PowerVR Series5

[Rodéric]

Could you please guys stop talking about things you don't have any information about ?
There's no point discussing whether Naomi3 (is it its name ?) would be faster or slower compared to XBox2, since we have the specs of neither !

 

[Simon F]

Could you please guys stop talking about things you don't have any information about ?
There's no point discussing whether Naomi3 (is it its name ?) would be faster or slower compared to XBox2, since we have the specs of neither !

Don't be silly. Don't you know B3D is powered by arrays of thermocouples and all this hot air is making it chug wonderfully well....

 

[Simon F]

I copied this job description from the Imagination Technologies career page at the beginning of January this year:

I'd go more for the research area

 

[Ailuros]

Could you please guys stop talking about things you don't have any information about ?
There's no point discussing whether Naomi3 (is it its name ?) would be faster or slower compared to XBox2, since we have the specs of neither !

Agreed. A performance/feature etc. guestimate from you Monsieur?

***edit:

Simon,

I just saw the Queen's award announcement. Was there/is there going to be some kind of dinner? I'd love to hear the details *snicker*

 

[Kristof]

I copied this job description from the Imagination Technologies career page at the beginning of January this year:

I'd go more for the research area

Naaah, I heard that team leader eats all the chocolates...

 

[JohnH]

Jodi,

Did you quit your job over there, or did they fire you?

I left to get married in NZ. I Loved my time at PowerVR, but I love my wife more ;-)

Obviuously somone forgot to put the bromein in your water ;-)

 

[JohnH]

I'm surprised everyone here is so pro TBDR. As time progresses I believe its advantages become less and less.

Addressing each one at a time:

Here are the problems I see with it:
+ Both Nvidia and ATI are convincing developers to use a zfill pass before doing any complex pixels. For games that do this a TBDR will have almost no advantage over a traditional raserizer with fast early z/hiz.

It is a common misconception that TBDR only offers a performance benifit where overdraw is present. This completly ignores the considerable BW advantages of keeping all FB read/modify writes on chip e.g. non of todays IMR's are able to hit anywhere near there theortical peak fill rates with simple translucency.

+ As geometry complexity increases creating an efficient TBDR becomes increasingly difficult/impossible.

Difficult yes, impossible no. We've been doing this a long time, we know what we're doing.

+ Pixel shaders and vertex shaders that modify depth make TBDR more difficult.

Why would you think this is any harder on an TBDR than an early Z IMR? Basically its a pain for everyone, and is bad for everyones performance, in fact more so for an IMR as it makes it rather difficult to apply heirarchical depth buffering.

Now to be fair I could list off some positives, but since everyone seems so supportive of TBDR you must already know about those so I won't go there.

Let me do an obvious one for you, where exactly do think an IMR would put the memory for 64bpp (HDR) 1024x768 64x FSAA'd image?

John

 

[Panajev2001a]

non of todays IMR's are able to hit anywhere near there theortical peak fill rates with simple translucency.

Cough... Graphics Synthesyzer... cough...

Hey, I am a fan of fat 2,560 bits data-paths .

I know your point will focus on having to store a HUGE frame-buffer for good AA and that will waste too many transistors, but intelligent tiling and primitive gropuing is not an exclusive rights of TBDR GPUs .

http://makeashorterlink.com/?P2EF36518

 

[Ailuros]

...64bpp (HDR) 1024x768 64x FSAA...

Dumb question: for N amount of samples a N*N grid?

 

[JohnH]

non of todays IMR's are able to hit anywhere near there theortical peak fill rates with simple translucency.

Cough... Graphics Synthesyzer... cough...

Hey, I am a fan of fat 2,560 bits data-paths .

I know your point will focus on having to store a HUGE frame-buffer for good AA and that will waste too many transistors, but intelligent tiling and primitive gropuing is not an exclusive rights of TBDR GPUs .

http://makeashorterlink.com/?P2EF36518

Hehe. Yes there's always embedded memory, where you're prepared to accept constraints on resolution and degree of AA applied.

John.

 

[JohnH]

...64bpp (HDR) 1024x768 64x FSAA...

Dumb question: for N amount of samples a N*N grid?

That 64 samples total, or any no of samples from 16x up really to make IMR's choke up their available memory.... Sorry that should read "any number of samples other than 1 really makes an IMR choke up its memory!"

John.

 

[Panajev2001a]

non of todays IMR's are able to hit anywhere near there theortical peak fill rates with simple translucency.

Cough... Graphics Synthesyzer... cough...

Hey, I am a fan of fat 2,560 bits data-paths .

I know your point will focus on having to store a HUGE frame-buffer for good AA and that will waste too many transistors, but intelligent tiling and primitive gropuing is not an exclusive rights of TBDR GPUs .

http://makeashorterlink.com/?P2EF36518

Hehe. Yes there's always embedded memory, where you're prepared to accept constraints on resolution and degree of AA applied.

John.

You had to put the final jab now didn't you ?

Now, be a good JohnH and follow the link .





BTW, I think IMRs + e-DRAM and TBDRs without e-DRAM are both valid approaches in theory so I am not trying to proove the TBDR solution has to suck.

x86 sucked since its inception, but few thought at that time that CPUs like the Pentium 4 ( Northwood ), the Pentium-M ( Banias ), the athlon 64 ( K8 ), etc... would have ever come to be.

 

[JohnH]

Thats one long link! Think I've heard about that one before, kind of like tiling really!

 

[Ailuros]

I understood your point perfectly well; my question was going elsewhere and that namely in the grid pattern direction.

I'd probably should ask first if past a certain amount of samples (let's say 16x) a sparse grid really makes any significant difference anymore, yet I can theoretically see a very high edge equivalent resolution even on just an 8x sparsely sample MSAA pattern for example ( 8*8 ).

Clearly the more samples the better, yet I don't think the resulting grid is irrelevant after all.

 

[Ailuros]

BTW, I think IMRs + e-DRAM and TBDRs without e-DRAM are both valid approaches in theory so I am not trying to proove the TBDR solution has to suck.

I didn't understand it otherwise; question would be whether by the time e-DRAM becomes "affordable" in larger quantities, why a TBDR couldn't theoretically use it too.

 

[Gubbi]

BTW, I think IMRs + e-DRAM and TBDRs without e-DRAM are both valid approaches in theory so I am not trying to proove the TBDR solution has to suck.

I didn't understand it otherwise; question would be whether by the time e-DRAM becomes "affordable" in larger quantities, why a TBDR couldn't theoretically use it too.

Indeed. Use it for a big low latency texture cache to boost those dependent texture reads.

Cheers
Gubbi

 

[Panajev2001a]

Thats one long link! Think I've heard about that one before, kind of like tiling really!

That is why I made it short .

Edit: you can do tiling too ( this is a new patent, btw ) if you divide bricks so that they do not over-lap at all.

This will cause you to waste a bit of RAM having to keep data for all the different tiles you are going to send to the other parts of the rendering engine, but it will save frame-buffer space.

You can mix tiling and geometry grouping: take bricks which do overlap ( super-bricks ) and then subdivide them in bricks which do not over-lap and process super-brick after super-brick.

The Network based rendering is another sweet point pf that patent, IMHO.

I hope you like the patent .

 

[Panajev2001a]

BTW, I think IMRs + e-DRAM and TBDRs without e-DRAM are both valid approaches in theory so I am not trying to proove the TBDR solution has to suck.

I didn't understand it otherwise; question would be whether by the time e-DRAM becomes "affordable" in larger quantities, why a TBDR couldn't theoretically use it too.

It could, of course, but it does not seem that they are pursuing it yet.

I am not aware of the R&D IMG Tech. has put behind e-DRAM, but given what they need for their designs I would not think they are investing into it that great amount of money as they simply do not need it that badly: they have other concerns which are more pressing ( geometry binning, Vertex and Pixel Shaders design, etc... ).

They could use some e-DRAM to cache textures and so on, of course.

 

[Simon F]

Simon,

I just saw the Queen's award announcement. Was there/is there going to be some kind of dinner? I'd love to hear the details *snicker*

I think that is for META so I wouldn't know if there is a dinner.

However, we won an award for DC/CLX a few years ago, and I was one of the lucky few who got to go to the palace. It was quite an amazing experience.

 

[Panajev2001a]

[double post]