cs161agc@sdcc10.ucsd.EDU (John Schultz) (11/24/88)
In article <13603@oberon.USC.EDU> papa@pollux.usc.edu (Marco Papa) writes: [...TMS34010 dialogue...] >Best case for BitBlt: 23.5 Million bits/sec. with VRAM (video RAM). >This is with the 34010 from 1986. I don't know if TI has upgraded the chip >with a faster clock. > >-- Marco Papa 'Doc' >-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= In 1988 we have the TMS34020, 32 bit graphics processor. Here is a short blurb from Byte, September 1988: The '20 is the newest member of Texas Instruments' (TI) TMS340 GSP Family. Depending on the instruction mix, it's between 6 and 50 times faster in key graphics operations than its predecessor, the TMS34010 (the '10). The '20 runs at 10 million instructions per second (MIPS) when executed from its 512-byte instruction cache. It's designed to connect directly to a second '20 as well as the 40-million-floating-point-operation-per-second (MFLOPS) TMS34082 graphics floating-point coprocessor (the '82 FPU). The '20 has instructions that can perform pixel- or bit-aligned block transfers at 142 megabits per second, and when using the TMS44C251 1-megabit video RAM (the 44C251 VRAM), the '20 can execute fills at up to 1.136 gigabits per second. ... [End of blurb] The article goes on to state how these chips interface and interact, etc. The 34082 FPU has _built in_: [from page 265] One-operand operations: Absolute value, 1's complement, Square root, 2's complement. Two-operand operations: Add, Divide, Subtract, Compare, Multiply. Conversions: integer to single, single to integer, integer to double, double to integer, single to double, double to single. Matrix Operations: 4x4,4x4 multiply, 3x3,3x3 multiply,1x4,4x4 1x3,3x3 multiply. Graphics Operations: Backface testing, Polygon clipping, polygon elimination, Viewport scaling and conversion, 2-D linear interpolation, 2-D window compare, 2-plane clipping (X,Y,Z), 2-D cubic spline, 3-D linear interpolation, 3-D volume compare, 2-plane color clipping (R,G,B,I), 3-D cubic spline. Image Processing: 3x3 convolution. Chained operations: Polynomial expansion, 1-D minimum/maximum, multiply/accumulate, 2-D minimum/maximum. Vector operations: Add, Dot product, Subtract, Cross product, Magnitude, Normalization, Scaling, Reflection. [...end of excerpt...] The TMS34020, TMS34082, and TMS44C251 are an awesome combination. Imagine real-time VideoScape, imagine a killer flight simulator, 3-D CAD in 3-D with real-time updates. I guess we could call this chipset the "RTCS" or "Real Time Chip Set". If someone is working with the old 34010, then a 34020 version could be right around the corner, as software written for the '10 is compatible with the '20. A board using these chips for the Amiga would definately bring the Amiga state of the art up to workstation power. Simply put, a board like this for the Amiga would be totally awesome [dude]. For more information, look up the September 1988 issue of Byte, "Taking the Wraps off the 34020", pp257-272. Hope I opened some sleepy eyes, John Schultz
srp@modcomp.UUCP (Steve Pietrowicz) (11/26/88)
in article <17@sdcc10.ucsd.EDU>, cs161agc@sdcc10.ucsd.EDU (John Schultz) says: > In article <13603@oberon.USC.EDU> papa@pollux.usc.edu (Marco Papa) writes: > [...TMS34010 dialogue...] >>Best case for BitBlt: 23.5 Million bits/sec. with VRAM (video RAM). >>This is with the 34010 from 1986. I don't know if TI has upgraded the chip >>with a faster clock. > In 1988 we have the TMS34020, 32 bit graphics processor. Here is > a short blurb from Byte, September 1988: > [stuff deleted] Does anyone have a guess at how much new board might cost? ---- Steve Pietrowicz ...!uunet!modcomp!srp
root@sbcs.sunysb.edu (root) (11/27/88)
In article <147@modcomp.UUCP>, srp@modcomp.UUCP (Steve Pietrowicz) writes: > in article <17@sdcc10.ucsd.EDU>, cs161agc@sdcc10.ucsd.EDU (John Schultz) says: > > In article <13603@oberon.USC.EDU> papa@pollux.usc.edu (Marco Papa) writes: > > [...TMS34010 dialogue...] > >>Best case for BitBlt: 23.5 Million bits/sec. with VRAM (video RAM). > >>This is with the 34010 from 1986. I don't know if TI has upgraded the chip > >>with a faster clock. > > In 1988 we have the TMS34020, 32 bit graphics processor. Here is > > a short blurb from Byte, September 1988: > > [stuff deleted] > > > Does anyone have a guess at how much new board might cost? This is actually almost funny. Has Commodore so spoiled us with followons & upgrades that we expect the followup product to be announced before its predecessor comes to market? In the spirit of the parent article of this posting, does anyone happen to know what the A3500 will cost? You know, the one with the 68040 and 16 mBytes ram. > Steve Pietrowicz > ...!uunet!modcomp!srp Rick Spanbauer SUNY/Stony Brook
srp@modcomp.UUCP (Steve Pietrowicz) (11/30/88)
in article <1856@sbcs.sunysb.edu>, root@sbcs.sunysb.edu (root) says: ] In article <147@modcomp.UUCP>, srp@modcomp.UUCP (Steve Pietrowicz) writes: ]> in article <17@sdcc10.ucsd.EDU>, cs161agc@sdcc10.ucsd.EDU (John Schultz)says: ]> > In article <13603@oberon.USC.EDU> papa@pollux.usc.edu (Marco Papa) writes: ]> > [...TMS34010 dialogue...] ]> >>Best case for BitBlt: 23.5 Million bits/sec. with VRAM (video RAM). ]> >>This is with the 34010 from 1986. I don't know if TI has upgraded the ]> >>chip with a faster clock. ]> > In 1988 we have the TMS34020, 32 bit graphics processor. Here is ]> > [stuff deleted] ]> Does anyone have a guess at how much new board might cost? ] This is actually almost funny. Has Commodore so spoiled us ] with followons & upgrades that we expect the followup product ] to be announced before its predecessor comes to market? ] In the spirit of the parent article of this posting, does ] anyone happen to know what the A3500 will cost? You know, ] the one with the 68040 and 16 mBytes ram. What's even funnier is the fact that you can't understand what I was asking. I was asking how much a the *new* board they're trying to bring to market will cost, not some yet to be dreamed up board. Certainly there are similar things on the market now for other machines, and I was asking how much something like that goes for on the market now. ] Rick Spanbauer ] SUNY/Stony Brook -------- Steve Pietrowicz ...!uunet!modcomp!srp
deraadt@dataspan.UUCP (Theo De Raadt) (11/30/88)
The bus interface is much nicer on the 34020, so I hope that we don't get a lot of C-A code which wires us in too close to the '010, cause it would indeed be real nice to be able to talk to a 34020 in the way that we are supposed to now.. Geez, I was thinking to myself.. that the 34020 might be fast enough to interpret copper lists and thus make everything 100% compatible.. <tdr. -- _____ _ ----------------------------------- / / / / \ _ _ /_/_ Theo de Raadt: (403) 289-4620 / /_ _ ___ __/_ /__/ _\ _\ __/ / DATASPAN / / /</_(_) (_/</_/ \_(_/\(_/\(_/_(_/ ..!alberta!calgary!dataspan!deraadt
thompson@savax.UUCP (thompson mark) (12/02/88)
> [...TMS34010 dialogue...] > > >Best case for BitBlt: 23.5 Million bits/sec. with VRAM (video RAM). > >This is with the 34010 from 1986. I don't know if TI has upgraded the chip > > > >-- Marco Papa 'Doc' Big problem with the 34010. It uses a packed-pixel architecture. Therefore BitBlt speed decreases with pixel depth. In a 8 bit plane amiga, thats under 3 Million pixels/sec. By todays standards, thats piss poor. Both the AMD and National chip set are pixel depth independent. The National is capable of over 60 Million pixels/sec. Although the AMD is not quite as fast, it has a reasonable price tag (~ $90 in plastic qnty 1000). > The TMS34020, TMS34082, and TMS44C251 are an awesome combination. > Imagine real-time VideoScape, imagine a killer flight simulator, 3-D > CAD in 3-D with real-time updates. I guess we could call this > chipset the "RTCS" or "Real Time Chip Set". If someone is working > with the old 34010, then a 34020 version could be right around the > corner, as software written for the '10 is compatible with the '20. > A board using these chips for the Amiga would definately bring the > Amiga state of the art up to workstation power. > > John Schultz The '20 is certainly a nice improvement over the '10 with some great new features and some respectable performance. But if by Real Time you mean 30Hz update...dream on! A fully loaded top of the line Silicon Graphics workstation stuffed to the gills with HUGE custom ASICs is not capable of this. TI has merely improved what was a DOG of a graphics part with something that can somewhat compete with the other off-the-shelf offerings. I would love to see the '20 in an Amiga 3000, but I'm sure its probably cost prohibitive. Sounds like a job for......Public Domain Engineering!!! -------------------------------------------------------------------------- | Mark Thompson | | decvax!savax!thompson Designing high performance graphics | | (603)885-9583 silicon today for a better tomorrow. | --------------------------------------------------------------------------
cs161agc@sdcc10.ucsd.EDU (John Schultz) (12/03/88)
In article <819@savax.UUCP> thompson@savax.UUCP (thompson mark) writes: >The '20 is certainly a nice improvement over the '10 with some great new >features and some respectable performance. But if by Real Time you mean >30Hz update...dream on! A fully loaded top of the line Silicon Graphics >| Mark Thompson | A 30Hz flight simulator, sure. A 30Hz ray-traced animation, uh, maybe .030Hz. I saw an AT&T board at NCGA crank out about 1 frame every 3 seconds. They called it real-time. Of course, frame rate is directly proportional to image complexity and the nature of the math involved to generate the image. We're getting 16-30 frames/sec with a standard Amiga in SpaceSpuds, just using the blitter efficiently and limiting the animation. It runs at over 45 frames/sec on an 030/020 machine, never slower than 30 f/s (only 30 frames/sec are displayed). All this in Stereo 3D. Look at all the tricks used in Starglider II. Runs very fast, with good polygon animation. Using those same techniques, a much more complex image can be generated at a higher frame rate with a powerful graphics board. Maybe not Evans and Sutherland quality... Yet. John Schultz Dreaming on...
thompson@savax.UUCP (thompson mark) (12/07/88)
In article <28@sdcc10.ucsd.EDU>, cs161agc@sdcc10.ucsd.EDU (John Schultz) writes: > In article <819@savax.UUCP> thompson@savax.UUCP (thompson mark) writes: > >The '20 is certainly a nice improvement over the '10 with some great new > >features and some respectable performance. But if by Real Time you mean > >30Hz update...dream on! A fully loaded top of the line Silicon Graphics > >| Mark Thompson | > > A 30Hz flight simulator, sure. A 30Hz ray-traced animation, uh, > maybe .030Hz. I saw an AT&T board at NCGA crank out about 1 frame > every 3 seconds. They called it real-time. Of course, frame rate I saw it at SIGGRAPH. Very nice! They use someting like 85 of their DSP parts working in parallel. > We're getting 16-30 frames/sec with a standard Amiga in > SpaceSpuds, just using the blitter efficiently and limiting the > animation. It runs at over 45 frames/sec on an 030/020 machine, > never slower than 30 f/s (only 30 frames/sec are displayed). All > this in Stereo 3D. Great job on SpaceSpuds, I love that game. When I said 30Hz was a pipe dream, I should have clarified my criteria. I am referring to doing double buffered animation from a display list by which the render buffer is cleared every 33msec and the image generation is started from scratch. A reasonable looking flight simulator would require about 2000 polygons per/frame (flat shaded). This would seem doable on 100k/s polygon machines but other overhead operations keep you from maintaining 30 Hz. SGI's simulator currently runs around 20Hz. > Look at all the tricks used in Starglider II. Runs very fast, ^^^^^^ Ahhh.. thats what I'm getting at, no tricks allowed. Seems I remember a quote about a '...suitabley rigged demo...' :-) > Maybe not Evans and Sutherland quality... > Yet. ^^^^^^^^^^^^^^^^^^^^ Hope you saw the real time texture mapped automobile simulator they did. Hot stuff. > > John Schultz > Dreaming on... Glad to hear it because dreamers make miracles possible. -------------------------------------------------------------------------- | Mark Thompson | | decvax!savax!thompson Designing high performance graphics | | (603)885-9229 silicon today for a better tomorrow. | --------------------------------------------------------------------------
king@client1.dciem.dnd.ca (Stephen King) (12/08/88)
In article <823@savax.UUCP> thompson@savax.UUCP (thompson mark) writes: >I saw it at SIGGRAPH. Very nice! They use someting like 85 of their >DSP parts working in parallel. From the context of the posting, I could not quite determine who has the DSPs working in parallel, Silicon Graphics or AT&T. Please clarify. I am aware that there is some parallel development at SGI, but know few of the available facts. Thanks. -- {utzoo|mnetor}!dciem!zorac!dretor!king or king%dretor@zorac.dciem.dnd.ca Stephen J King =-= DCIEM Human Factors Division =-= (416) 635-2149
thompson@savax.UUCP (thompson mark) (12/17/88)
In article <1262@client1.dciem.dnd.ca> king@client1.dciem.dnd.ca (Stephen King) writes: >In article <823@savax.UUCP> thompson@savax.UUCP (thompson mark) writes: >>I saw it at SIGGRAPH. Very nice! They use someting like 85 of their >>DSP parts working in parallel. > >From the context of the posting, I could not quite determine who has the >DSPs working in parallel, Silicon Graphics or AT&T. Please clarify. I am >aware that there is some parallel development at SGI, but know few of the >available facts. Thanks. >-- > Stephen J King =-= DCIEM Human Factors Division =-= (416) 635-2149 It was the AT&T pixel-machine, a massively parallel graphics system. SGI hardware relies primarily on pipelining. They have a custom floating point ASIC which they download configuration code into at boot time. They have strung a bunch of these together (4 of them actually) and each performs a separate rendering task. I'd elaborate on the two machines but I'm in a bit of a rush at the moment. Drop me email if you are interested. -------------------------------------------------------------------------- | Mark Thompson | | decvax!savax!thompson Designing high performance graphics | | (603)885-9229 silicon today for a better tomorrow. | --------------------------------------------------------------------------