🚀 go-pugleaf

Paul DeMone wrote:
>aa158@valleynet.on.ca wrote:
>>
>> Two current news items at EETimes:
>>
>> "The Playstation 2's CPU, jointly developed by Toshiba
>> and SCE, is an enhanced version of the device described
>> at ISSCC. The device has floating-point performance of
>> 6.2 Gflops..."
>> http://www.eetimes.com/story/OEG19990302S0026
>
>  It would be interesting to know if this peak performance
>  is obtained with IEEE compliant FP or shortcuts
>  acceptable for 3D graphics.

The specs include 10 FMAC units, so that's where
6 of the GFLOPS come from. I'd expect that they're
sticking to IEEE compliance there.

They might be cutting corners on (some of) the
dividers, though. The FPU is probably compliant,
but the vector units are for MPEG/3d use.

>> "Intel said Merced would achieve 3-Gflops extended-
>> precision floating-point performance and 6-Gflops
>> single-precision performance."
>> http://www.eetimes.com/story/OEG19990226S0012
>
>  This figure is interesting if only to suggest a clock rate
>  target of about 750 to 800 MHz.
>
>  Both of these figures are peak performance values
>  which essentially means you cannot achieve higher
>  performance.  For real applications you typically
>  achieve a lot less.  If the Toshiba chip sustains 20%
>  and the Merced sustains 50% of these peak values
>  on real customer applications it will be impressive.

The Toshiba/SCEI chip should be able to do far better
than 20%, at least in the PSX2. The graphics chip that's
going in there is (theoretically) capable of rendering 75m
polys/s, so it's likely that the CPU will be doing a lot of
geometry - pretty much an ideal application for SIMD
FP.

---
Russ

aa158@valleynet.on.ca writes:

> 
> Two current news items at EETimes:
> 
> "The Playstation 2's CPU, jointly developed by Toshiba and SCE, is an enhanced
> version of the device described at ISSCC. The device has floating-point
> performance of 6.2 Gflops..."
> http://www.eetimes.com/story/OEG19990302S0026
> 
> 
> "Intel said Merced would achieve 3-Gflops extended-precision floating-point
> performance and 6-Gflops single-precision performance."
> http://www.eetimes.com/story/OEG19990226S0012
> 
> 

And it will probably just about as usefull in either of them.  Quoting
Gflops in a vacum is about the most useless thing that I have ever
heard of.  

Aaron Spink
not speaking for Compaq
         

aa158@valleynet.on.ca wrote:
> 
> Two current news items at EETimes:
> 
> "The Playstation 2's CPU, jointly developed by Toshiba and SCE, is an enhanced
> version of the device described at ISSCC. The device has floating-point
> performance of 6.2 Gflops..."
> http://www.eetimes.com/story/OEG19990302S0026

  It would be interesting to know if this peak performance is obtained
  with IEEE compliant FP or shortcuts acceptable for 3D graphics.

> 
> "Intel said Merced would achieve 3-Gflops extended-precision floating-point
> performance and 6-Gflops single-precision performance."
> http://www.eetimes.com/story/OEG19990226S0012

  This figure is interesting if only to suggest a clock rate target
  of about 750 to 800 MHz.

  Both of these figures are peak performance values which essentially
  means you cannot achieve higher performance.  For real applications
  you typically achieve a lot less.  If the Toshiba chip sustains 20%
  and the Merced sustains 50% of these peak values on real customer
  applications it will be impressive.

All opinions strictly my own.
--
Paul W. DeMone       The 801 experiment SPARCed an ARMs race of EPIC
Kanata, Ontario      proportions to put more PRECISION and POWER into
demone@mosaid.com    architectures with MIPSed results but ALPHA's well
pdemone@igs.net      that ends well.

In article <rnbn21tmlsg.fsf@nortelnetworks.com>, Chris Ebenezer
<chriseb@nortelnetworks.com> writes
>
>aa158@valleynet.on.ca writes:
>: Two current news items at EETimes:
>: 
>: "The Playstation 2's CPU, jointly developed by Toshiba and SCE, is an 
>: enhanced
>: version of the device described at ISSCC. The device has floating-point
>: performance of 6.2 Gflops..."
>: http://www.eetimes.com/story/OEG19990302S0026
>
>I notice they say that "The core of the I/O processor is the CPU of the
>current Playstation", so i guess its a R3000 based core ? 
>
>Still, what do they mean by the term I/O processor, in this context ?
>Surely, it means that this processor also has to be used as the main
>processor for using older games ?
They'll almost certainly emulate the PSX1 in software (if a 133Mhz PC
can do it, then why not this behemoth).  They won't make the hardware
backwards compatible, esp. if this thing can churn out 75 million
triangles, because of the PSX design.  

There is no way the PSX1 architechure could handle that, so it has to be
in software, with the GPU commands remapped to the new hardware.  It'll
certainly fix the perspective warping problem that has plagued so many
PSX games.

The I/O processor probably handles the firewire interface and DVD drive.
Make it a R3000 because they are cheap, and Sony probably has a lot of
'em lying around in stock.

---
In the beginner's mind there are many possibilities. In the expert's mind, there
are few.  -Shunryu Suzuki-

Joat Mon, joatmon@auia.demon.co.uk

Chris Ebenezer <chriseb@nortelnetworks.com> wrote:
[...]
>I notice they say that "The core of the I/O processor is the
>CPU of the current Playstation", so i guess its a R3000
>based core ?

Yup. Details are at http://www.lsilogic.com

>Still, what do they mean by the term I/O processor, in this context ?

Glorified DMA controller, AFAIK.

By making it a programmable CPU, they get all sorts of benefits.
Some uses might be:
* disc effects, eg: FAT look-ups, transparent decompression,
    scatter-gather (to reduce seek times), intelligent caching...
* compression/encryption of (FireWire) network signals.
* asynchronous saving of game data.
* interpolation of controller/network inputs.
* generating random numbers (which is pretty time consuming if
    you need a lot of them).
* running game logic, leaving all 6 GFLOPS for geometry...

Basically, if it's available to program, then it'll be used for any
low-priority background tasks that happen to be lying around
waiting for CPU time...

>Surely, it means that this processor also has to be used as the main
>processor for using older games ?

Looks that way - the lower clock speed mode certainly implies that.

Given the power of the CPU, however, it could certainly emulate
PSX1 games. Their stated approach, however, is to provide 100%
compatability (ie: no better or worse). Maybe they'll release a
software emulator later, one that makes use of higher-res graphics
and fancy effects when playing (most) PSX1 games...

---
Russ

John Chasey <john@ioproductions.co.uk> wrote:
>>Speaking of which: I saw an article a couple of days ago
>>saying that dual-PIII systems were going to fall foul of some
>>supercomputer export laws in the US, so a single PSX2
>>might well be in the same situation...
>
>The PSX2 will be manufactured in the Far East, and Sony are
>a Japanese company, so at what point is anything exported
>from the US?

IIRC the PSX1 is made all over the place. SCEA's territory
includes Mexico and Canada, so if some are made in the
US, they'll need exporting. (OTOH, they might be made in
Brazil or Mexico...).

Sony probably wouldn't have an problems getting an export
license, though, in any case...

---
Russ

In article <36DDB3AB.378F5C87@igs.net> Paul DeMone <pdemone@igs.net> wrote:
>   Both of these figures are peak performance values which essentially
>   means you cannot achieve higher performance.  For real applications
>   you typically achieve a lot less.  If the Toshiba chip sustains 20%
>   and the Merced sustains 50% of these peak values on real customer
>   applications it will be impressive.

I'd be very surprised if the Toshiba chip didn't run well over 80% for
a significant amount of time and I'd expect it to hit its speed of light
"often enough"  when running its real customer applications.  Those might
not be your applications, and it might not do well on other applications,
but I'd bet Sony+Toshiba optimized it for the application they care about.

The attack of the killer micros is over - they won.  But, the fact that
killer micros took better advantage of certain technologies than did
"big iron" doesn't imply that killer micros are the best use of that
technology.  The killer micros are now prey.

-andy


-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own    

Factory wrote in message ...

>  Actually from looking at the specs of the psx2 I would guess that the
>complete system will perform as well as a tnt equipped celeron 400, but
>that is just a guess.

Hmm, that's interesting.  For the first time in a while (ever..?) the new
batch of consoles will be far behind consumer PC technology by the time
they're released.  I wonder how much a PII450+TNT will cost by the time the
PSXII or Dreamcast debuts?

- Mike

"Russ Williams" <russ@algorithm.demon.co.uk> writes:


> Speaking of which: I saw an article a couple of days ago saying that
> dual-PIII systems were going to fall foul of some supercomputer
> export laws in the US, so a single PSX2 might well be in the same
> situation...

It would be hard for the PSX2 to fall foul of the US export limits,
since it isn't a US made/designed device.  It is a Japanese device and
as such the US law is kinda funky.  The limit is something like
2billion theoretical units of performance.  AFAIK, the dual-processor
EV6 system slide under the limit(ie, it has a "little" basis in
reality for the number).


Aaron Spink
not speaking for Compaq

In article <TwNT+VAWzr32EwJ0@auia.demon.co.uk>,
  Joat Mon <joatmon@auia.demon.co.uk> wrote:

> >I notice they say that "The core of the I/O processor is the CPU of the
> >current Playstation", so i guess its a R3000 based core ?
> >
> >Still, what do they mean by the term I/O processor, in this context ?
> >Surely, it means that this processor also has to be used as the main
> >processor for using older games ?
> They'll almost certainly emulate the PSX1 in software (if a 133Mhz PC
> can do it, then why not this behemoth).  They won't make the hardware
> backwards compatible, esp. if this thing can churn out 75 million
> triangles, because of the PSX design.


"In a press conference held today via conference call, Sony VP of Third Party
Relations and Research & Development, Phil Harrison and Executive VP and COO
Kaz Hirai confirmed that the machine's I/O processor uses the same core as the
PlayStation. While that core will play all first generation PlayStation games,
it will not draw upon the machine's powerful graphics processor to do anything
more than rasterize the images produced by the old system. In other words, the
system will not add anti-aliasing, better fill rates, more impressive effects,
or anything else to the first generation PlayStation games."
<http://www.next-generation.com/jsmid/news/5975.html>


Richard



-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own    

On Thu, 4 Mar 1999 14:21:44 -0500, "Link" <mikemaz@interlog.com>
wrote:

>Factory wrote in message ...
>
>>  Actually from looking at the specs of the psx2 I would guess that the
>>complete system will perform as well as a tnt equipped celeron 400, but
>>that is just a guess.
>
>Hmm, that's interesting.  For the first time in a while (ever..?) the new
>batch of consoles will be far behind consumer PC technology by the time
>they're released.  I wonder how much a PII450+TNT will cost by the time the
>PSXII or Dreamcast debuts?

I really dont think the PC will be doing what we saw the PSX2 demo
doing in real time within this year, I certainly havent seen any
programs out currently that will do that or anything CLOSE to it in
geometry flexibility and smoothness of surfaces and clarity.  Perhaps
the best PC system now trying to render at television screen sizes
could do something similar, but I certainly havent seen anything like
it yet.  Even the PIII enabled games that look incredible like Outcast
pale in comparison.

In fact the only machine Ive seen capable of doing that type of
animtaion so far is the SGI Onyx series, which the PC is still a good
deal behind in actual capabilities even if the speeds are catching
up/caught up in Mhz.

-Geoff Howland
Lupine Games               http://www.lupinegames.com/

In article <MPG.1148ef176b46cee6989681@news.melbpc.org.au>,
Factory <faqtori@hotmail.com> wrote:
>  The caching on the psx2 was fairly unimpressive too, it had a 8k/16k 
>instruction/data cache and 16k scratchpad mem, no mention was made of any 
>other forms of caching. Considering that the current low end pentium 
>class chips have much more caching, I do not doubt that the merced 
>processors will have better caching.
>  If those specs are roughly right (I'm doing this from memory..) I would 
>expect the merced to out perform the psx2's cpu in real world work.

   Caching is most important when the internal bandwidth of the CPU is
significantly faster than the external memory bandwidth. If you've got
a very good memory throughput system, you don't need half as much
cache.  With a nice wide (128 bit? more?) rambus setup, you can shovel
in huge amouts of data into the processor nearly as fast as it wants.

   Remember, in consoles, they can design nice fat busses to shovel
around more data, while PCs tend to be more limited in their memory
i/o setups due to compatability reasons. 

Nathan Mates
--
<*> Nathan Mates - personal webpage http://www.visi.com/~nathan/  
# Network Programmer, Battlezone 2: see http://www.pandemicstudios.com
# NOT speaking for Pandemic Studios or Activision, ONLY myself
# "What are the facts, and to how many decimal places?" -R.A. Heinlein

In article <7bmlv8$9kf$1@news.interlog.com>, Link <mikemaz@interlog.com>
writes
>Factory wrote in message ...
>
>>  Actually from looking at the specs of the psx2 I would guess that the
>>complete system will perform as well as a tnt equipped celeron 400, but
>>that is just a guess.
Probably could run rings around a celeron for matrix and geometry work
(with that seirous FPU), but the integer performance should be similar.
Also, a TNT could churn out what, 10 million benchmark (read: small)
triangles?, and the PSX2 is supposed to handle 75 million, so I think
the 3d might just be a bit better ^_- than a TNT.

>Hmm, that's interesting.  For the first time in a while (ever..?) the new
>batch of consoles will be far behind consumer PC technology by the time
>they're released.  I wonder how much a PII450+TNT will cost by the time the
>PSXII or Dreamcast debuts?
The Dreamcast has already been out for 4-5 months, and costs around �150
($250), and it has the power of a pretty decent consumer machine
($800-$1000 machine), but it won't be out here until Fall (don't know
about the states).

I'd like to see PSX2 workstations (hehe), with a usb keyboard and mouse,
and a firewire flatpanel video display, as powerful as SGI's from a few
years ago, but for a couple hundred dollars (+display, since LCD panels
are outrageously priced).

---
In the beginner's mind there are many possibilities. In the expert's mind, there
are few.  -Shunryu Suzuki-

Joat Mon, joatmon@auia.demon.co.uk

Geoff Howland wrote in message <36dee3cd.76130691@news>...

>I really dont think the PC will be doing what we saw the PSX2 demo
>doing in real time within this year, I certainly havent seen any
>programs out currently that will do that or anything CLOSE to it in
>geometry flexibility and smoothness of surfaces and clarity.

No kidding.  Which demo is this?

>Perhaps
>the best PC system now trying to render at television screen sizes
>could do something similar, but I certainly havent seen anything like
>it yet.  Even the PIII enabled games that look incredible like Outcast
>pale in comparison.


Okay, so it looks good I guess :)  (personally, I think it's the low-res,
blurry TV signal.. but I'll have to see some demos :)


- Mike

> I'd not be amazed if it actually achieved that - what they seem to be
> claiming is one-tick-throughput matrix-vector multiply, which is
> probably achievable. Essentially it's a 300MHz geometry accelerator.

Prediction: within 5 years, a single instruction that performs a 4-vector
by 4x4-matrix multiply, of 32 bit FP numbers, will be as common as
vector FP and integer (MMX, KNI, 3DNow) are now.

Probably less than 5 years, but I want to have a bit of slack.

Paul DeMone wrote in message <36DDB3AB.378F5C87@igs.net>...
>aa158@valleynet.on.ca wrote:
>>
>> Two current news items at EETimes:
>>
>> "The Playstation 2's CPU, jointly developed by Toshiba and SCE, is
>> an enhanced version of the device described at ISSCC. The device
>> has floating-point performance of 6.2 Gflops..."
>> http://www.eetimes.com/story/OEG19990302S0026

I'd not be amazed if it actually achieved that - what they seem to be
claiming is one-tick-throughput matrix-vector multiply, which is
probably achievable. Essentially it's a 300MHz geometry accelerator.


Tom

In article <7bjr4a$cq2$1@nnrp1.dejanews.com>,
  aa158@valleynet.on.ca wrote:
> "The Playstation 2's CPU, jointly developed by Toshiba and SCE, is an enhanced
> version of the device described at ISSCC. The device has floating-point
> performance of 6.2 Gflops..."
> http://www.eetimes.com/story/OEG19990302S0026
>
> "Intel said Merced would achieve 3-Gflops extended-precision floating-point
> performance and 6-Gflops single-precision performance."
> http://www.eetimes.com/story/OEG19990226S0012

From the Merced numbers we can guess that it has 2 extended-precision FMAC
units and 2 single-precision FMAC units (the latter for the exclusive use
of the SIMD instructions).

The Playstation CPU, on the other hand, has 10 FMAC units (probably all but 1
single precision) and 4 FDIV units. What a BEAST! And two RDRAM channels,
so it will almost certainly have greater memory bandwidth than Merced too.

Imagine networking a bunch of these puppies via the Firewire interface
Poor man's supercomputer!

Remember when videogame systems used 10 year old hand-me-down CPUs?


-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own    

<cbm80@my-dejanews.com> wrote:
[...]
>The Playstation CPU, on the other hand, has 10 FMAC units
>(probably all but 1 single precision) and 4 FDIV units. What a
>BEAST! And two RDRAM channels, so it will almost certainly
>have greater memory bandwidth than Merced too.
>
>Imagine networking a bunch of these puppies via the Firewire
>interface Poor man's supercomputer!

Speaking of which: I saw an article a couple of days ago saying that
dual-PIII systems were going to fall foul of some supercomputer
export laws in the US, so a single PSX2 might well be in the same
situation...

>Remember when videogame systems used 10 year old
>hand-me-down CPUs?

Yup. I wonder how long it'll be before someone gets a PSX2, USB
keyboard/mouse and a FireWire hard disk and calls it a high-end
graphics workstation...

---
Russ

>Speaking of which: I saw an article a couple of days ago saying that
>dual-PIII systems were going to fall foul of some supercomputer
>export laws in the US, so a single PSX2 might well be in the same
>situation...


The PSX2 will be manufactured in the Far East, and Sony are a Japanese
company, so at what point is anything exported from the US?

John

In article <7bjr4a$cq2$1@nnrp1.dejanews.com>, aa158@valleynet.on.ca 
says...
> "The Playstation 2's CPU, jointly developed by Toshiba and SCE, is an enhanced
> version of the device described at ISSCC. The device has floating-point
> performance of 6.2 Gflops..."
> "Intel said Merced would achieve 3-Gflops extended-precision floating-point
> performance and 6-Gflops single-precision performance."

  Hmm I don't know about the merced, but from the specs I saw for the 
psx2 system, it has less registers (erm from what I remember 32 general 
regs, I assume the fp regs are similar), vs. the merced's 128 regs set 
(if I remember rightly).
  The caching on the psx2 was fairly unimpressive too, it had a 8k/16k 
instruction/data cache and 16k scratchpad mem, no mention was made of any 
other forms of caching. Considering that the current low end pentium 
class chips have much more caching, I do not doubt that the merced 
processors will have better caching.
  If those specs are roughly right (I'm doing this from memory..) I would 
expect the merced to out perform the psx2's cpu in real world work.
  
  Actually from looking at the specs of the psx2 I would guess that the 
complete system will perform as well as a tnt equipped celeron 400, but 
that is just a guess.

On Thu, 4 Mar 1999 21:38:44 -0500, "Link" <mikemaz@interlog.com>
wrote:

>Geoff Howland wrote in message <36dee3cd.76130691@news>...
>
>>I really dont think the PC will be doing what we saw the PSX2 demo
>>doing in real time within this year, I certainly havent seen any
>>programs out currently that will do that or anything CLOSE to it in
>>geometry flexibility and smoothness of surfaces and clarity.
>
>No kidding.  Which demo is this?

Its the PSX2 animation that shows the woman from the cover of the new
Ridge Racer game walking down a runway-catwalk thing, her hair bounces
and turns in near individual strands (well they are individual, but
they are thicker than a single hair), then it switches to a fighting
scene with about 20-30 guys jumping around that are pretty well
detailed and very smooth and curved surfaces (looks like it at least)
and two guys fighting in the middle, which shows that it can do a lot
of animated geometry at the same time while filling the whole screen.

Then it shows a old mans face that looks bump mapped animating with
real time mesh/NURB/whatever-theyre-using deformations that show
muscle moves to different facial features, then the make like 100 of
these bop around the screen.

Then they have cars from Ridge Racer racing in the inside of a city,
that are definitely curved surfaces and just look incredibly smooth
and the light is specularly lighting them just AMAZINGLY.

Ive seen nothing like it on anything less than an Onyx, like I said.

>>Perhaps
>>the best PC system now trying to render at television screen sizes
>>could do something similar, but I certainly havent seen anything like
>>it yet.  Even the PIII enabled games that look incredible like Outcast
>>pale in comparison.
>
>Okay, so it looks good I guess :)  (personally, I think it's the low-res,
>blurry TV signal.. but I'll have to see some demos :)

I dont think that a PIII 500 with the top 'o the line consumer
graphics card could do this at TV resolutions.  

Especially not the geometry as we all know we have that geometry limit
and the PSX2 looks like its a geometry/matrix pumping machine. Of
course this is only my speculation so take it for what its worth, but
I think PCs currently have a geometry bottleneck which this one
doesn't or at least has a bigger opening for the bottle.

Its a significantly cool machine.  

As other people have said, this may be the next low-priced graphics
workstation.  In fact, if it could be, it would be the next low priced
graphics workstation priced even at $1500 for the console since it
seems to do its thing better than PCs could  currently.

-Geoff Howland
Lupine Games               http://www.lupinegames.com/

On Thu, 04 Mar 1999 21:29:11 -0600, Andy Glew <glew@cs.wisc.edu>
wrote:

>> I'd not be amazed if it actually achieved that - what they seem to be
>> claiming is one-tick-throughput matrix-vector multiply, which is
>> probably achievable. Essentially it's a 300MHz geometry accelerator.
>
>Prediction: within 5 years, a single instruction that performs a 4-vector
>by 4x4-matrix multiply, of 32 bit FP numbers, will be as common as
>vector FP and integer (MMX, KNI, 3DNow) are now.
>
>Probably less than 5 years, but I want to have a bit of slack.

Prediction: In 15-20 years almost all specific computer problems we
have now will seem ridiculously pathetic.

-Geoff Howland
Lupine Games               http://www.lupinegames.com/

In article <36df551a.105140012@news>,
Geoff Howland <ghowland@lupineNO.SPAMgames.com> wrote:
>As other people have said, this may be the next low-priced graphics
>workstation.  In fact, if it could be, it would be the next low priced
>graphics workstation priced even at $1500 for the console since it
>seems to do its thing better than PCs could  currently.

   For $400, I'll pay for the PSX2 to be my next *VIDEO* board for my
PC. :) 

   Strip out the DVD interface, just put the CPU and graphics
subsystems on a card with 32+MB RAM and a VGA port, and it'll be a
VERY nice system addon.

Nathan Mates
--
<*> Nathan Mates - personal webpage http://www.visi.com/~nathan/  
# Network Programmer, Battlezone 2: see http://www.pandemicstudios.com
# NOT speaking for Pandemic Studios or Activision, ONLY myself
# "What are the facts, and to how many decimal places?" -R.A. Heinlein

In article <36DF4F87.FFDEFF0C@cs.wisc.edu>,
Andy Glew  <glew@cs.wisc.edu> wrote:
>> I'd not be amazed if it actually achieved that - what they seem to be
>> claiming is one-tick-throughput matrix-vector multiply, which is
>> probably achievable. Essentially it's a 300MHz geometry accelerator.

>Prediction: within 5 years, a single instruction that performs a 4-vector
>by 4x4-matrix multiply, of 32 bit FP numbers, will be as common as
>vector FP and integer (MMX, KNI, 3DNow) are now.

   Prediction: within 2 years, a lot of that functionality for matrix
multiplies will be sitting on your 3D video board, as off-CPU geometry
processors get more and more common. Even a slightly stripped down
version of the PSX2 CPU on the video board would be a serious
improvement to overall system speeds. If you're doing NURBS and
dynamic tessellation, that's a boatload of polygons generated, and
it's far better to not bog down the system bus with that data-- keep
it on the 3D card where mega-bandwidth is available.

   Profiling by folks of common 3D games (see also one of Brian Hook's
recent .plan updates) show that in a large number of apps, the app is
getting less than *half* of CPU time. The 3D API {OpenGL|D3D} and the
3D card's drivers are the culprits here, and the more work they do
that can be shifted to the 3D card, the better. Intel and AMD are
gonna hate geometry accelerators, as that sinks their marketing plans,
but if the system is overall faster withOUT a new CPU, I won't mind.

Nathan Mates

--
<*> Nathan Mates - personal webpage http://www.visi.com/~nathan/  
# Network Programmer, Battlezone 2: see http://www.pandemicstudios.com
# NOT speaking for Pandemic Studios or Activision, ONLY myself
# "What are the facts, and to how many decimal places?" -R.A. Heinlein

In comp.arch Russ Williams <russ@algorithm.demon.co.uk> wrote:

: Speaking of which: I saw an article a couple of days ago saying that
: dual-PIII systems were going to fall foul of some supercomputer
: export laws in the US, so a single PSX2 might well be in the same
: situation...

Perhaps the American taxpayers will send inspectors around the world to
read the serial numbers off of the PSX2 boxes. A few years ago, I had
two men from the US Dept of Commerce Bureau of Export Administration
from Washington DC come to read the serial number of our 1 cpu 182MHz 
21064 Digital 7000 system. They first read our own serial number and 
declared that it was not the correct serial number - until we led them
to the actual Digital serial number. The official definition of a 
supercomputer isn't always very useful - unless you are a civil servant
wanting to see the world.
-- 
John McCallum            School of Computing, Department of Computer Science
johnm@comp.nus.edu.sg    National University of Singapore,  Singapore 119260
//www.comp.nus.edu.sg/~johnm      tel: 874-2990       NUS, DISCS  S15 #03-13

On Fri, 05 Mar 1999 04:03:13 GMT, ghowland@lupineNO.SPAMgames.com
(Geoff Howland) said:

> I dont think that a PIII 500 with the top 'o the line consumer
> graphics card could do this at TV resolutions.

You keep repeating this, as if TV resolutions were amazingly high or
something.

I hope that's not what you mean, since only some TV sets are good
enough to handle 800x600.


The PSX2's specs sure seem impressive enough, though.  But I won't
make any assumptions as to its quality vs graphics cards for PCs until
the PSX2 is something more than marketingware.  (There really isn't
any point in comparing it with existing PC products; compare it with
other vapor-/marketingware.)


J<>I
-- 
Many an ancient lord's last words had been,
"You can't kill me because I've got magic aaargh."
- Magic armour is not all it's cracked up to be.
  (Terry Pratchett, Interesting Times)

In article <36DF4F87.FFDEFF0C@cs.wisc.edu>,
  Andy Glew <glew@cs.wisc.edu> wrote:

> Prediction: within 5 years, a single instruction that performs a 4-vector
> by 4x4-matrix multiply, of 32 bit FP numbers, will be as common as
> vector FP and integer (MMX, KNI, 3DNow) are now.
>
> Probably less than 5 years, but I want to have a bit of slack.


Can AltiVec (PowerPC G4) do this?

How does its vector unit compare to the dual vector units of the Toshiba chip?



Richard






-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own    

Chris Ebenezer <chriseb@nortelnetworks.com> wrote:
[...]
>I went to http://www.playstation.com, according to the information
>there.  The I/O processor has a R3000 core, but then heres a
>suprise, the 128 bit CPU is implementing a MIPS IV subset!

Yup. Plus the 107 extra vector/MM ops they've added.

>I thought Sony had stated that they weren't going to go with MIPS
>for the next generation console ?

Probably just a rumour. Until a couple of days ago SCEI refused
to even admit that they were working on a PS2...

---
Russ

Factory wrote in message ...
>In article <7bjr4a$cq2$1@nnrp1.dejanews.com>, aa158@valleynet.on.ca
>says...
>> "The Playstation 2's CPU, jointly developed by Toshiba and SCE, is an
enhanced
>> version of the device described at ISSCC. The device has floating-point
>> performance of 6.2 Gflops..."
>> "Intel said Merced would achieve 3-Gflops extended-precision
floating-point
>> performance and 6-Gflops single-precision performance."
>
>  Hmm I don't know about the merced, but from the specs I saw for the
>psx2 system, it has less registers (erm from what I remember 32 general
>regs, I assume the fp regs are similar), vs. the merced's 128 regs set
>(if I remember rightly).
>  The caching on the psx2 was fairly unimpressive too, it had a 8k/16k
>instruction/data cache and 16k scratchpad mem, no mention was made of any
>other forms of caching. Considering that the current low end pentium
>class chips have much more caching, I do not doubt that the merced
>processors will have better caching.
>  If those specs are roughly right (I'm doing this from memory..) I would
>expect the merced to out perform the psx2's cpu in real world work.
>
>  Actually from looking at the specs of the psx2 I would guess that the
>complete system will perform as well as a tnt equipped celeron 400, but
>that is just a guess.

The Cpu sounds pretty yummy alone...
    Emotion Engine Features and General Specifications
    CPU core 128bit RISC (MIPS IV-subset)
    Clock Frequency 300MHz
    Integer Unit 64bit (2-way Superscalar)
    Multimedia extended instructions 107 instructions at 128bit width
    Integer General Purpose Register 32 at 128 bit width
    TLB 48 double entries
    Instruction Cache 16KB (2-way)
    Data Cache 8KB (2-way)
    Scratch Pad RAM 16KB (Dual port)
    Main Memory 32MB (Direct RDRAM 2ch@800MHz)
    Memory bandwidth 3.2GB/sec
    DMA 10 channels
    Co-processor1 FPU (FMAC x 1, FDIV x 1)
    Co-processor2 VU0 (FMAC x 4, FDIV x 1)
    Vector Processing Unit VU1 (FMAC x 5, FDIV x 2)
    Floating Point Performance 6.2GFLOPS
    Geometry
    + Perspective Transformation 66Million Polygons/sec
    + Lighting 38Million Polygons/sec
    + Fog 36Million Polygons/sec
    Curved Surface Generation (Bezier) 16Million Polygons/sec
    Image Processing Unit MPEG2 Macroblock Layer Decoder
    Image Processing Performance 150Million Pixels/sec
    Gate width 0.18 micron
    VDD Voltage 1.8 V
    Power Consumption 15 Watts
    Metal Layers 4
    Total Transistors 10.5 Million
    Die Size 240 mm2
    Package 540pin PBGA

But then the GP makes you go oooooooooohhhhhhhhhhhhh
    Graphics Synthesizer � Features and General Specifications
    GS Core Parallel Rendering Processor with embedded DRAM
    Clock Frequency 150 MHz
    No. of Pixel Engines 16 (in Parallel)
    Embedded DRAM 4 MB of multi-port DRAM (Synced at 150MHz)
    Total Memory Bandwidth 48 Giga Bytes per Second
    Combined Internal
    Data Bus bandwidth 2560 bit
    Read 1024 bit
    Write 1024 bit
    Texture 512 bit
    Display Color Depth 32 bit (RGBA: 8 bits each)
    Z Buffering 32 bit
    Rendering Functions Texture Mapping, Bump Mapping
    Fogging, Alpha Blending
    Bi- and Tri-Linear Filtering
    MIPMAP, Anti-aliasing
    Multi-pass Rendering
    Rendering Performance
    Pixel Fill Rate 2.4 Giga Pixel per Second
    (with Z buffer and Alphablend enabled)
    1.2 Giga Pixel per Second
    (with Z buffer, Alpha and Texture)
    Particle Drawing Rate 150 Million /sec
    Polygon Drawing Rate 75 Million /sec (small polygon)
    50 Million /sec (48 Pixel quad with Z and A)
    30 Million /sec (50 Pixel triangle with Z and A)
    25 Million /sec (48 Pixel quad with Z, A and T)
    Sprite Drawing Rate 18.75 Million (8 x 8 Pixels)
    Display output NTSC/PAL
    Digital TV (DTV)
    VESA (maximum 1280 x 1024 pixels)
    Silicon process technology 0.25 � 4-level metal
    Total number of transistors 43 Million
    Die size 279mm2
    Package Type: 384 pin BGA


mmmmm

>  Actually from looking at the specs of the psx2 I would guess that the 
>complete system will perform as well as a tnt equipped celeron 400, but 
>that is just a guess.

p2/3/celerons basically have 1mflop/mhz aka celeron 400 = 400mflops or
.4 gflops vs the playstation 2's 6.2gflops.  If you could tie four of 
these together you'd make the top 500 supercomputer list.

On 05 Mar 1999 12:24:48 +0100, Jan Ingvoldstad <jani@ifi.uio.no>
wrote:

>On Fri, 05 Mar 1999 04:03:13 GMT, ghowland@lupineNO.SPAMgames.com
>(Geoff Howland) said:
>
>> I dont think that a PIII 500 with the top 'o the line consumer
>> graphics card could do this at TV resolutions.
>
>You keep repeating this, as if TV resolutions were amazingly high or
>something.
>
>I hope that's not what you mean, since only some TV sets are good
>enough to handle 800x600.

Umm, I meant it means "TV screens are like 3xx by 2xx pixels in
dimension, so thats MUCH smaller than PC games have to render to"
Meaning that you could theoretically do much more in-depth graphic
wise PC games if you only had to render to 3xx by 2xx pixels, but it
still wouldnt look as good as the PSX2 because of the PC's geometry
limit.

>The PSX2's specs sure seem impressive enough, though.  But I won't
>make any assumptions as to its quality vs graphics cards for PCs until
>the PSX2 is something more than marketingware.  (There really isn't
>any point in comparing it with existing PC products; compare it with
>other vapor-/marketingware.)

MarketingWare by be MarketingWare, but it still has to DO the
calculations, and right now I think I can safely say (from everything
Ive seen) that a PC couldnt keep up or do that quality.

PC product MarketingWare is normally prerendered 3D Studio stuff, this
is hardly the same thing.

-Geoff Howland
Lupine Games               http://www.lupinegames.com/

In article <36df551a.105140012@news>,
  ghowland@lupineNO.SPAMgames.com (Geoff Howland) wrote:

> Then they have cars from Ridge Racer racing in the inside of a city,
> that are definitely curved surfaces and just look incredibly smooth
> and the light is specularly lighting them just AMAZINGLY.

You didn't mention the motion blur... Very, very nice.

> I dont think that a PIII 500 with the top 'o the line consumer
> graphics card could do this at TV resolutions.

There isn't a chance in Hades it could... That comment about it being
comparable to a Celeron-with-a-TNT-card is ludicrous to the point of being
surreal. (Maybe he was talking about a *magic* Celeron. ;)

> Its a significantly cool machine.

It'll be interesting to see if it lives up to its (possibly rigged?) demos. 
It sure looks "for real".  I want one.	Perhaps two...


Dave Nagy&#137;

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own    

>Mahlzeit
>
>
>Geoff Howland (ghowland@lupineNO.SPAMgames.com) wrote:
>> Umm, I meant it means "TV screens are like 3xx by 2xx pixels in
>> dimension, so thats MUCH smaller than PC games have to render to"
>PAL is 768x576 with 25 fps (actually 50 halve-frames per second).
>So it is not that much from todays PC games.

The graphic chip spec lists digital TV output and VESA 1280x1024.  Shouldn't
affect the graphic rate too much with a higher res.  1280x1024, textures,
alphas, z-buffer with 4Mb of embedded DRAM.... hmmm.  Maybe not.  Could be
done with 16-bit colour, Talisman/VR style z-tiling* and texture
compression.  Maybe not.

matt.
ex-techo

* This would require a pre-render hardware sort of the display list with
each region being sequentially split between the 16 parallel pixel engines.

Matthias Bruestle <m@mbsks.franken.de> wrote:
>Geoff Howland (ghowland@lupineNO.SPAMgames.com) wrote:
>> Umm, I meant it means "TV screens are like 3xx by 2xx pixels
>> in dimension, so thats MUCH smaller than PC games have to
>> render to"
>
>PAL is 768x576 with 25 fps (actually 50 halve-frames per second).
>So it is not that much from todays PC games.

The half-frames part is very significant: it halves the vertical
resolution. Since dealing with non-square pixels is a major
hassle for artwork, the horizontal res will be halved as well.
Hence the PSX1 most often uses resolutions of 320x240
and 320x256 non-interlaced. The PSX2 might use a higher
res and antialias it down (eg: 640x480 => 640x240), but
PC games are heading towards 1024x768 (and higher).

---
Russ

In comp.arch Nathan Mates <nathan@visi.com> wrote:

:    Caching is most important when the internal bandwidth of the CPU is
: significantly faster than the external memory bandwidth. If you've got
: a very good memory throughput system, you don't need half as much
: cache.  With a nice wide (128 bit? more?) rambus setup, you can shovel
: in huge amouts of data into the processor nearly as fast as it wants.

Caches reduce both latency and bandwidth.  For a graphics rendering engine,
bandwidth is more important than latency.  For a PC, it will usually be a
matter of latency and not bandwidth.

:    Remember, in consoles, they can design nice fat busses to shovel
: around more data, while PCs tend to be more limited in their memory
: i/o setups due to compatability reasons. 

I can't understand why a PC would be limited in i/o setups for anything 
involving compatability.  Could you give an example?  

Mark

: Nathan Mates
: --
: <*> Nathan Mates - personal webpage http://www.visi.com/~nathan/  
: # Network Programmer, Battlezone 2: see http://www.pandemicstudios.com
: # NOT speaking for Pandemic Studios or Activision, ONLY myself
: # "What are the facts, and to how many decimal places?" -R.A. Heinlein

-- 
  ~~~~~~~~~~~~~~~~~~~~~~~brehob@cps.msu.edu~~~~~~~~~~~~~~~~~~~~~~~~~
  | | The reports of SIMD's death have been greatly exaggerated  | |
  | -=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- |
  ~~~~~~Mark Brehob: Ultimate Player, Gamer, Computer Geek~~~~~~~~~~

Mark W Brehob (brehob@cse.msu.edu) wrote:
: I can't understand why a PC would be limited in i/o setups for anything 
: involving compatability.  Could you give an example?  

Because consoles are fixed, whole-cloth designs, one can design point to
point I/O for all components. As opposed to a PC where one has to use
something like PCI or AGP for graphics. Keeping in mind that AGP is point
to point so its more or less out from under one of the major constraints on
PCI performance. But you can't just arbitrarily widen the processor to
graphics bus in a PC. Nor can one crank up the clock rate. In a console,
these things can be done because both components (CPU, graphics processor)
are custom built for the box.

The next generation of switched fabric interconnects pushes further on this
point as well. But those will start out in the server space and move down
to "gaming PCs" in the year to two year timeframe...

-Z-

In article <7buekv$ik4$1@msunews.cl.msu.edu>,
Mark W Brehob  <brehob@cse.msu.edu> wrote:
>:    Remember, in consoles, they can design nice fat busses to shovel
>: around more data, while PCs tend to be more limited in their memory
>: i/o setups due to compatability reasons. 

>I can't understand why a PC would be limited in i/o setups for anything 
>involving compatability.  Could you give an example?  

   Easy: slots (ISA, PCI, AGP, etc) have a fixed bit width throughout
their lifetime. However, video boards based on them have upped their
internal bandwidths over time-- 64 and 128 bit internal pathways on
video boards are quite common these days. If you've got a 64 bit card
on a 32 bit PCI slot, wouldn't it be better to redesign the PCI to be
64 bits as well for a better pathway to system RAM? Oops, can't do
that. Want to make a new PCI card that talks to the motherboard at
50Mhz rather than 33Mhz? Oops, can't do that.

   On a console, you don't have these issues-- the designers pick the
bus width and speed upfront to best balance performance, cost, and the
like. They're not limited to PCI/ISA/AGP or whatever bus their last
console had; they could pick a 48-bit wide bus running at 78.2Mhz if
they wanted. 

Nathan Mates
--
<*> Nathan Mates - personal webpage http://www.visi.com/~nathan/  
# Network Programmer, Battlezone 2: see http://www.pandemicstudios.com
# NOT speaking for Pandemic Studios or Activision, ONLY myself
# "What are the facts, and to how many decimal places?" -R.A. Heinlein

Marc Coupar <marc@taste.demon.co.uk> wrote in message
news:01be699e$2cfed500$8b47e4d4@xqiowghw...
> I would expect even the first batch of games on Psx 2 to be 640x480.

And surely with the amount of graphic grunt they have they'll be looking to
HDTV, as well. Are there any specs on the maximum resolution of the PSX2?
I'd expect it to be able to deliver games at 2000x1000x72Hz (which I believe
is HDTV res) if the PSX2 actually achieves the advertised specs. At a
minimum, it should be able to do 1024x768 on a digital TV.

Colin

In article <920810365.12598.0.nnrp-04.c2de56b7@news.demon.co.uk>,
Russ Williams <russ@algorithm.demon.co.uk> wrote:
>The half-frames part is very significant: it halves the vertical
>resolution.

And any Amiga user can tell you that you just render into the full
screen at 25 (or 30) FPS and let the video subsystem generate interlaced
output.

In other words the halved resolution doesn't impact programming.

The main reason for using a non-interlaced mode (or scan-doubling) is to
avoid flicker when you have aliasing. If the rendering is properly
antialiased this isn't an issue (alternate frames don't present the
same information in this mode, mind, you really do get better resolution
than with scan-doubled screens).

The second issue is that on a normal TV screen the luminance resolution
is significnatly below the full 768 pixels per scan line, so you get
color artifacts if you try and render anything finer than about 300-400
pixels per line.

We're already seeing higher resolution TVs in the store, which means the
PSX2 will likely be able to make effective use of 640x480 up to 768x512
(I was able to coax almost this many displayed pixels out of my Amiga 3000
on a Commodore 1942 analog RGB monitor).

-- 
In hoc signo hack, Peter da Silva <peter@baileynm.com>
 `-_-'   "Heb jij vandaag je wolf al geaaid?"
  'U`
         "Tell init(8) to lock-n-load, we're goin' zombie slaying!"

nathan@visi.com (Nathan Mates) writes:
>[...] If you've got a 64 bit card
>on a 32 bit PCI slot, wouldn't it be better to redesign the PCI to be
>64 bits as well for a better pathway to system RAM? Oops, can't do
>that. Want to make a new PCI card that talks to the motherboard at
>50Mhz rather than 33Mhz? Oops, can't do that.

I have a lot of Sun computers at work with 66 MHz/64 bits PCI slots.
You point may be a valid one, but not the example used.

--Pete

On 8 Mar 1999 19:07:47 -0600, "Colin Fleming" <colin@caverock.net.nz>
said:

> I'd expect it to be able to deliver games at 2000x1000x72Hz (which I
> believe is HDTV res) if the PSX2 actually achieves the advertised
> specs.

I find that hard to believe, considering Andy Mucho's article, which
had the following line innit:

    Display output NTSC/PAL
    Digital TV (DTV)
    VESA (maximum 1280 x 1024 pixels)

(Message-ID: <920657570.28485.0.nnrp-09.d4e4fdbd@news.demon.co.uk>)


> At a minimum, it should be able to do 1024x768 on a digital TV.

That would appear to be possible, yes.  But that's a far cry from what
you expect, which is 2.5 times that resolution, and probably at a
higher refresh rate, too.  Besides, you don't know what frame rate
you'll get, or how good the textures will be...


I don't have a hard time believing that it will perform well at
1280x1024.  We already have a PC add-on card manufacturer[0] who
claims to reach nice frame rates at that resolution with SLI
configured Riva TNTs, and besting that shouldn't be impossible.


(But when do I get my InfiniteReality^2 add-on card for my home
computer?  Hmm?  Hmm?)


---------------
[0] Metabyte's marketingware (of course), see also (if you care)
    <URL:http://www.sharkyextreme.com/hardware/metabyte_tntsli_p/>.


J<>I
-- 
Many an ancient lord's last words had been,
"You can't kill me because I've got magic aaargh."
- Magic armour is not all it's cracked up to be.
  (Terry Pratchett, Interesting Times)

"Colin Fleming" <colin@caverock.net.nz> writes:


> And surely with the amount of graphic grunt they have they'll be looking to
> HDTV, as well. Are there any specs on the maximum resolution of the PSX2?
> I'd expect it to be able to deliver games at 2000x1000x72Hz (which I believe
> is HDTV res) if the PSX2 actually achieves the advertised specs. At a
> minimum, it should be able to do 1024x768 on a digital TV.
> 

If the info is correct and the Rasterization engine contains 4MB of
embedded dram, then the possible resulutions are as follows.

All assuming 32bit color(8b,8r,8g,8a) and  16, 24, 32 bits of Z

		16	24	32
640x480		1.8MB	2.1MB	2.5MB
800x600		2.88MB	3.36MB	3.84MB  
1024x768	4.7MB	5.5MB	6.29MB

I'm going to assume that they will support roughly 800x600x32x32 or 
640x480x32x32.  Anything higher than 800x600 is wishfull thinking
unless you want bad quality or want to pay ~2k per PSX2.

Aaron Spink
not speaking for compaq

In article <7blp77$8en$1@flex.london.pipex.net>,
  "Russ Williams" <russ@algorithm.demon.co.uk> wrote:

> Speaking of which: I saw an article a couple of days ago saying that
> dual-PIII systems were going to fall foul of some supercomputer
> export laws in the US, so a single PSX2 might well be in the same
> situation...

Since when was PS2 going to be US-made? Or does Japan has similar
export laws as US?

> >Remember when videogame systems used 10 year old
> >hand-me-down CPUs?
>
> Yup. I wonder how long it'll be before someone gets a PSX2, USB
> keyboard/mouse and a FireWire hard disk and calls it a high-end
> graphics workstation...

Come on, that was already claimed about Nintendo64. 'A high end
SGI graphics workstation on your desk for $200!'. N64 didn't live
up to that hype, and frankly I don't think PS2 will either, even though
it will probably be very impressive.

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own