engineer forgot bilinear filtering unit in PowerVR PCX1 ?

ram

Newcomer
http://www.firingsquad.com/print_article.asp?current_section=Hardware&fs_article_id=1422

The PCX was supposed to have bilinear filtering. The engineers knew it was an important quality feature but during chip development, the transistor block required for bilinear filtering cycle was inadvertently left out by an engineer. By the time they realized the mistake, the decision was made that it would have been too costly and too time consuming to fix it. They had to go with the flawed chip.
 
How could that possibly happen?! Sounds far more likely that there was a bug that was overlooked.

MuFu.
 
MuFu said:
How could that possibly happen?! Sounds far more likely that there was a bug that was overlooked.

MuFu.

Well, it IS possible. Keep in mind the diagrams are very complex, and that at some point the design has to be submitted to a foundry for production.
 
Seriously smells of BS to me. A single engineer made the mistake and nobody else noticed? Both the foundry and PVR missed it, possibly across several ASIC revisions? The physical layout was verified despite there being an entire block of functionality missing? Come on.

...although I can already almost hear the likes of Kristof etc chiming in with "It happens all the time! Once we forgot to include a DB15 port and it caused all sorts of monitor-incompatibilty problems, haha".

MuFu.
 
Re: engineer forgot bilinear filtering unit in PowerVR PCX1

"Livid" does not even begin to describe my state at the moment!

ram said:
http://www.firingsquad.com/print_article.asp?current_section=Hardware&fs_article_id=1422
The PCX was supposed to have bilinear filtering. The engineers knew it was an important quality feature but during chip development, the transistor block required for bilinear filtering cycle was inadvertently left out by an engineer. By the time they realized the mistake, the decision was made that it would have been too costly and too time consuming to fix it. They had to go with the flawed chip.

Utter rubbish. I imagine that a strongly worded letter to firing squad to detract this is in order.

At the outset PCX1 was intended to have linear MIP mapping but NOT bilinear filtering. It was a conscious decision to keep the number of texture accesses low (2 compared to 4) while eliminating the transistions between MIPmap levels.


Given the state of available consumer HW at the time of design (i.e. nothing) and that linear MIP mapping would be a relatively huge boost to texturing quality of the available software games (e.g. picture the hideous aliasing in Doom), it was felt this was sufficient.

Simon (speaking just for myself but from 1st hand knowledge)

[Edit: Made use of Zaphod's suggestion]
 
Re: engineer forgot bilinear filtering unit in PowerVR PCX1

Simon F said:
Given the state of available consumer HW at the time of design (i.e. nothing)

I guess the Voodoo Graphics card was non existant for PowerVR then :rolleyes:
 
Re: engineer forgot bilinear filtering unit in PowerVR PCX1

xGL said:
Simon F said:
Given the state of available consumer HW at the time of design (i.e. nothing)

I guess the Voodoo Graphics card was non existant for PowerVR then :rolleyes:

Uhmm... Probably, yes... :rolleyes:

I think you might have missed the five keywords here: At the time of design. If my memory serves me the PCX1 was announced before the Voodoo, and released at about the same time.
 
Simon, while we're on the subject, make sure you don't forget to include "wanktastic speed" in the Series 5.

MuFu.
 
There are tons of errors in that article. I didn't really take stock in anything the guy said as a result.

Why would he cover PowerVR and not Rendition (in other than the last paragraph)? Where was the Matrox G200? Uhgg.... there was a much better historical article over at http://accelenation.com/?ac.id.123.1
 
swaaye said:
Where was the Matrox G200?

Forget that...how about the original Matrox Mystique? (AKA Matrox Mistake?) I can still see the box-art on that one...a jester if I'm not mistaken. It was Matrox's first foray into 3D, and first new brand since the original "Millenium".

Lacked bilinear filtering, and if memory serves, lacked alpha blending as well. Matrox PR touted "screen door transparency" (dithering) as a performance feature. Only problem was that other cards (Voodoo, PCX, Verite...) handled alpha blending just fine. ;)
 
I'm not doubting the "at the time of design" part.
But the announcement time doesn't fit with what I remember.

I remember my early days of www.dimension3d.com when the fights about VooDoo vs Rendition raged. Occationally there were questions to the PoverVR resident (Simon F) if there ever where going to be a PC version of PVR. I had spent (what I then though) quite a long time at that site before he answered with the official news that there would be a PC version called PCX1.
(And the crowd rejoiced.)

But I guess the (core) design was the same in that as in the earlier (arcade) chips.


About Matrox:
Their first 3D card was actually the last Millenium that came before Mystique. (I don't remember if there were more than one Millenium version before Mystique.) It had Z-buffering and Gouraud shaded triangles. It did however lack some minor features (like texturing :D).
 
Basic said:
But the announcement time doesn't fit with what I remember.
As I said, it could have been just my mind playing tricks on me... So I had to dig around for a couple of minutes.

The oldest reference I found to the PCX1 is this USENET 3d graphics FAQ from 19 March 1996 that references www.videologic.com as its source of information. It also has information on 3Dfx, so the Voodoo anoouncement seems to have been earlier than the Accelenation.com article linked above would indicate.

Edit (with a little help of The Internet Archive WayBack Machine): Apparently VideoLogic first announced it's deal with NEC on 19 September 1995, and the chip itself (sampling) on 26 February 1996. 3DFx announced the Voodoo as early as 6. November 1995. Can't tell if they had hardware at that point though.
 
Zaphod said:
Basic said:
But the announcement time doesn't fit with what I remember.
As I said, it could have been just my mind playing tricks on me... So I had to dig around for a couple of minutes.

The oldest reference I found to the PCX1 is this USENET 3d graphics FAQ from 19 March 1996 that references www.videologic.com as its source of information. It also has information on 3Dfx, so the Voodoo anoouncement seems to have been earlier than the Accelenation.com article linked above would indicate.

Edit (with a little help of The Internet Archive WayBack Machine): Apparently VideoLogic first announced it's deal with NEC on 19 September 1995, and the chip itself (sampling) on 26 February 1996. 3DFx announced the Voodoo as early as 6. November 1995. Can't tell if they had hardware at that point though.

Checking the WayBack Machine for my NEC/VideoLogic page on my Dimension3D.com site I found the following press release...

NEC Electronics Inc. Debuts 3D Graphics Processor Family Based on NEC and VideoLogic's PowerVR Technology
(2/26/1996)

This is the press release announcing the PCX1. VideoLogic later announced the card using the PCX1.

VideoLogic Targets PC OEMs with PowerVR 3D Accelerator Card
(5/9/1996)

They didn't release it then, but said they would release the retail board to coincide with the release of DirectX later that summer. They announced the same day that the card would show up in Compaq Presarios. I believe the Presarios had them way before they ever hit retail. The actual press release gave the following info...

Press Release said:
NEC’s single-chip PCX1 is slated for sampling in Q2 ‘96 and volume in Q3 ‘96. Volume price is expected to be less than $50 in quantities of 10,000 per month by the end of 1996 or the beginning of 1997. Packaging is 208-pin, plastic quad flat pack (PQFP).

Tommy McClain
 
Zaphod said:
Edit (with a little help of The Internet Archive WayBack Machine): Apparently VideoLogic first announced it's deal with NEC on 19 September 1995, and the chip itself (sampling) on 26 February 1996. 3DFx announced the Voodoo as early as 6. November 1995. Can't tell if they had hardware at that point though.

Almost certianly 3dfx had hardware. SST1 was being touted around in the arcade and sim space for some time before (as a triple texturing part). Its was only a matter of luck that memory prices dropped and they decided to target it for the consumer space with only one filtering unit.
 
swaaye said:
There are tons of errors in that article. I didn't really take stock in anything the guy said as a result.

Why would he cover PowerVR and not Rendition (in other than the last paragraph)? Where was the Matrox G200? Uhgg.... there was a much better historical article over at http://accelenation.com/?ac.id.123.1

i totally agree, especially the bit about Rendition not being included - imo second only to 3Dfx, yet still there were more times i saw partcle effects filtered on my V2x00 than on my friends Voodoo cards - this pleased me immensly at the time considering it was my defence against the bragging ;)
 
MuFu said:
Simon, while we're on the subject, make sure you don't forget to include "wanktastic speed" in the Series 5.

MuFu.

He mostly develops algorithms doesn't he? That should rather read something like wanktastic AA algorithm....*runs*
 
Is he the only employee of PowerVR ?

because I've only heard the name of Simon Fenney
since the time I made my PVRSG page (which never came..).

LeGreg
 
Back
Top