PSP2 features - the handheld version *renamed

Looks like I got the rug pulled out from under me on this one.

I didn't know you meant new games. If they have to distribute digitally to begin with exactly what do they gain in selling a dual-platform binary?

And how about PS1 games that have been hand improved for a new platform? I don't know of any.

And there'll be more to GPU shaders than just transforming vertices (and lighting pixels) too.

Lots and lots of Android apps do run native code, Android off of ARM is probably not much of a starter at this point...

Fortunately developers do have access to the proprietary stuff, where performance matters for ARM talking about GCC is kind of irrelevant. The best ARM compiler is beating GCC by a margin as much as 2x in many apps. PS2 and PS1 are ancient history now and there are probably more devs experienced with the myriad of ARM devices than PSP.

Interesting. Is it lower numbers on the new part doing it?

 

I wasnt talking bout 2 binaries, rather an embedded configuration file for the "PSP emulator" (game works with antialising and higher res so use that)... or some stricter conformance rules on new games so those enchantments work.

If you are talking bout something like "Compute Shaders", I think thats still way off to gain big traction. And even longer on Handhelds where absolute performance is not the only concern but power-efficiency too. eg. running physics on a GPU is fast, but not really efficient for the power used compared to a CPU - compared to how much more efficient a GPU is at shading pixels at the very least. Putting a good vector extension onchip only takes transistors (rather laughable die-space nowadays), which might not be used but is preferable to pushing data back and forth to the GPU if it can be used.
Just try enabling Physics if you have a single Gfx-Card, usually you will prefer scaled down effects running on the CPU and a smooth experience. I think we wont see a liberal use of "Compute Shaders" until CPU and GPU are tied together very closely.

Err, I thought everything is bytecode at the app-level (the ones you buy in the store)?

There are proprietary compilers for PSP aswell (and hence MIPS), just that I have no experience with them, but I expect them to trounce GCC aswell. I think we are talking about different groups of devs here, Im more concerned about the big game developers and not 2$ app developers.
And I dont see how it matters much, the overall system and libraries will be the big stuff to learn, not the assembly-code the compiler spits out.

Err, Im not sure what you mean. the numbers are coremark 74K vs A8, the "new part" (1074K) is practically the same cores with some coherency-modules added - so perf/mhz should be the same or higher (addition of L2-Cache).

So for the 1074K all numbers I can come up with are:
74K - 2.5 coremarks/MHz From That link
1074Kf 0.36 mW/MHz (dualcore 40nm TSMC automated design, optimized for speed not powerdraw) Mips`Homepage

A8 has 2.4 coremarks/MHz and Im lacking power numbers.

 

The only place I've seen such a thing really happen was GBC, and that was largely because of the design being such a one-off from the original. I still say, if you're going to put any effort into making new games designed for old hardware run on new hardware it'd be effort much better spent doing a version for the new hardware.

I agree with you, but please consider the context in which I made the statement - nVidia believes in this sort of model, even for handhelds, and if they're the ones involved in graphics IP for PSP2 then they'll probably try to sell this - as is the case with them not including NEON on Tegra 2. The point I'm making is that if PSP2 is using an nVidia solution (like is currently rumored) they'll have a harder time emulating PSP due to not having VFP than due to not being MIPS.

I do think that for a handheld dedicating vector resources on the GPU and off the GPU on the level of PSP's VFPU is too excessive in terms of die space and power draw. Something on the level of NEON, if anything at all, is more appropriate. 3DS clearly doesn't believe in off-GPU vector processing, but that might be a bad example since they don't seem to believe in much in terms of CPU at all.

No, Google did make native options available, and the better apps tend to use it. Probably because Google was too late in coming in making Dalvik's performance acceptable.

We'll see - I just think that the ARM ISA has more to offer compilers than MIPS. I think it's kind of unfortunate that an academic ISA is being supported to this extent when it has pretty obvious deficiencies in the name of simplicity (allowing quick implementations to be made). But I might be overstating something merely from the perspective of someone who has had to write too much ASM in both.

We'll see; chances are the system development will be pretty standard. Not a lot of heavy OS interaction and pretty OGL like access to the GPU, like usual.

How is a comparison vs A8 fair? Compare with A9, the faster architecture. Or do you think Sony can't use it for some reason?

Coremark runs in L1 cache. Adding L2 won't improve anything.

 

First hand experience? Or are you just referring to the only public comparison I know of, the one from the FFmpeg guy?

Anyway my personal experience is that gcc 4.5 generates good code for ARM which, performance wise, is similar to a proprietary compiler. There are of course cases where the proprietary compiler will be significantly faster than gcc, but gcc will win as often.

 

and if its not much effort other than testing the game on PSP2?

Nvidia is not doing a "solution", Nvidia is providing their IP and Sony will manufacture it as they see fit. The GPU in PS3 is a custom tailored one with FlexIO, not because Nvidia already had some existing parts laying around but because Sony said "get us a part with FlexIO". I dont see how rumors and existing Socs play a big part in this.

I dont think diespace is the biggest concern, IMHO they are already pretty limited by powerdraw so that CPU+GPU will be rather small. Making the chip 1-2 mm^2 bigger probably isnt an issue - 3DS also gone (if rumors are true) into wasting diespace for 2 (relatively) slow CPUs instead of a single fast one.
I dunno, but a solid vector processing seems to be a logical thing to do if you make a device for media-processing.

Ok, dint knew that.

thats the strangest use of the word "academic" I heard in a long time.
the RISC concept as whole was academic once, while MIPS went pretty much everywhere in the last 2 decades.

I dont disagree but please look at the context. I said that up to and including the Cortex A8, MIPS had parity or advantage in perf/watt.

 

Sure, but if it doesn't pass the test? Of course, the real work is going to be on Sony to make these enhancements available, and they'd have to determine if it'll work enough of the time for vendors. If carefully engineered they can probably manage something, but Sony themselves would likely prefer vendors pushed new software that better sells the system.

Strictly speaking you don't know that, just because that's how it was on PS3 doesn't mean it'll be like that on PSP, assuming nVidia has a role to begin with - the rumors are specifically that it'll be "powered by Tegra", not "will contain nVidia GPU IP"

PSP's VFPU has got to be relatively huge (compared to the rest of the stuff on there) and it's definitely optimized for 3D. 3DS using two CPUs instead of a fast one is justifiable (in terms of power draw), Sony including a full blown VFPU over a weaker SIMD extension ISA, if anything at all, is a luxury that might not win them enough. Diespace isn't the biggest concern, but it does add cost for the manufacturer.

What I'm trying to say is that MIPS overemphasized simplicity for implementation purposes, and while the "RISC concept" might have been academic altogether the more relevant details of specific ISAs generally were not, save MIPS. The instruction set encoding isn't very efficient and lacks basic things like any kind of register + register addressing, which every other RISC ISA I know of has.

Yeah, my bad, I didn't even realize I said Cortex-A8 in the first place. I meant A9.

FFmpeg guy, but of course qualified by my mention of "as many as." My personal experience has continued to be that GCC is pretty easy to beat with hand-ASM in ARM, and it's pretty easy to see where it's deficient.

Do you have any comparisons that show GCC winning often?

 

The particular x2 speedup you mentionned is a failure of gcc to detect a widening multiplication (32b x 32b -> 64b) resulting in using 64b x 64b -> 64b. I've been told this has been fixed.

Obviously a talented assembly programmer will always beat a C compiler, provided he's given enough time

The end-user license agreement prohibits the publication of results.

 

rumors are just that, 3DS was supposed to have a tegra aswell, and the PSP2 having 4 SPUs. I dont think that there is any truth to it, because

• I expect a MIPS based solution (feel free to ignore that)
• I expect Sony to add fixed or programmable hardware for security
• Sony manufactures most or all of their high-volume SoCs (in shared fabs), I dont think that will change.

But I wouldn't be surprised if the GPU-Core is quite similar to the one in Tegra.

Yep, MIPS is probably the most "RISCiest" ISA architecture around, for the better or worse. But I think it has its merits aswell, given that it competes rather nicely even if you sometimes need an additional instruction over other architectures should tell you its no big issue. Personally I think the lack of PC-relative addressing and (long) jumps is annoying if you want to write PIC-code.

@Laurent06: given what I know gcc doesnt handles 8/16 bit variables well on architectures which only operate on 32bit, adding alot of unnecessary truncation instructions. I only have experience with gcc 4.3 & MIPS but the issue seems to sit pretty deeply.

 

No one can assure you that easily that NV was at some point of time negotiating with Nintendo, but I'm as certain as I can be that they were albeit it must have been quite a long time ago. "Off the record" it wasn't exactly that some folks could hide their pre-mature excitement.

No idea to the first two points and most likely right on spot for the last (which doesn't really take a crystal ball to guess).

Each GPU-core might have a few similarities to the Tegra GPU (but only if you're willing to oversimplify things to a ridiculous level). However if you want to start seeking differences start with the fact that Tegra isn't tile based.

Well for what it's worth both MIPS and IMG announced in the past a strategic alliance but it sounded in the end like something for a single deal with Sigma:

http://www.imgtec.com/Newsletters/with_imagination/issue8_10.pdf

Or maybe not:

http://technews.tmcnet.com/business...ic-alliance-with-imagination-technologies.htm

 

What are you saying? That PSP is using SGX543MP4 like the original (now quite old) rumors claim, or that nVidia is doing a tiler for them now?

Because the most recent rumor from SA is that they're using something Tegra-derived. It's possible that this is slanted towards an opportunity for them to sling mud at nVidia (not sure how they feel about Sony), but it all seems about as credible as any of the other rumors, and more recent is more likely, especially if we're to believe that dev kits are circulating.

Granted, SGX543MP4 sounds impressive, a lot more so than something Tegra 2-like. Hopefully if nVidia is involved they're shooting for more a preview of a Tegra 3-like GPU, and that's a more substantial improvement over Tegra 2 than that was over the original Tegra.

The whole rumor of Nintendo originally using something Tegra 2 derived themselves seems credible given all the citations and the sighting of older 3DS boards that had "TEG2" in their part numbers. Which makes it all the more bizarre that Nintendo ended up with ARM11s. Maybe this was part of a knee-jerk reaction to an nVidia solution having higher than expected power numbers.

I do think right now nVidia needs some big console vendor win to help stay afloat. They could have made hundreds of millions on 3DS, that's a pretty hard loss.

 

Well how are Nvidia-Sony relations currently?

 

Probably better than nVidia's relationships were with Microsoft over the original XBox, at least.

 

Isn't Sony moving away from MIPS? (PS2 IOP, etc)

http://www.qj.net/qjnet/playstation-3/ibm-talks-ps3-processor-backwards-compatibility.html

 

The first.

I can't know what's in the heads of the SA authors, but if they heard at some stage that NV was working or hoping on a handheld console design win, they could have speculated upon the rest.

http://www.xbitlabs.com/news/multim...Starts_to_Supply_PSP2_to_Game_Developers.html

Development of that SoC couldn't have started recently but several years ago. Then there's that link here:

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

Announced 2 years ago.

Only real insiders and especially at NINTENDO can know what was really going on. Is the current 3DS a collection of IP licenses and Nintendo has contracted a manufacturer to build the SoC for them? If yes then if NV intended or wanted to sell them the entire SoC themselves I could think of a few very good reasons why Nintendo backed out hypothetically in the end if and whenever that happened.

In hindsight why would any console manufacturer would want to have anything but IP deals for consoles or in extension handheld consoles anyway? And even more important if for NVIDIA licensing RSX IP to SONY for the PS3 didn't necessarily hurt them after all, why would it be any different with a handheld console deal?

Any market needs obviously the right approach; if anyone should try an alternative approach it naturally could involve quite a few risks.

I haven't heard anything about any problems between the two firms, but that wouldn't had stopped SONY to go for any other IP providing source if they felt they had something better on offer. If that "PSP2" ends up having a SGX543 MP then I wouldn't say that SONY made a wrong decision at all.

 

http://www.semiaccurate.com/2010/07/14/sonys-psp2-powered-nvidias-tegra-line/

Either PSP2 is using nVidia or someone who is otherwise credible is deliberately lying. There is no potential in this scenario for "Sony was considering nVidia at some point and the information got out", this is recent news. I do think the article makes a lot of assumptions (ie, "Tegra" is being provided, there are issues due to power consumption, etc), but it seems pretty evident that nVidia is contributing to PSP2, meaning they're supplying the GPU.

Why would PSP's current SoC have to have its GPU determind years ago when Nintendo was free to switch to PICA200 quite recently? The only actual new information from this is a claim from a developer that PSP2 will be "pretty powerful." That says next to nothing.

IMG's announcement is vague and could be referring to pretty much anything.

There's the usual rehashing of both the SGX543MP and Cell rumors, but the latter sounds particularly hard to swallow. The former would be nice but I'm not really convinced in light of the new nVidia information... Outside of that, isn't SGX543MP4 kind of a big sudden jump in power consumption for 3D? Wouldn't it be at least 4x over SGX540 at similar clocks? Not that that's really breaking the bank, since the SGX power numbers do seem really low, but it still seems like a big increase.

 

I'm willing to bet a good amount of money that there was never or will be any sign of Tegra in SONY's next generation handheld. I don't care where Charlie got that piece from and who and why is feeding him such nonsense (which he isn't really responsible for anyway), but here are a couple of questions regarding that entire story:

1. Why would SONY on God's green earth be interested in an "off the shelf" applications processor and not license IP? Because I can't read anything in that piece that suggests otherwise.

2. Regarding power consumption: perf/W aside the power consumption hints in that article sound like utter nonsense to me for Tegra2 with one single GPU core consisting of 2 FP24 PS ALUs, 2 VS units and 2 TMUs, while otherwise what has been proposed so far is a SGX543 4MP. That equals 4 GPU cores with 4 USC ALUs, 2 TMUs each. However you turn or twist it the latter scenario is going to burn more power overall even more so if it contains a quad core CPU too.

3. When console manufacturers investigate IP or whatever else for their future products they have a very specific performance target in combination with power consumption in mind. The ballpark between a T2 GPU and a SGX543 4MP is so big that either way I switch it it doesn't make sense. If we'd be talking about something like Tegra2 GPU vs. SGX535 or SGX540 things would be vastly different.

What's recently exactly? If NV ever was in negotiations with Nintendo then it must have been several years ago and rumors about that collaboration were pestering around ever since. But there was never ever any other speculation or rumor outside that.

I'm keen to hear all the possible scenarios. Remember that any possibility must have a very good reason for such a long term secrecy. And once we're at the secrecy point let's see why console manufacturers are so tight lipped about technology aspects of future projects and why it seems that Nintendo picked PICA on any hypothetical short notice.

I can think of a VERY good reason why SONY would want such a jump: SGX54x 2MPs appearing in smart-phones sooner than many would expect. Under that light wouldn't you say that a hypothetical 4MP would place a handheld in a far better position especially as time goes by?

Since 540 has 8 z/stencil units after all and the 543/544's have over twice the ALU throughput, 4x sounds conservative.

I can't of course know what SONY has in mind exactly, but:

http://www.xbitlabs.com/news/multim..._Specs_and_Launch_Dates_for_Nintendo_3DS.html

http://www.imgtec.com/News/Release/index.asp?NewsID=556

 

I agree, tegra always seemed unlikely.
Backwards compatibility seems like it'd be a high priority, especially after Go's failure.
At the very least, they'll want the ability to run everything on PSP's PSN store.
Though it defeats the point of BC when it only supports the extreme minority of both content and users.
Which makes UMD2/BLUMD seem all the more likely, especially since a gig of flash costs as much to make as a 50 GB bluray disc right now.

 

http://kotaku.com/5672410/psp2-will-be-bigger-out-fall-2011-currently-overheating

 

And another:

http://kotaku.com/5675525/psp2-will-ditch-the-umd-may-rival-xbox-360-in-horsepower

 

I don't know about any handheld rivaling the 360 in 2011, but it should be able to beat the wii.
Maybe they meant CPU-wise, which might be possible, but there is no way GPU-wise.