : D-DAVE H BIX : hazy "I can't relax, 'cause I'm a Boinger!" -- Dave Haynie Commodore-Amiga Usenet: {ihnp4|caip|rutgers}!cbmvax!daveh "The B2000 Guy" PLINK : D-DAVE H BIX : hazy "I can't relax, 'cause I'm a Boinger!"
harald@ccicpg.UUCP ( Harald Milne) (11/25/87)
In article <1718@netsys.UUCP>, bathkey@netsys.UUCP (Scott Bethke) writes: > OK folks, I just got a 68020, with a 68881 (20 Mhz) for my B2000 > REV 3.9 . Its the CSA Turbo Board, and I must say FASTMAN is VERY FAST! But > thats the ONLY thing that was fast. I expected perhaps dos to speed up a bit > maybe even dare I say math functions should go much faster even if the > software doesnt directly address the 68020. I think you mean the 68881. The 68020 is needed to interface cleanly to the 68881. The 68020 will pass the operation to the 68881, if present. But, and a big but at that, the 68881 will only give you a speedup if you use 68881 instructions (an extension of 68020 instructions). I have asked a few questions about CSA boards a few months ago, but got no response. I guess nobody had any. Now that I have got you on the horn so to speak, did you perchance get a mathieedoubbas.library to replace your stock library? Any program that uses this library (excluding amiga basic) should notice a drastic speedup. For example Videoscape3d. > Well since I only have the > very first version of Sculpt 3-d and its a slow dog, I decided to run that > to benchmark.. On a picture that normaly takes 12 hours to make with a 68000 > it took sculpt 11.5 hours with the 68020 and 20 Mhz 68881. I have this version also, and I am sorry to say that it does not support 68881 instructions. The speedup you are seeing is the 68020 only. But according to an update letter and disk I recieved for Sculpt3d, you can get a version that does use the 68881 for $30.00. Lucky you! > Mow I must admit > I tried using MANX 3.4b and the +2 and +f8 options on several c programs and > on one, I noticed instead of taking Perry's balls.c demo 80 seconds to creat > and run the balls, it took 30 seconds.. kind-of impresive.. but I dont > believe this HYPE about 60 Hour Scuplts taking 45 minutes .. Well that does sort of sound like hype, but not by much. From what I figure, a 25 Mhz 68881 should take a 15 hour rendering to 30 minutes. Thats a rough guess. But it should be VERY noticeable! > Also if only > a handfull of Amiga applications have to address the 68020 and 68881 directly > what good is a 68020 in my amiga? If anyone has any Serious speed benchmarks > and would not mind sending (email) them to me, please do so. Id love to see > the difference > If there is anyway you can get a hold of Amazing Computing Volume 2, Number 7, this should answer a lot of your questions. Lots of benchmarks. > Scott Bethke, BATHKEY AT NETSYS.UUCP -- Work: Computer Consoles Inc. (CCI), Advanced Development Group (ADG) Irvine, CA (RISCy business! Home of the CCI POWER 6/32) UUCP: uunet!ccicpg!harald
andy@cbmvax.UUCP (Andy Finkel) (11/25/87)
In article <2858@cbmvax.UUCP> daveh@cbmvax.UUCP (Dave Haynie) writes: >That, of course, assumes that you have this RAM where the OS can get at it. >How does CSA configure their 32 bit RAM? If it's not at the head of the >FAST memory list, you may be getting other memory first. If it's >autoconfigured, you definitely want to run SlowMemLast on a B2000. If not, >that may or may not put the 32 bit RAM up front, depending on their AddMem >command. Should be really noticable if it's wrong vs. right. Warning...I think the CSA memory is out of the normal autoconfig address range. This means that DMA devices (like our hard disk) can't see it. If you use such a combination, don't addmem the 32 bit memory until after the driver has been bound (ie after binddrivers) and accessed, so all the buffers are allocated where the disk controller can get to them. (this is one of the reasons that our 68020 board puts its 32 bit ram in autoconfig space) -- andy finkel {ihnp4|seismo|allegra}!cbmvax!andy Commodore-Amiga, Inc. "Any sufficiently advanced technology is indistinguishable from a rigged demo." Any expressed opinions are mine; but feel free to share. I disclaim all responsibilities, all shapes, all sizes, all colors.
perry@well.UUCP (Perry S. Kivolowitz) (12/02/87)
In article <2856@cbmvax.UUCP> grr@cbmvax.UUCP (George Robbins) writes: >In article <1718@netsys.UUCP> bathkey@netsys.UUCP (Scott Bethke) writes: >> >> OK folks, I just got a 68020, with a 68881 (20 Mhz) for my B2000 >> (stuff deleted) >> it took sculpt 11.5 hours with the 68020 and 20 Mhz 68881. > >Did you invest in any of their "fast 32 bit memory"? A 68020 executing out >of 16-bit memory, or worse 16-bit chip memory isn't going to blow away the >world or anything... I might add that the static memory boards from CSA do infact greatly inhance the speed of their co-processor boards. The draw back is of course their price. These are their latest prices as I understand them: 1/2 meg board $999 2 meg board $3999 May I suggest that a good way to go is to get one CSA 32 bit memory board and then supplement your ram with large amounts of 16 bit ram. For example, get a half meg from CSA for $999 and an 8MI 8 megabyte board from ASDG for $2599. The ASDG Recoverable Ram Disk will use 16 bit memory first by the way giving you the best of both worlds. Another idea is to get the CBM 68020 board when available. That board can be populated with 2 megabytes of 32 ram on card appearing in the standard autoconfig space (correct me if I am wrong George). With 2 meg gone from the total of 8 allowable by the autoconfig standard an 8MI with 6 megabytes installed will autoconfig the remaining 6 meg. BTW: autoconfiguring a six meg memory board is an ASDG feature not part of the autoconfig standard. Thanks, Perry Kivolowitz ASDG Incorporated
liu1@puff.wisc.edu (Clif Liu) (12/05/87)
Is the ASDG 20X board available? Is anyone planning a UNIX implementation on it? I'm looking for a UNIX workstation based on the Amiga. Thanks in advance... -CLiu