How come Matrox never tried to improve the Parhelia?

Akumajou

Regular
I mean ok the Parhelia was a .15nm GPU and AGP 4X, how come they never tried to do a die shrink to .13nm process and add AGP8X while leaving the 80million transistors the same kind of like the way AMD went from Palomino core to Thoroughbred core.

What and why would it be difficult for Matrox to do that?

I mean at .15nm the P's clock was somewhere around 220 or more, a .13nm could have definetly gotten them to 300Mhz safely or probably 375Mhz and I am being conservative here.

Wouldn't it have been worth it for Matrox just for the sake of improving their weakness without spending money to support DX9

Maybe Matrox would have been in better shape if they would have done that and maybe, just maybe they would have had another chance to go and support Dx9 later with a mature .13nm process...

I do hope they make a come back, not an expensive one like Parhelia but one to at least keep them in the market.
 
They did do a Mk2 IIRC

There is the Parhelia 8X which is the same thing, still .15 I think, but they added AGP 8X and fixed the banding issue. I think they also changed the FAA to fix a couple of errors.

The main problem with Matrox is they lost most of their engineers to nVidia and ATI. Im not sure if they didnt pay enough, or were just a shit company to work for. I hear the boss man (Dan Woods?) is a bit of an arse.

Ali - ex-MURCer
 
Parhelia probably basically proved to them that the cost of development to stay competetive was now more than they could realistically bear.
 
Didn't they have a refresh internally coded Pitou that they shelved after Parhelia's poor sales?
 
Parhelia simply didn't make enough money for Matrox to continue to fund the development of gaming cards. It's performance and features were simply not up to the competition. Given what Matrox would have had to have spent to compete with the behemoths of ATI and Nvidia, they simply dropped away from the market.
 
Randell said:
Didn't they have a refresh internally coded Pitou that they shelved after Parhelia's poor sales?

From what was reported on MURC, yes, but they shelved it "until they'd sold out their current Parhelia stock".

If Matrox were to have taken a gamble and gone ahead with it (or just fixed the major problems with the original Parhelia) instead of waiting to clear their stock, they might have managed to pull it off imo.

Quite a few of us where impressed about the features (mostly SurroundGaming, but still), but the speed left most of us lacking when it finally were finished. Though if they'd been able to get an "ok" amount of fps in surround gaming, I'd have picked one up.
Add to that the image they had of giving excellent IQ (;)) in 2d and you have something that could've gotten/kept a small part of the business. (Granted, not high-end, but high-mid-end possibly)

When that didn't happen, I'm, regretfully, agreeing with Dave that they found the market to expensive for their taste and that they're focusing on medical and military markets instead.

WRT to the engineers going over to ATI and nvidia, true, but, afaik, Matrox actually fired a lot of them when they closed down and restructured their organisation...

Nappe1> Isn't this your forté? It's been too long since I did any (real) reading of Matrox history...
 
One of the reasons I asked was that I was looking through my old Maximum PC magazines and re-read the Matrox Strikes back article again and I remember that the writer said that 80 million transistors (aka Matrox making a 512bit GPU) was way too many transistors for a .15nm process GPU, yet they expected somewhere around 250Mhz to 275Mhz but Matrox was only able to bring it out at around 220Mhz which really kills framerates.

They also said that idealy it would be better if Matrox were to do the Parhelia core on the .13nm process, but this is back in mid-2002 making kind of impossible.

Then I found out that Matrox's GPUs were fabbed by USMC, not TSMC and therefore I assume that USMC's process is not as advanced as TSMC circa 2002.

I understand that Parhelia was a flop as far as "Surround Gaming" was concerned but MPC leads me to assume if Matrox would have just done a die shrink to .13nm in 2003, maybe Matrox would have been in better shape specially in the professional market with a faster Dx8.1 card.

What I want to know is how was it that no one at Matrox could have though of doing that.

On that same note the first ATI Radeon and the Radeon 7500 core are basically the same only the latter was on a .15nm process giving really high clock speeds, I believe it was 260Mhz or so on the R100.

Nvidia did a similar thing the .22nm Geforce256, the .18nm GeForce2, the .15nm GeForce2 Ti and the mature .15nm Geforce 4MX, all the same GPU cores transistor wise.

I know that ATI and Nvidia were very fierce but still I can't help but wonder that Matrox had a chance and somehow they blew it and refused to fix the problem or maybe doing a die shrink would have been too expensive for them even if they left the 80 million transistors alone.

Either that or they just wanted to shoot themselves in the foot.

BTW I just wanted to mention that I am not saying I wanted to have matrox on equal grounds with ATI & Nvidia I just wanted to see them at least maintain themselves and eventually, like in late 2004 or 2005 be able to produce an update to keep in the race but not on the bleeding edge that belongs to ATI and Nvidia.
 
Akumajou said:
One of the reasons I asked was that I was looking through my old Maximum PC magazines and re-read the Matrox Strikes back article again and I remember that the writer said that 80 million transistors (aka Matrox making a 512bit GPU) was way too many transistors for a .15nm process GPU, yet they expected somewhere around 250Mhz to 275Mhz but Matrox was only able to bring it out at around 220Mhz which really kills framerates.

What would you expect some 2-bit writer dude at maximumpc to know about CMOS engineering and what is "too many" transistors for any particular process? The guy is typing out of his ass, there's no such ACTUAL limit. Considering ATi did chips on .15 with upwards of 50% more trannies than parhelia that should say something about the guy's claims.

Any real limit is more of a limit on die size, which will vary a lot for a fixed number of transistors, depending on how said transistors are arranged. The larger the chip, the bigger the risk of having a manufacturing defect crop up somewhere on it, and the risk of defects doesn't really (afaik anyway) correspond to the number of transistors on the chip. It's more of an area-based function.

Then I found out that Matrox's GPUs were fabbed by USMC, not TSMC and therefore I assume that USMC's process is not as advanced as TSMC circa 2002.

That seems like quite a stretch to me. Rather, any problems of the Parhelia is likely due to understaffing and lack of competence in their analog/digital design departments, rather than USMC lacking advanced manufacturing processes.

but MPC leads me to assume if Matrox would have just done a die shrink to .13nm in 2003, maybe Matrox would have been in better shape specially in the professional market with a faster Dx8.1 card.

What I want to know is how was it that no one at Matrox could have though of doing that.

What makes you think they didn't think of it?

FYI, direct optical shrinks between different processes are either not possible, or at least rather inefficient. It's not like putting a page in a photocopier and hit the minification button, the process of creating the masks needed for .13u manufacturing easily racks up many hundreds of thousands of dollars even if one doesn't hit a single snag. I'd think Matrox have thought of it and rejected the idea as not being cost efficient.

Nvidia did a similar thing the .22nm Geforce256, the .18nm GeForce2

GF256 isn't the same core as GF2 GTX.

.15nm GeForce2 Ti

Afaik, GF2 Ti is the same core as GTX, just rebranded along with slightly optimized production techniques.

I know that ATI and Nvidia were very fierce but still I can't help but wonder that Matrox had a chance

You REALLY like Matrox don't you? :LOL: Sorry man, their days of glory are over. They haven't been a real contender since the Millennium/Mystique days.
 
Actually, their G400 was an amazing peice of hardware. IF you had the MAX version then you had the fastest card on the market at higher resolutions, with only the tnt2 ultra beating it at the lower res's.

I still have mine in a computer, and ran it fromt he day it was released till the day I gave up on the Parhelia and bought one of the first 9700pros. In between there I bought a geforce 1 and returned it, and bought a radeon 7500 and put it in my wifes machine because neither of them were as good all-round as the G400.

Ali
 
Akumajou said:
They also said that idealy it would be better if Matrox were to do the Parhelia core on the .13nm process, but this is back in mid-2002 making kind of impossible.

That wouldn't have necessarily saved Parhelia. It also has problems inherant in its design, in that it was built with the wrong design considerations in mind for the period it eventually reached the market - it had massive multitexturing capabilities, but we were moving away from a fixed function multitexturing period into a more programmable period and the shader pipeline wasn't really that great. Curiously, it didn't really perform as great in multitexturing as its specifications would have suggested.
 
Going to a new process costs a lot of money and it's not worth it if the performance increase is not enough to increase sales. Parhelia trailed 9700 Pro quite a lot in terms of features and performance. Plus it likely would have had to compete against the 9800 Pro. If you consider shrinking Parhelia to .13um would also reduce the cost per chip the analysis becomes a little trickier.

Regarding what transistor count is possible on a particular process. It's probably theoretically possible to make a chip the size of a wafer, but that's not practical because the yield will be nil. When Maximum PC reported the maximum number of transistors at .15um what they really meant was the maximum number of transistors practical at .15um. What's practical depends on the situation. I remember 3dlabs said floating point pixel shaders weren't practical at .15um. Ati obviously disagreed and made a 100+ million transistor chip at .15um (9700 Pro).
 
to clear some mistakes up:
'pitou': was not the chip put on ice to wait sells of old parhelia. Pitou was parhelia 2 that would have had most likely DX9 feature set. 'pitou' was trashed as project way before tape out.

The Chip that was put on hold waiting for old parhelia chips to be sold was Newer Revision of Parhelia, including AGP 8x support, FCBGA package, new memory controller and fixed Stencil with FAA. This chip is now on sell and it quietly replaced the old parhelia. Matrox didn't make any announcement about it. and yes, it is considerably faster ('TI4600 killer') and it is what parhelia was ment to be in autumn 2001.

there's several posts about new revision parhelia by me as well as other fellow MURCers when the new chips started to arrive.
edit: as an example: http://www.beyond3d.com/forum/viewtopic.php?p=260957&highlight=parhelia#260957
 
Someone asked for a history? welll, here we go, enjoy the ride...

from PM to one indivdual in this board somewhere in late 2002
I maybe will release censored edition of this story later on... ;) it was pretty fun to write. :) and btw, you are the first one to read it, so there most likely are some mistakes.

M decided to take all the emplyees to see, what's on the wild side. Trip started nicely with picking up emplyees on the M bus model G2 and everyone was happy, because lot of work was made to take this bus to run. it's brand new 32 bit windscreen and 32 bit Binocular system made it pretty easy to drive. Sky wasn't never looked so good before from windscreen and Binocular system guranteed maximal viewing distance forward on the road. Bus' problem was that it's speed dropped too much when using new windscreen and Binocular system. Many other busses passed it busses with V2 decals as well as Busses with TNT decals. when they at last arrived, they were laughed at, because everyone else were there many hours ago.

Everyone get angry and decided that it was last time when M -bus is in the laughing stock.They took a bus back to garage and started planning. new M-bus would be the fastest. they re-tuned the engine and upgraded the fuel system. They also found nice Bumped hood (instead of flat one that everyone else were using.) from the old Sauna. Because the owner of sauna was scraping his bus project, they got easily a deal with M using same kind of hood on their bus. after they announced using Bumped hood, major bus body work maker, also licensed the bumped hood and announced full support for it on their next new body version called D6.

work on continued with great speed and before next "bus companies drive to the picnic", they finished first working prototype busses. instructions to technical part maker were sent about constructing new technical parts and brand new assembly line for this G4 Buses was opened. next "wild side" trip of bus companies came and this time M-bus was first on picnic place. all the other wer back, some more some less. sadly, that sauna owner was still strugling with his new Bus, that was rumoured run on methanol. In any case, this time M-bus was first on the place and after other arrived they came to take look of G4. it had many new interesting features like, two windscreens and Bumped hood. Everything was looking good.

Few months later things didn't look good anymore; technical parts maker still was not able to deliver and new assembly line was standing still. while waiting this, M decided continue selling G2 busses because they stiil sell pretty well for older ppl use. those customers weren't so much on speed, but more like how the scenery looked. TNT bus company already surpassed them with G-type on highway whenever they were going somewhere, though they didn't have bumped hood and dual windscreens. But at last, tech part maker started to deliveries and new manufacturing line started to run smoothly. Now because some competitors were again faster, started M updating once more the G2 based G4. but this time they decided do it properly. until they would be ready with G8, they decided quickly update G4 for better price/quality ratio for older ppl market.

At the same time, A -bus company started to gain speed improvments; they even tested dual engine type, but after major bus body maker (that same which earlier lisenced bumped hood from sauna owner.) made new body model, they dropped it because there wasn't room for two engines.

On the trip to Bus makers annual picnic on model G4.5, M's management happy and convinced that G8 would be next year able to attract more younger customers. Everything was fine until they arrived to picnic place and saw what was happening. TNT company had bring some hound dogs with them and before M's Management even noticed, those dogs were chased most of their emplyees on their bus. Now TNT was starting the engine and drive away. Of course M tried to go on pursuit, but G4.5 was just too slow to keep on with TNT's new GD - type. After a quick look, they noticed that practically whole G8 design team was gone. Only one guy was left from it. Quietly they drove back on their garage. situation looked very bad. the sauna owner made another announcement on bus magazine that he decided not release methanol bus, but continue it's development even further.

M had not much choices left. they decided that only one guy that knews G8, continues developing it. Later on it was launched as another older ppl model, G5.5. all the others and every employee that was hired, were put on work with new P-model bus. This time, it was made from scratch. Everything was going fine until first tech parts should have been arriving from new tech part maker. then something went badly wrong and it's not known even now, why it took tech part maker to make first working tech parts so late. they even missed one picnic because of that. In any case, during this time, now biggest bus companies A and TNT had lowered their prices quite lot. Now older ppl started to buy faster busses too, because there wasn't anymore much difference on scenery quality and lowest priced faster busses were cheaper than M's older ppl busses. M started to lose their money. and P -type was having serious problems on technical parts. 4 months later M at last got P-type finished and launched it to the market. P-type got good reception though there were some ppl thinking it being too underpowered to use new D9 body work that was coming soon. Press crushed it almost every test; tests only considered the top speed, while M tried to market Bus' Hill race features. engine had 4 cylinders and 4 valves on each. those were fed by only two carburators, but fuel pump and fuel pipes were highly optimized. also M's marketing made a kind of mistake by rushing bus out with bad air-cleaners. so usually, bus always losed to GD4 from TNT and R8 from A. later on, M got better aircleaners, but it was too late. youngsters market was already gone. A introduced soon their R9.7 with 8 cylinder engine (but having only one valve per cylinder. they new idea was use one valve more than once in the burning process.) Everyone was waiting TNT's new GX-type but it came very late and company said it had problems with tech maker, which was changing lathes more efficient ones to get higher quality parts.)

when M first time sent P-type construction instructions to tech part maker, they decided to arrange a new group of designers around that guy who finalized G8. they got their new group and these guys started to make new, more D9 body work compliant version of P-type. this newer P-type was finalized some time ago and first construction instructions were sent. after this it isn't quite clear what has happened. Still It is known that M decided to drive new P-type proto busses to scrap yard and make another older ppl model. BUT, scrapyard is pretty far away from M's garage and story tells that on way to there they picked up that sauna owner on ride. sauna owner had one more design try after that one he announced developing methanol bus further. Then he decided to change target market. Taxis were becoming quickly popular and it fitted more precisly on his qualifications. After discussing with the driver the sauna owner grasped the handbrake and said "hold on a buddy. pull over..."

How the story ends? I don't know, because rest part does not exist yet. I am now hoping the best, fearing the worst. in any case, that last chapter gives a pretty good insight what's happening behind the stage.

as it has been passed so long, it is cleary how the story ends. the hand brake most likely was broken and nothing changed. Sauna owner builds now Taxis and is making profit in first time in years. M-Company has decided to pull back on niche markets, with very special needs on busses.
 
that was a good story... kinda sad that the whole g8 team left but there we are... pfft those N guys are deadly.

There must have been something wrong with the bus company, that the whole team left though?!?
 
Druga Runda said:
that was a good story... kinda sad that the whole g8 team left but there we are... pfft those N guys are deadly.

There must have been something wrong with the bus company, that the whole team left though?!?

well, they "kind of" left... M did have lawsuit against N because of N using hound dogs, but they most likely settled it because it never went on court.

six months after the happening, N proudly released nScenery, which (suprise suprise!) had identical feature set to M's Dual Windscreen that had been used since G4.
 
tEd said:
How does matrox financialy? I'm actually surprised that they are still alive.

no problem being alive, they have such a strong foothold on special cards. (yeah yeah yeah, I know there's bunch of Quadros available, but QID and other parhelia special versions seems to keep them going.)

they have PCI-E support coming on next product, so even their R&D has something left. (don't ask when and what... I don't have a slightest idea.)

But do not expect them to come back to mainstream anywhere near future. competition is just too hard for them here right now. (but that also keeps the niche markets almost exclusively on them.)
 
Thanks Nappe1, you really nailed it.

Thanks for the information, its too bad that the Dx9 Parhelia was scrapped but now my opinion leads me to assume that indeed the chip fab Sucks with a capital S. I personally blame the chip fab for their incompetence, as well as some decision making on the M brass's part.

If only the chip fab would have been able to reach full production by December 2001 cause that would have given Matrox a good 6 to 7 months window prior to the launch of the R300 unless the press would have crushed them reguardless.

Then again I could start conspiracy rumours about N and A agents in the P's chip fab.

ATI's Rage 128 Pro Fury MaXX did not kill them.

Its too bad that like the story said, the press only believes in "speed instead of image quality" and proper 32bit color rendering with no missing colors (GF3ti & 4Ti)

Someone asked if I really like Matrox, yes I do, I liked the G400 Max I own alot, I also like the ATI Fire GL Radeon 8800 that I also own. I don't hate Nvidia but I have used GF3Ti & GF4Ti and disliked their image quality since I got the ATI card prior to knowing about Parhelia's launch I decided not to get the P since I had already invested my non volatile money.

As far as the current cards and my opinion, I think that they are great, I personally will not buy them untill after the R5X0 and Nv5X based GPUs are released.

I am hoping for Matrox to also have a product that can support Dx9 or eventually Dx10 as well as some other card, like a PowerVR based card or even 3dLabs or S3 based technologies to also be there just so we can have a nice little price war along with the technology war.
 
Parhelia doesn't have any kind of memory bandwidth saving technique - this made it DOA.

PS: I still have my Parhelia on my shelves somewhere. I'm thinking about putting together another PC for editing, it'll be perfect... :)
 
Back
Top