Android x86 phone

I haven't looked into this, but a Tegra3 is faster than a PS3? Really? Wow. I always knew the Tegra3 had superior quality (Anandtech compared the ASUS TF201 to the iPad 3 and concluded that the TF201 had better texture rendering albeit at a slightly slower frame rate), but didn't realize it was comparable to a PS3! I was thinking more like a PS2, but I never played any games on it.

Last week I bought an HDMI adapter so I can hook up my ASUS Transformer Prime TF201 (with NVidia's Tegra3) to my TV. I did this to stream movies (the PS3 won't play Cinavia watermarked content, but Android will :D). I also paired up a PS3 controller with it, which surprisingly works well over bluetooth. So ya, it seems like that tablet is all I need to surf the web, play games and do whatever on my big screen TV. My PS3 should be worried.
 
Ah yes, you're right. The next xbox is rumored to have an AMD GPU which got me confused there. Still, Sony's experience with the CELL processor is likely what's making them consider x86.

If it uses an AMD processor it'll only be the control processor - their tech boss talked about something a lot more complex that sounds to me like a CPU surrounded by GPUs, Cells and an FPGA thrown in for good measure.

A traditional processor will let them get stuff running quickly but like the PS3 and 360 you have to do work to make it fast.

And the xbox is rumored to be getting a 16 core PPC. From what I know of game development, multi-cores usually isn't what you want. So that does surprise me a bit.

The Xbox dev kits supposedly contain a 16 core PPC.
 
given the graphical capabilities of Tegra 3 (faster than a PS3)

The mobile GPUs might be similar but they only have a fraction of the memory bandwidth and that's shared with the CPUs.

OTOH even a quad core Tegra won't match a single SPE so it has quite a way to go to match Cell. The mobile GPUs are a lot closer but even they wont match it yet you'll need a high end next gen part to match it - but it will have the same bandwidth problem.

All that said mobile GPUs are getting pretty impressive and the next gen even more so. The same goes for the processors.
 
The next xbox and PlayStation will be x86 based. Surprise surprise, the PS4 will likely be AMD based. Turns out CPUs aren't that important anymore, GPUs are.

For mobile this is very true, which is interesting given the original topic of this thread. Intel doesn't even have their own mobile graphics.
 
If it uses an AMD processor it'll only be the control processor - their tech boss talked about something a lot more complex that sounds to me like a CPU surrounded by GPUs, Cells and an FPGA thrown in for good measure.
Sony took it in the shorts with the PS3. One way to look at it is the PS3 ate all the PS2 profits. That's not a position Sony is going to want to maintain. Heck they can't. Techincally there are no PS3 profits to eat. I think Sony has some hard choices ahead. Either go Cell 2 - perhaps a 16 PowerPC core. Or risk a blog standard x86 and a whole new architecture, which in the Sony world has killed backwards compatibility in the past. Part of the problem I see is Sony's strength was more so the hardware ideas. Gaming has mostly moved away from hitting the hardware directly. Which in turn means the CPU is less important.

Microsoft is in a better position. Their strength is software. And they're actually fairly CPU neutral. The NT core ran on PowerPC, Alpha, along with X86. Now we're seeing that same core run on ARM. The 360 had much better backwards compatbility than the PS3 due to this ability. (Sony has to include a PS2 inside the PS3.)

The Xbox dev kits supposedly contain a 16 core PPC.
Didn't early 360 dev kits contain an Intel CPU? Though I know the Macs were used too as the time neared.
 
The mobile GPUs might be similar but they only have a fraction of the memory bandwidth and that's shared with the CPUs.

Tegra 2 is capable of putting out 90 million triangles/sec verses PS3's 250MT/s, Tegra 3 is five times faster than Tegra 2 in it's current incarnation. Obviously the more complex the scene, the more the memory speed difference will show. But be under no illusions as to the capability of Tegra 3.

OTOH even a quad core Tegra won't match a single SPE so it has quite a way to go to match Cell. The mobile GPUs are a lot closer but even they wont match it yet you'll need a high end next gen part to match it - but it will have the same bandwidth problem.

If you're talking the full package verses a single SPE, depending on workload and more precisely how well tuned the code is for the SPE, a Tegra 3 could well outpace it for more general purpose tasks. One of the single biggest issues with the Cell was and is an absolute PITA to get any performance out of it. Given the right task it absolutely flies, try to get it to do something outside of one of it's party pieces and it'll leave you with a rapidly receding hairline.

All that said mobile GPUs are getting pretty impressive and the next gen even more so. The same goes for the processors.

Indeedy. If someone 15 years ago had told me I'd have a colour LCD device that could do ps2 graphics, had multiple cpus (ok, multi core), run for 8 hours and have gigabytes of storage I'd have laughed at them.
 
Having a Tegra Zune I love that thing. I'm interested in a Tegra4. It's supposed to ship in late 2012 and if it has Windows 8 on it I may have a companion to my iPadHD.
 
Tegra 2 is capable of putting out 90 million triangles/sec verses PS3's 250MT/s, Tegra 3 is five times faster than Tegra 2 in it's current incarnation. Obviously the more complex the scene, the more the memory speed difference will show. But be under no illusions as to the capability of Tegra 3.

Unfortunately triangles per second is even less useful than MIPs for measuring performance. I wouldn't be surprised if it's can match the PS3 in numbers but reality can be rather different. The GPU may be more advanced with better shaders though.

That said, Tegra 3 appears to have one of the weakest mobile GPUs, it does pretty badly in benchmarks. Tegra uses immediate mode rendering - this is the completely wrong solution for mobile as it is completely dependant on bandwidth. All the other GPUs are deferred, tile based so hit the memory a lot less. AFAIK Tegra is moving to the same approach in the future.

If you're talking the full package verses a single SPE, depending on workload and more precisely how well tuned the code is for the SPE, a Tegra 3 could well outpace it for more general purpose tasks. One of the single biggest issues with the Cell was and is an absolute PITA to get any performance out of it. Given the right task it absolutely flies, try to get it to do something outside of one of it's party pieces and it'll leave you with a rapidly receding hairline.

I worked for a pre-startup who were going to build Cell workstations. I spent a year reading everything about Cell there was to read. Almost all of the comments about Cell being difficult to program came from people who never touched it. As far as I can tell it was almost entirely FUD.

It is more "involved" to program, you do need to know how to optimise code, but the techniques you use are exactly the same as on every other processor. You do have to manage the local memory but that's just the same as cache management - which is critical for high performance code.

I did a little Cell programming and the guts of the code was copy-pasted AltiVec code from a Mac (that I originally wrote with a whopping 2 weeks of vector coding experience). To get high performance on the SPE all I needed to do was unroll the loop. It really wasn't very difficult at all.

I'm involved with GPGPU these days - it's more complex than programming Cell.
 
Almost all of the comments about Cell being difficult to program came from people who never touched it. As far as I can tell it was almost entirely FUD
John Caramack commented on the difficuly of programming the PS3. For example, RAGE looks about the same on both consoles. It did take the PS3 team longer to optimize for the Cell. Perhaps some was just the natural learning stage? Though I did want to note PS3 difficulty complaints did come from some programmers.

Perhaps part of the problem is the industry has moved to a model closer to the 360's design than the PS3's design. I understand that often the PS3 is the lead console now. But, I do wonder if the less restrictive 360 design is quicker to optimize coming off the PS3 than it is to do it the other way.
 
Perhaps part of the problem is the industry has moved to a model closer to the 360's design than the PS3's design. I understand that often the PS3 is the lead console now. But, I do wonder if the less restrictive 360 design is quicker to optimize coming off the PS3 than it is to do it the other way.

Other way around. It's easier to get something running on the 360 because it's more conventional. It's when you go to optimise, that it gets complicated. With Cell you have to optimise from the very beginning so it requires more work.

The difficult bit was probably the fact it's multi core, most PCs are the time were single core, switching to a 6 core vector machine is quite a jump.

I suspect the whole x86 in PS4 is an answer to this. That will let them get stuff up and running quickly. Surrounding it with a Cell or 2 will give them a lot of room to grow beyond that.
 
Ah yes, you're right. The next xbox is rumored to have an AMD GPU which got me confused there. Still, Sony's experience with the CELL processor is likely what's making them consider x86. But I'm not sure what makes them think any AMD chip is a good idea right now, unless they know something we don't. And the xbox is rumored to be getting a 16 core PPC. From what I know of game development, multi-cores usually isn't what you want. So that does surprise me a bit.

Don't forget Sony's experience with ARM via PSP.
 
Back
Top