PowerVR Series5

[aZZa]

Most of my thoughts & ideas came about from speculated rumours and roadmaps. If the Kyro 3 was to be released around geforce 2-geforce 3 time, i dont think pixel shading was needed, like the situation with ps3.0 and nv40 & r420 today. If it was to come out in the geforce 4-radeon9700 era i think that shaders would have been required more for the longevity, as they havent become a prerequisite until recently. As an integrated solution i dont think shaders would have been a worthwhile requirement up until the last six months, where games have started really using them, and only ati offers a dx8 solution.

Adding shader 4.0 or 5.0 tech to a chip is not going to be much use until everyone can catch up and implement it in software/programs. Im sure this is why ati are striving for performance first in the coming chip(s). Once the games become available the hardware will no doubt show up in droves.

I dont think the real issue was gf4mx not having pixel shading tech it was more the labelling it as a geforce 4 variety. I have a friend who cant run Prince of Persia...but he says he has a geforce 4 (mx)... This was the real problem. The better budget card was a radeon9000 (i know), because it had more features (and better overall performance)... but can one say a gf4ti is not better than a gffx5200 for gaming because it has less features??

There has to be compromises, and for the smaller companies performance must be as if not more important than trying to drive new features to the market. If these features dont become adopted (or not straight away) as in the past by the competition as with EMBM or displacement mapping etc then why not use these transistors for speed enhancements.

 

[Dave B(TotalVR)]

On shaders:

You have to remember that game developers cant develop games using pixel shader hardware when they have no pixel shader hardware. Game development cycles are bloody long nowadays, how long have we been all waiting for Doom3? and Half Life 2....

 

[aZZa]

I think having powerful enough shaders is a must though. One must give congrats to ati for producing the radeon 9500-9800 series. These allow users to run the games with the 2.0 shaders enabled. The low-main to high end cards offer reasonable to good performance especially for a first generation product series for dx9, something the fx5200 and 5600 series cards do not seem to produce.

Its hard to get the right balance between performance and features, and hopefully S3, XGI, ATI, PowerVR, Nvidia etc can work with developers to bring out/develop a minimal standard to shader performance (an idea for next 3dmark??) so people dont get stuck with another generation of checkbox features which will go unused or fail to perform well enough to be useful.

 

[Teasy]

At a completely uninformed guess, I would say TSMC.

I'd guess NEC, after all NEC produced the GPU for Naomi 1, 2 and DC. So Sega and IMGTEC have a very good relationship with them (especially in the arcade/console market).

 

[Roger Kohli]

Do we know who actually licensed series 2, NEC or SEGA? I had the impression that it was SEGA, but if so, how did the PC version of the chip come about, surely SEGA wouldn't have paid for the development of that?

Can we draw any conclusions from that previous agreement and the latest IMG/SEGA tie up?

 

[Pottsey]

â€￾KYRO II was a good move, there was supposed to be a KYRO II SE but that never happened because STM pulled out the market.â€￾
I thought the Kyro II SE did happen. It made it out in shops in very small numbers and reviewers also had final cards. At lest I remember the old Pky website posting news about a shop that had some in stock in limited numbers.

Personally I think the Kyro II should never have come out and the Kyro II SE should have replaced it. Kyro II SE was too late with too little difference over a Kyro II.

 

[Ailuros]

Series4 was designed as a mainstream product from what I understand it, while Series3 was rather aiming for the value/budget segment.

To avoid misunderstandings I had something entirely different in mind; after the initial KYRO I would had prefered to see in early 2001 the 4 piped STG5000 aka K3 even at a meger 166 or 175MHz. That one later on pulled through a die shrink and at 250MHz alongside with a KYRO2 for the lower end.

You people are concentrating way too much on raw numbers like clockspeeds, amounts of pipelines etc. One generation away from Series3 indicates more than just a few changes and there's no guarantee either that fill-rate=fill-rate between those two. For early 2001 and instead of KYRO2, a same clocked KYRO3 would had been good enough.

Well, there is no denying the level of sucess of the KYRO series, they sold loads of em. They just didn't follow up. KYRO II was a good move, there was supposed to be a KYRO II SE but that never happened because STM pulled out the market. Then KYRO III could not be built at all (i.e. nor ith another AIB manufacturer) because of licensing issues with STM. They wanted royalties.

2Million units in 2001 or more? The result might had been satisfactory for PowerVR/IMG but that doesn't mean that things couldn't had been way better too. The higher you set your targets the better chances you have to succeed. ST Micro didn't obviously have high targets or ambitions which shows not only from the delays but from the amounts invested.

Scroll up and re-read my paragraphs above your quote. 4x times the fill-rate is 4x times better and that will T&L on board and cube map support. Include the other improvements and they also would had sold times more units too.

They sold buttloads of KYRO's. Thats a successful product, it was only a budget card.

As I said for IMG's expectations that might even stand a chance, yet the "buttload" description is still a vast exaggeration. Let's keep things on a level here. ATI and NVIDIA sell "buttloads" of value products, even more if you include the OEM market.

 

[mboeller]

it seems that's now the "How I would have liked the PowerVR-Kyro History" Thread. So here is my entrance to the competition :

Kyro1:
Release March 2000 @ 133MHz instead of 115MHz; would have made the Kyro1 really competitive with the GF-SDR and the GF-MX later on.

Kyro2:
Release October 2000 @ 200MHz with maybe 18Mio Transistors @ 0,18µm technology ( a little high I know, but it's my "History" ). Fully DX7-compliant but without an T&L-unit; slightly enhanced vertex-handling, other tweaks like improved AF.

Kyro2-Plus
Release October 2000 @ 250MHz ( 18 Mio @ 0,18µm ) with an Elan-based T&L-unit @ 125 MHz or 250 MHz as an companion chip. Fully DX7-compliant; slightly enhanced vertex-handling, other tweaks like improved AF. This chip would have been better than the GF2-Ultra; so this would have been an really highendchip!

Kyro3:
Release October 2001 @ 350 MHz ( ~40 Mio ; 0,15µm ) with 2 fully complaint DX8.1 (PS1.4) Pipelines ( MSAA; Trilinear for free; 16bit per component ) and an integrated Vertex-Shader Unit with 2 Vertex-Shader. The complete design would be enhanced to handle large amounts of polygons efficiently. Memory : 128bit DDR-RAM with 233MHz (should be enough for an TBDR, and the 4/3 offset should handle the DDR-efficiency defizit well )

Kyro3-Mainstram :
Release October 2001 @ 275 MHz ( ~40 Mio ; 0,15µm ) with 2 fully complaint DX8.1 (PS1.4) Pipelines ( MSAA; Trilinear for free; 16bit per component ) and an integrated Vertex-Shader Unit with 2 Vertex-Shader. The complete design would be enhanced to handle large amounts of polygons efficiently. Memory : 128bit DDR-RAM with 183,3 MHz (should be enough for an TBDR, and the 4/3 offset should handle the DDR-efficiency defizit well )

So I hope you like it

 

[Dave B(TotalVR)]

You hav to understand that a lot of descisions were made with regards to the specs of the KYRO to make it as cheap as possible.

For example every KYRO has a locked ram/core speed. Moebeller just suggested they have a 4/3 ratio on the KYRO III were it released. Well a synchronised memory controller is a damn sight simpler than one that is asynchronous and able to be set at arbitry speed ratios (which would have been demanded too for overclockers). Think about the complexity involved of sending (in the case of a 4:3 ratio) 24 bits of information (say 128bit chunks) which would be sent over 32 clock cycles if synchronous but instead you have 32 memory clock cycles to send it over. Sure you can do it easy but how do you use that bandwidth efficiently? You are going to have to skip memory clock cycles so the information is ready from the core. Sure you could send out other data but then you will bugger up your memory efficiency due to page switching.

In short it was just not worth making the bus asynchronous. The same is true for anisotropic filtering. It was only when the KYRO II was old and probably not selling anymore that aniso became big hits. KYRO could do it, just slowly.

Also, the TnL unit absence on the KYRO II didn't hurt its performance so much, it just hurt its street cred (so to speak). At the time of release and for time afterwards games were being held back by fillrate, not TnL power. This was clearly evidence when Nvidia's dagoth moor zoological gardens ran faster on the KYRO II than on the geforce it was written to show off. The only thing was the KYRO didn't support the cubemaps used on the little bugs that walked about (boohoo, I stamped on them anyway).

Oh and MSAA? IMO IMGTEC has no excuse for not introducing this earlier, although the KYRO III was supposed to sport this IIRC. Speculation on saying KYRO III should have come in place of KYRO II is a tad acedemic though because we dont know what the specs of KYRO III were. Hey it could have been a super budget card;p.

For KYRO III though, I suspected.
150-180Mhz.
the pipes with free trilinear on KYRO II split into 4 without free trilinear ala geforce 2.
DX7 TnL unit (yes fixed function)
No pixel shaders.
128bit synchronous DDR RAM
MSAA (free(ish) AA)

Not fantastic because its behind on features, but it would be pretty damn fast. At 166Mhz that would be 666Mpixels of fillrate, with free AA thats 1280x1024 gaming at 60FPS with 16x AA turned on, plus anisotropy.

But hey that was just a guestimation, still dont know how right I was...

Dave

 

[Ailuros]

We are close to delivering our very high-end next generation graphics technology that targets arcade, PC and console. We have already licensed this technology to SEGA for use in arcade systems and intend to explore relevant partnerships for other markets.

http://www.imgtec.com/Investors/HTMLResults/Prelims04/index.asp

Uhmmmm what's the definition of "close" in british terms exactly?

 

[Bjorn]

We are close to delivering our very high-end next generation graphics technology that targets arcade, PC and console. We have already licensed this technology to SEGA for use in arcade systems and intend to explore relevant partnerships for other markets.

http://www.imgtec.com/Investors/HTMLResults/Prelims04/index.asp

Uhmmmm what's the definition of "close" in british terms exactly?

Interesting to say the least. And what do they mean by deliver, perhaps a demonstration card to send to Dave B ?

 

[Dave Baumann]

Interesting to say the least. And what do they mean by deliver, perhaps a demonstration card to send to Dave B ?

I guess I could live with that. Heck, they're oly a couple of junctions on the M1 and a junction on the M25...

 

[Tim Murray]

Interesting to say the least. And what do they mean by deliver, perhaps a demonstration card to send to Dave B ?

I guess I could live with that. Heck, they're oly a couple of junctions on the M1 and a junction on the M25...

Dave, ignore the black van in front of your house. And the black helicopters that follow you when you leave. And the SWAT team in your closet. They are only there for your own protection and are not actually mercenaries I hired in order to seize your Series 5 sample.

 

[Hanners]

I guess I could live with that. Heck, they're oly a couple of junctions on the M1 and a junction on the M25...

So, they're a few days drive away then...

 

[Jabbah]

We are close to delivering our very high-end next generation graphics technology that targets arcade, PC and console. We have already licensed this technology to SEGA for use in arcade systems and intend to explore relevant partnerships for other markets.

http://www.imgtec.com/Investors/HTMLResults/Prelims04/index.asp

Uhmmmm what's the definition of "close" in british terms exactly?

um, well, the earth is close to the sun, and ive been close to finishing my project for the last 3 months something to do with browsing these forums maybe.

I dont know why, considering my nightmare experience with the pcx1, but I really want to see series 5. I have a certain fondness for its architecture, most intriguing. However when they say they are "intending to explore" that sounds like saying "at some point we may ask if anyone wants to make a chip with this stuff, if were bored and nothing else to do"

 

[aZZa]

pure speculation...

I wonder if Sega/Sammy might become involved with other partner(s) and attempt to bring some graphics hardware to the pc which could possibly be bundled with some sega software/ported titles as a means of attracting some consumers, and help push the new arcade Tech they now have licenced. I guess they could make extra money on the hardware they will use in the arcades, re-using partially buggy chips (ala rad 9500,x800pro etc) and get the pick of the crop to use in their own machines.

Its a long shot, but if their partners can make chips compatible to multiple segments, maybe disabling certain specific functions/parts to suit each market, im sure bundling a few sega titles might help attract customers to a new performance pc card.

Ok this might not even be anything close but say Sega/and/or partner did ... (with such an architecture it may not be arranged it in tradidtional quads)

16 pipes ....ARCADE (best chips, maybe a limited edition review version)
12 pipes ..high-end pc (slighty buggy, partially disabled)

8 pipes ...performance pc (either buggy-disabled or new chip)
followed by a much lower-cost high-volume
4 pipe mid->low-end as a mainstream offering.
(Or this might simply apply to an r300 type design 8 pipe/4 pipe chip)

I think this would require 2 chips to be developed either way (16/12/8 & 4) or (16/12 & 8/4) with necessary redundancies to improve yields to be commercially successful on the PC. They could attack mainstream and low-end (k1,k2 etc), but without a performance part its hard to get the message across to the general public who only hear everything is either geforce or radeon.

Reusing buggy arcade chips might be a useful way of making some extra $$$, but then again these chips might be a completely uncommercial product for the home user and even be incompatible. Another partner might step up and become involved in series 5 if they arent already. I guess only time will tell.

 

[psurge]

From the link above...

Amusement - We have made continued progress in the amusement market with existing customers Aristocrat and IGT as they ship advanced products using our graphics technology. Developments with Renesas/Sammy, and the latest SEGA partnership are expected to lead to volume shipments in the amusement machine market starting from the second half of this year and early next year respectively.

 

[Dave Baumann]

pure speculation...

...

Yeah, you're right. Pure speculation.

 

[Bjorn]

5 partner devices in or close to production (compared with only one in production at the beginning of FY03/04) and 2 in prototype silicon;

Interesting.

 

[Teasy]

Uhmmmm what's the definition of "close" in british terms exactly?

There isn't one in British terms since there is no such thing as the British language..