IMG sign up a portable gaming partner ?

[tangey]

http://www.imgtec.com/News/Release/index.asp?NewsID=412

In what has become common-place for IMGs news releases not only does it not mention the company, but it also doesn't mention the market segment.

However looking at the clues:-

"a new partner"

"major international consumer electronics company"

"another high-volume consumer device segment."

"forthcoming member of Imagination's POWERVR SGX graphics processor family."

There aren't too many"high-volume" markets available to IMG that it is not represented in currently, the only one of note is portable gaming consoles.

IMG currently have no relationship with either Sony or Nintendo, the "international consumer electronics company" fits Sony better than Nintendo in my opinion.

Interesting that its not a current member of the SGX roadmap, is this why th high-end device disappear from the roadmap ?

 

[Panajev2001a]

http://www.imgtec.com/News/Release/index.asp?NewsID=412

In what has become common-place for IMGs news releases not only does it not mention the company, but it also doesn't mention the market segment.

However looking at the clues:-

"a new partner"

"major international consumer electronics company"

"another high-volume consumer device segment."

"forthcoming member of Imagination's POWERVR SGX graphics processor family."

There aren't too many"high-volume" markets available to IMG that it is not represented in currently, the only one of note is portable gaming consoles.

IMG currently have no relationship with either Sony or Nintendo, the "international consumer electronics company" fits Sony better than Nintendo in my opinion.

Interesting that its not a current member of the SGX roadmap, is this why th high-end device disappear from the roadmap ?

I think it is Sony, I hope it is Sony.

TBDR's are the best there is when it comes to portable gaming IMHO...

The keys in that market are cost and Image Quality (for me at least ):

1.) TBDR's allow you to save on expensive external super-high-speed VRAM and thus you can more easily move to a UMA approach which makes for a simpler programming environment (PSP's logical-UMA was good too, but it requires two memory pools mobile DDR + e-DRAM) and easier data sharing between CPU and GPU.

2.) more than tons of polygons per second and layers upon layers of textures and custom shaders, portable gaming machines with relatively small LCD screens need to have very high quality pixels (AA is a must) as well as low ghosting (better to save money elsewhere and deliver screen-wise).

 

[Ailuros]

I think the seat for NINTENDO has been already taken, but could be wrong also. Last time they announced a licensing for a major consumer electronics company I wanted to bet on SONY yet it turned out to be Apple apparently after all.

Interesting that its not a current member of the SGX roadmap, is this why th high-end device disappear from the roadmap ?

Their recently updated SGX whitepaper ends with SGX545. SGX555 has been obviously canned and replaced by something we'll have to wait and see what it is. The latter was either too ambitious or too weak for such a project; or better phrased depends what would die size would be more reasonable for a handheld gaming device. 555 was to capture a wee bit more than 20 square millimeters under 65nm.

 

[Arun]

Yeah, that's very likely the license for Sony's PSP2. The problem with the SGX 555 might be that some of those DX10+ features are rather useless on a handheld and they do waste power (and obviously die area). So who knows, could effectively be a 'SGX 550', or it might be even higher-end than that!

 

[Panajev2001a]

Yeah, that's very likely the license for Sony's PSP2. The problem with the SGX 555 might be that some of those DX10+ features are rather useless on a handheld and they do waste power (and obviously die area). So who knows, could effectively be a 'SGX 550', or it might be even higher-end than that!

I wonder how PSP1 would be emulated .

I wonder about what kind of CPU they would use... next-gen ARM ?!? No more MIPS nor PowerPC?

IMHO, PSP1's BC will be software only and PSP2's main SKU might not even offer an UMD slot (maybe there could be a more expensive UMD based PSP2 with HW assisted BC or maybe they have a good software based solution already) and its compatibility list will be initially based at the PSP titles available on PSN.

With that said, SCE would be free to mix and match their preferred CPU and GPU and move to a full UMA solution (which could even simplify PSP emulation [logical UMA anyways... CPU and GPU can read freely in each other's memory]... PSP's GPU uses some GS-like tricks [e-DRAM could provide parallel frame-buffer read and write operations as well as a texture read], but its e-DRAM bandwidth is nothing crazy like the GS and I do not think devs have been doing uber-crazy tricks with it like they could with the GS).

.

 

[Ailuros]

Yeah, that's very likely the license for Sony's PSP2. The problem with the SGX 555 might be that some of those DX10+ features are rather useless on a handheld and they do waste power (and obviously die area). So who knows, could effectively be a 'SGX 550', or it might be even higher-end than that!

Since SGX545 is already 10.1 compliant, I severely doubt that they'd name something like that SGX55x-anything if it would end up with lower compliance.

Besides I as a layman can see more benefits to support as many as possible features on such a core than the other way around.

By the way honest question: what do you think would be a sensible amount of die area (for graphics) for any upcoming handheld device?

***edit:

I wonder how PSP1 would be emulated

I don't see any particular problems in that. While I can understand your point, I'm far more interested in possible PS3 to PSP2 ports

 

[Arun]

I wonder about what kind of CPU they would use... next-gen ARM ?!? No more MIPS nor PowerPC?

If they stick to MIPS, this is likely the core they'll use: http://www.mips.com/products/processors/32-64-bit-cores/mips32-74k/index.cfm#summary - I'm not sure what is the PSP's dmips/mhz rating, but being optimistic for the PSP: 1GHz vs 333MHz, 2.0DMips/MHz vs 1.5DMips/MHz -> 4x as fast (per core). Sounds good enough for a next-gen to me; of course they could decide to switch to ARM or whatever since the Cortex-A9 is also a very very good solution on paper.

but its e-DRAM bandwidth is nothing crazy like the GS and I do not think devs have been doing uber-crazy tricks with it like they could with the GS).

Here's the dirty little secret: it's not eDRAM, just on-package DRAM (i.e. more like Xenos really...) - link (including PSP die micrograph): http://www.semiconductorblog.com/index.php?tag=sram

Since SGX545 is already 10.1 compliant, I severely doubt that they'd name something like that SGX55x-anything if it would end up with lower compliance.

*cough* SGX 535 vs SGX 540 *cough*

Besides I as a layman can see more benefits to support as many as possible features on such a core than the other way around.

Tell that to, say, the R520 or NV30 engineers!

By the way honest question: what do you think would be a sensible amount of die area (for graphics) for any upcoming handheld device?

PSP's SoC seems to be ~50mm² based on the link above. Assuming ~40mm² for the non-IO part and assuming a substantial part of that to be non-3D, maybe ~20mm²? For reference's sake, SGX 530 in OMAP3 is ~5.5mm² (although I think it'd be even smaller on TSMC's process rather than TI's); so about 4-10 times as fast plus higher clocks would seem realistic to me on 45nm (with the catch ofc that just about any current smartphone platform is absurdly slower than even the lowest-end PC GPU). Although for power consumption reasons they might want to be more on the conservative side anyway...

 

[Ailuros]

*cough* SGX 535 vs SGX 540 *cough*

SGX53x = DX9.0L
SGX54x = DX10.1

While I know where you're getting at, it hardly has anything to do with compliance. The higher the second digit (up to now) the higher the compliance. If they'd name the anannounced high performance core "SGX55x" it logically couldn't be less than 10.1 compliant. Of course have we seen stranger things with naming scemes/codenames, but I'm merely following the up to now reasoning for their codenames.

Tell that to, say, the R520 or NV30 engineers!

LOL; while die area might be more important in small form factor devices (as opposed to high end GPUs), in the case of SGX it's more like a "yes we can" (within area budget X) kind of story

PSP's SoC seems to be ~50mm² based on the link above. Assuming ~40mm² for the non-IO part and assuming a substantial part of that to be non-3D, maybe ~20mm²? For reference's sake, SGX 530 in OMAP3 is ~5.5mm² (although I think it'd be even smaller on TSMC's process rather than TI's); so about 4-10 times as fast plus higher clocks would seem realistic to me on 45nm (with the catch ofc that just about any current smartphone platform is absurdly slower than even the lowest-end PC GPU). Although for power consumption reasons they might want to be more on the conservative side anyway...

20mm^2 on 65 or 45nm?

 

[Arun]

SGX53x = DX9.0L
SGX54x = DX10.1

Hmm really? My understanding was:
SGX 520/530/531: OpenGL ES 2.0
SGX 535: DX9.0L (or even DX10?)
SGX 540: OpenGL ES 2.0
SGX 545: DX10.1
SGX 555: DX10.1
i.e. the 5 implies it's not aimed exclusively at handhelds... I could be horribly wrong though!

20mm^2 on 65 or 45nm?

45, but as I said that's a very optimistic estimate (i.e. it's not because they could afford ~20mm² for the PSP1 that they could for the PSP2 on 45nm, since cost/mm² and power/mm² have both increased from 90 to 45 AFAIK...) - I'd say a minimum estimate that's still very realistic would be 4 TMUs @ 200MHz, while the maximum I could still believe is 8 TMUs @ 300MHz.

 

[Ailuros]

Hmm really? My understanding was:
SGX 520/530/531: OpenGL ES 2.0
SGX 535: DX9.0L (or even DX10?)
SGX 540: OpenGL ES 2.0
SGX 545: DX10.1
SGX 555: DX10.1

In terms of API support 535 is 9.0L and 545 10.1 "only". I've no idea if and by what means 520 has been castrated, but I think their baseline is SM3.0+ for all.

45, but as I said that's a very optimistic estimate (i.e. it's not because they could afford ~20mm² for the PSP1 that they could for the PSP2 on 45nm, since cost/mm² and power/mm² have both increased from 90 to 45 AFAIK...) - I'd say a minimum estimate that's still very realistic would be 4 TMUs @ 200MHz, while the maximum I could still believe is 8 TMUs @ 300MHz.

Considering 555 was supposed to be 20.3mm^2 @65nm with I suppose 8TMUs/8ALUs, if they'd go for roughly 20 square millimeters under 45nm it would result into a monster for a handheld. If my layman's math isn't off base it would result into something north of 16 ALUs. Sounds way too much IMHLO.

 

[Arun]

In terms of API support 535 is 9.0L and 545 10.1 "only". I've no idea if and by what means 520 has been castrated, but I think their baseline is SM3.0+ for all.

Ah yes, ignore me. For a second I had a brainfart and thought things like FP16 filtering were required for SM3.0; obviously that's not the case, and that is exactly the kind of thing that would be missing on handheld-oriented variants (unlike SGX535/545/...) - the ALU cores in SGX are probably DX10-level, what's missing would tend to be in the non-programmable portions of the chip.

Considering 555 was supposed to be 20.3mm^2 @65nm with I suppose 8TMUs/8ALUs, if they'd go for roughly 20 square millimeters under 45nm it would result into a monster for a handheld. If my layman's math isn't off base it would result into something north of 16 ALUs. Sounds way too much IMHLO.

Consider the fact that's likely pre-layout (add 10%+++) and doesn't include a display subsystem and might benefit from more power optimization (rather than area optimization), the real number would likely be nearer 15 than 10. So yeah SGX 555-level is pretty much what I'd expect and what I meant by 8 TMUs; perhaps in fact that's why it got 'canned' (lack of interest by others and Sony wanted to have exclusivity too)? Who knows.

 

[tangey]

There is an extremely vague reference to "series 6" PowerVr (Scheduled to be released in 2009) over on wikipedia, whether than is just notional or relates to any firm info is another matter.

http://en.wikipedia.org/wiki/Powervr#Series_6


Could some of the mods correct the title of this thread, my spelling mistake is startin' to look embarassing !

 

[Captain Chickenpants]

Haven't looked at that wiki page for a long while. Gosh we are in a lot of products!

 

[Ailuros]

Consider the fact that's likely pre-layout (add 10%+++) and doesn't include a display subsystem and might benefit from more power optimization (rather than area optimization), the real number would likely be nearer 15 than 10. So yeah SGX 555-level is pretty much what I'd expect and what I meant by 8 TMUs; perhaps in fact that's why it got 'canned' (lack of interest by others and Sony wanted to have exclusivity too)? Who knows.

I have that gut feeling though that whatever is going to end up in handhelds is not going to be the highest end variant of SGX. Besides that I honestly hope that if all the above speculations are close to reality that we'll see a tad of higher image quality on handhelds and that's obviously not as much in the hardware designer/engineer's hands. If IQ improving features X,Y or Z can be enabled without a significant cost, then developers should take advantage of it. There's no way to measure performance on a handheld; a significant difference in IQ will be noticable though....

There is an extremely vague reference to "series 6" PowerVr (Scheduled to be released in 2009) over on wikipedia, whether than is just notional or relates to any firm info is another matter.

That's what the old roadmap we've seen stated. I'd dare to speculate that it won't take all that long before they announce Series6. Eurasia was first mentioned in the Intel licensing deal in 2005 if memory serves well.

 

[Panajev2001a]

Here's the dirty little secret: it's not eDRAM, just on-package DRAM (i.e. more like Xenos really...) - link (including PSP die micrograph): http://www.semiconductorblog.com/index.php?tag=sram

Here is a downright pr0n secret (much dirtier) :

The CoC set-up (3D IC) happened with the 90 nm to 65 nm design transition for the PSP SoC AFAICS.

There are already cases where firms have made a major switch away from pursuing smaller rules and to 3D IC technology. Sony Computer Entertainment Inc (SCE) of Japan recently stopped working on developing smaller design rules for merged DRAM technology and instead introduced chip-on-chip (CoC) technology sandwiching two chips face-to-face with microbumps. The microcontroller for the firm's PSP handheld game system was originally fabricated using 90nm merged DRAM process technology. At the end of 2005, though, SCE switched to CoC, sandwiching the DRAM and logic together as separate chips.

One of the key reasons for the change in policy was that SCE was unable to find any way to make the chip cheaper by switching from 90nm to 65nm merged DRAM process technology. A source at the firm explained: "A massive capital investment would have been needed to drop the design rule to 65nm. And even if we had single-chipped with DRAM and the logic, each demanding a different manufacturing process, it would have taken a long time to get the yield up."

http://techon.nikkeibp.co.jp/article/HONSHI/20070328/129633/

merged DRAM-logic --> CoC

 

[tangey]

Just been advised that IMG have since Aug 12th added Navigation, STB/DTV and game consoles to the SGX C.V.

14th of august SGX "C.V." paragraph looked like:-

"The POWERVR SGX architecture is uniquely scalable, with IP (Intellectual Property) cores in the family targeting entry level handsets, through smartphone and MID, up to netbooks and laptops."

Today, IMG issued P.R. relating to SGX540 now being in silicon and the C.V. looks like:-

"The POWERVR SGX architecture is uniquely scalable, with IP (Intellectual Property) cores in the family targeting entry level handsets, through smartphone and MID, navigation and DTV/STB products, up to netbooks, laptops and games consoles"

Differences are
1) Navigation (IMG signed a licence with SiRF since 14th of August)
2) DTV/STB (IMG signed a licence with NXP since 14th of August, citing these markets)
3) GAMES CONSOLE
This can only be yesterdays licence announcement which cited "another high-volume consumer device segment"

Clearly the list just isn't aspirational, its stating segments where cores are committed by licencees.

 

[Ailuros]

Mod : Please add an explanation of what these links are! eg.
Have Sony bought into IMG to power the PSP's successor?


Fudo:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=10653&Itemid=1

EEtimes:

http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=212200099

XBitLabs:

http://www.xbitlabs.com/news/video/...ower_Sony_PlayStation_Portable_Successor.html

 

[Ailuros]

Sorry I hadn't seen the mod remark; yes I truly believe that EEtimes hit the nail on its head and SONY is the company that licensed an upcoming SGX variant for their next generation handheld. I've been waiting for something like that a very long time and I ended up with any vague announcement that didn't mention names sounding like a broken record: "is it SONY? is it SONY? is it SONY?"....

 

[Lazy8s]

The higher performance of this unidentified SGX variant could be the result of support for a wider bus like with the SGX531, as well as having more ALUs/pipelines.

While Sony feels like the mystery partner being mentioned, nVidia seemed to drop hints about a PSP2 contract and Nintendo is the console maker whose graphics partner recently quit the handheld business. Nintendo is one design win that Imgtec shouldn't let slip away, regardless.

The fate of AMD's handheld graphics IP will be interesting to observe considering both Freescale and Qualcomm still have licenses and NXP recently became an SGX licensee despite their business combination with STM.

The design wins for quite a few SGX variants which are getting closer to launch are still unclear. While SGX520 is probably being used by Renesas, at least, for their SH-Mobile handheld SoCs, SGX531 is far along with lead partners, and SGX540 was being demoed in silicon in Japan recently.

 

[Ailuros]

While Sony feels like the mystery partner being mentioned, nVidia seemed to drop hints about a PSP2 contract and Nintendo is the console maker whose graphics partner recently quit the handheld business. Nintendo is one design win that Imgtec shouldn't let slip away, regardless.

I wouldn't be in the least surprised if Nintendo's seat is already captured. Reverse your thoughts, it might be the other way around.

The fate of AMD's handheld graphics IP will be interesting to observe considering both Freescale and Qualcomm still have licenses and NXP recently became an SGX licensee despite their business combination with STM.

Question would be what any semiconductor or OEM manufacturer will have to do if it wants a high end chip for those markets. I think but could be wrong that there's only one variant from AMD and it doesn't look like a convincing high end sollution to me.

The design wins for quite a few SGX variants which are getting closer to launch are still unclear. While SGX520 is probably being used by Renesas, at least, for their SH-Mobile handheld SoCs, SGX531 is far along with lead partners, and SGX540 was being demoed in silicon in Japan recently.

I was puzzled at first why they released so many variants of SGX. It seems like each variant has been crafted for specific needs of various markets. With MBX I recall 4 variants (if you encount with and w/o VGP). If you include the not yet announced SGX55x for the handheld market the amount of SGX variants is 7.