The PS4 may well make for a better video game machine and more profit for Sony, but it's too bad the very interesting and powerful cell processor concept will be relegated to the dustbin.
The Cell design was interesting, but it was obvious that chip multiprocessors would win. The argument is similar to why x86 continues to prevail vs RISC and VLIW: supporting x86 is only a modest constant factor in transistor count but preserves access to the existing software ecosystem. The latter trumps the pennies per part the x86 decoders add.
Likewise with chip multiprocessors vs systolic arrays: making a core fully general and with proper cache coherant memory access isn't that expensive in transistors, and allows you to leverage existing multi-threaded code.
Systolic arrays still have a future in true embeded devices: radar processing is a good example.
Oh, and before you point to the AMD architecture as being Cell-like, it's not at all. Unified virtual memory instead of Cell's complex 2 level store. Cell requires the CPU to manage DMA's to the APU's, on Fusion you can just pass a pointer. CPU and APU instruction sets are entirely decoupled on Fusion. APU supports thread pre-emption.
In any case, the lesson here is obvious: design hardware in opposition to the existing software ecosystem in your industry, and you'll make something that is academically interesting but struggles to deliver value to end products.
The x86 vs RISC debate is a perfect example of how technological advances can confound seemingly simple tradeoffs. There's simply no such thing as a CISC processor anymore. All "x86" processors have RISC internals and use micro-op translation to present an x86 ISA at a software level. As you point out, the same sort of dynamics play out at every level. There's only so much advantage that simplified core designs get you, since the overhead for "legacy" support tends to be small.
It'll be interesting to see how Intel's x86 SoC development pans out. With the Atom they weren't really aiming for the SoC market and only recently have they actually bothered to try to do so. Even with their first generation stuff they've produced pretty competitive, to ARM, results, given Intel's research muscle and their FAB capacity it wouldn't be surprising to see a lot more x86-SoC based smartphones and tablets hitting the market in the coming years.
The APU in the PS4 picks up many ideas from the Cell processor. You have some heavy cores for program logic and many small ones for graphics and computation tasks. Both on the same piece of silicon and attached to the same memory space.
Maybe, but it does not seem to be very powerful at all. I've seen most of the upcoming games for launch (and played about half of them) and I was far from being impressed. It feels like a mid-range gaming PC kind of power at best, and many games apparently struggled to run in full HD, with the framerate low enough to make you feel you were back in the PS2 days. If they are planning to sell these consoles solely on the graphics, they'll have a tough time.
Launch titles are generally either carry-over concepts from earlier consoles and/or are drastically rushed to hit a very hard deadline, the titles hitting in the next couple of years will up the ante significantly.
That aside, I'm personally looking forward to next Friday quite a bit, as my preordered PS4 from Amazon is supposed to be arriving that day and having played around with some of the games already I'm really looking forward to more time with Killzone, Knack and Assassin's Creed 4.
FWIW, I'm one of those gamers who was a huge fan of the Xbox, mostly due to Live as a multiplayer platform, but I'm looking much more forward to the PS4 than the Xbox One, in fact I have no immediate plans to buy a One, though I'm sure I'll probably buy one eventually (much like I bought a PS3 eventually, but that wasn't until this year).
Thanks for posting at least one pic of what launch games looked like last generation. I urge people to go look at some others on the web.
Anyone arguing that games don't look that much better than current gen need to remember how much progress we've made this generation, and then look at how good games like Killzone: Shadow Fall already run and look at launch.
That's another topic. It's still a lot more powerful than the original Cell in the PS3. (And a lot easier to program to boot)
Sony originally intended to have the cell do all the rasterization work in the PS3. They had to add an additional NVidia GPU in the last moment once Cell proved too slow to do that at competitive speed. That was one of the reasons why the PS3 was so expensive at launch.
I suspect the $400 PS4 contains roughly $400 worth of hardware, so it's not surprising that it would be similar to a mid-range PC. This generation isn't as "generous" as the $600 PS3 that cost $800 to make.
Honestly, you'd spend $600-$650 to build a comparable computer. And that's not counting the fact that the PS4's software, both OS and games, are built for a single target. And that's assuming you'll get cheap peripherals and cheap case and stuff. And probably doesn't include the cost of a Windows license either.
Specifically video cards don't start passing the PS4 until the $200 range, at least when comparing via synthetic benchmarks.
I'm also unaware of how to get memory like GDDR5 for PC RAM.
Yeah, this is not true at all. You can get video cards that are better than the PS4's (by enough to counter any initial optimization, if not end-of-life optimization) for $150. Specifically, 7870 GHz editions.
The only thing you can't get in a PC is the high-bandwidth shared memory. You can probably make up for it by having tons of RAM on your PC's video card, but I still think the shared memory architecture will be the most interesting part of the current generation of consoles.
Haha sorry buddy, the 7870 editions that beat the PS4 are $200. You're right that they are better than the PS4 GPU, but wrong on the price. The 7850's get down to $150, but they're also benchmarking lower than the PS4 in synthetic benchmarks.
Also, a $50 price difference doesn't invalidate my answer of $600-$650 --- in fact you'll notice that it lands exactly in my stated error range of $50.
EDIT: My bad, I found a super off-brand cheap bad-warranty version of the 7870 for $170.
But speaking from experience, don't buy a shit card from a shit company with a shit warranty.
You don't put an experimental design into production on day 1. This is similar to the classic "let's rewrite the whole thing" problem. Maturity of a platform is an important factor, and it's often difficult to pinpoint all of the particular aspects behind the maturity of a given platform.
Intel made the same mistake with the IA64/Epic/Itanium architecture. They were banking on that to be the future of high performance computing and then AMD kicked their ass by using an iterative approach with x86-64. The irony is that today the Itanium architecture is actually pretty good, finally, but it's been relegated to a tiny market niche.
Likewise with chip multiprocessors vs systolic arrays: making a core fully general and with proper cache coherant memory access isn't that expensive in transistors, and allows you to leverage existing multi-threaded code.
Systolic arrays still have a future in true embeded devices: radar processing is a good example.
Oh, and before you point to the AMD architecture as being Cell-like, it's not at all. Unified virtual memory instead of Cell's complex 2 level store. Cell requires the CPU to manage DMA's to the APU's, on Fusion you can just pass a pointer. CPU and APU instruction sets are entirely decoupled on Fusion. APU supports thread pre-emption.
In any case, the lesson here is obvious: design hardware in opposition to the existing software ecosystem in your industry, and you'll make something that is academically interesting but struggles to deliver value to end products.