What do people think of P.A. Semi CPUs ?

Beastyben1

Distinguished
Mar 2, 2006
46
0
18,530
With a name like PWRficient, how could they not be awesome?

On a more serious note, one of those articles was written in late 2005 meaning this chip is slated for mid/late 2007 and is to compete against IBM and FreeScale Semiconductor. Tough market.

I haven't seen any performance notes but the power consumption numbers are great. 2 cores at 2.0ghz at 5 watts and max 25.

I'm not sure what specific market their targeting but good luck to um.

Ben
 

plankmeister

Distinguished
Sep 7, 2006
232
0
18,680
Looks pretty awesome... I guess they have working products if they're due for release soon.

I think the idea of integrating the entire chipset into the die is definitely the way forward. Hopefully this will give Intel and AMD a kick in the ass and get them to do the same. The advantages are numerous!

However, I'm certain that (at least for the near future) there won't be a suitable desktop processor from them to challenge the Big Two. Maybe if the company takes off and they start kicking ass in their target market they'll consider creating a C2D beater? (and - wow - what a beater it would be!)
 
I heard rumors of a new chip maker hitting the market in '07 with an eye on cutting into the low-end market. I like the idea of integrating the north and south bridge and sounds ideal for the low power embedded consumer market. By partnering with TI they are pretty much guarenteed a great fab process and a foot in the door with the embedded chip market. I can't imagine this chip making it's way into the enthusiast market and don't think we'll be seeing the PWRficient overtaking C2D or Athlon, but I do see great potential for it in the consumer appliance market. This would be an excellent choice for increasing the functionaility and usability of smart devices.

I remember reading a Popular Mechanics article where they hypothisized about "connected" appliances like the refrigerator, stove/oven, TV, heater/AC, and phone linked up to a broadband connection opening the possibilities of remotely heating/cooling your home, turning the lights on and off, even recording ala PVR while away on vacation. This type of processor would play into that "automated home" and "appliance management" scenario.

I'll have to keep an eye out for any further news from PAsemi. Thanks for the article. Happy New Year!
 

TabrisDarkPeace

Distinguished
Jan 11, 2006
1,378
0
19,280
Yeah but imagine, 20 cores, using 10 dual-cores, using around 135 watts.

Consider that most server chips from IBM (Power 4,5,6), and high end consumer CPUs (Athlon 64 FX, Pentium Extreme Edition, Xeon 700 series) all use around 135 watts too (sometimes even more).

That is some pretty sweet performace potential right there, and it scales down to fully functional (eg: It runs a real OS) PDA sized handhelds, and even 6 core laptops at these power envolopes.

I would seriously pay good money for such a laptop, if it supported 8 GB RAM and ran Virtual Machines.
 

CaptRobertApril

Distinguished
Dec 5, 2006
2,205
0
19,780
Given the tiny size of the company, you can bet that their CPUs will be quite pricey as they cannot match the economies of scale of the big guns. That approach is 180 degrees away from what I lust after. I want massive processing power CPUs that are affordable/cheap and use up so much electricity that I have to run a cable next door and steal his juice and cool it with a 50 ton cooling tower! :lol:
 

darious00777

Distinguished
Dec 15, 2006
687
0
18,990
This reminds me of what I've heard about AMD. They buy up companies and use the technology to get ahead. Intel seems to do everything internally.
 

TabrisDarkPeace

Distinguished
Jan 11, 2006
1,378
0
19,280
These chips are not x86 compatible, they are PowerPC / G5 compatible, so I doubt nVidia has an interest in them. Unless they want to start manufacturing PPC machines.

nVidia have used their video cards in PPC machines before though, so I wouldn't call it an impossibility, just highly unlikely.
 

ajfink

Distinguished
Dec 3, 2006
1,150
0
19,280
These chips are not x86 compatible, they are PowerPC / G5 compatible, so I doubt nVidia has an interest in them. Unless they want to start manufacturing PPC machines.

nVidia have used their video cards in PPC machines before though, so I wouldn't call it an impossibility, just highly unlikely.

Indeed. I wouldn't doubt another large microelectronics producer of some sort buying them, though. The IP portfolio -might- be worth it in itself.

If these things use as little power as an above poster said, they could be used in tons of different applications with varying mobility.
 

Bozzkins

Distinguished
Jan 2, 2007
5
0
18,510
It would be an interesting option for Microsoft's next generation of console since its PPC based. Eight of these chips running at 2Ghz would still only consume 104watts and give 16 ppc cores! Couple that with process improvements that would run the cores at 2.5Ghz (based of their site) and its starts to look like a pretty hairy beast. Add a reasonable GPU (say a quad core r600) and you're looking at something that will probably be pretty good! :D
 

TabrisDarkPeace

Distinguished
Jan 11, 2006
1,378
0
19,280
Tabris, now that Apple has gone Intel, is there any non-server market left for PPC?

I don't know for sure, and I can't really find out. Unless 'we' are the market. (Or at least people thinking along the same lines as me, looking to move from XP, but not to Vista, instead to Linux, OS X.5, OS X.6, Solaris, and other OS's, using Virtual Machines. Get a clean slate and try and make / port / refine 'good, quality' software for a 'differet' [well, non-Windows, non-x86] platform - Give IT a nice clean slate for one.
Cut the 'backwards compatibility cord' and let fly.

If you count me as part of that market, then yes, There is a market.
Even Cyrix / National Semiconductor / Via are doing well in the Asian market, with a similar design. Sure FPU performance sux, but FPU performance on the Sun SPARC has typically been pretty bad compared to x86 and it has a huge market.

- The irony is these P.A. Semi redesigned G5 / Power5 CPUs have strong FPU performance, and strong everything else performance. - I would call it a CPU without a weakness, and coming from me I feel that really means something. - The only things that comes to mind is (1) 'It isn't an x86 design, and (2) the royalties don't go to Intel, which isn't exactly a weakness until you bring into account that Intel is the 800 pound Gorilla with a giant legal team.

Could be large in Asian markets, which is good news for Australia since we can tap their 'grey' market more easily than the US. Assuming they sell exclusively to Asia, which is also unlikely.

But I wouldn't exlude these chips from server installations, 216 watts for 32 G5 like cores at 2 GHz, it was only a few years ago that was thought impossible by Apple, which is why they went to Intel.

That is damn sight better performance all up, performance/watt, and performance/cubic inch, etc than most other systems on the market. Including IBM 'Mainframes', and Sun UltraSPARC or CoolThreads designs.

Frankly that alone deserves heaps of attention, Heck if most x86 software needs a re-write to take advantage of 64-bit, but most Power5 / G5 software is already 64-bit, and we are going down both the multi-core path, and performance/watt path, then I can only see these guys shaking the market up.

Think about it:

- A PDA that can run most fully fledged Operating Systems using just 1 or 2 cores.

- A laptop with 3, 4, 5 or 6 'G5' cores and still within laptop power designs.

- Servers with 32+ cores using only 216 watts (typically) for processors.

The source code and binaries for PowerPC software (Operating Systems included) still exists, it may even become free / open-source, and it already takes full advantage of 64-bit PPC (G5 style) processors, will scale to multi-core designs easily, and most of it supports the SIMD instructions on offer. - The software did all this years ago, just back when the only G5 systems made the PreScotts look good (performance per watt wise).

A 2 GHz G5 core is enough for most 'normal' single threaded apps, with 2, 4, 6, or 32+ of them running in a 'normal' SOHO sized PC, within 'normal' power and cooling specs is a huge jump.

A Core 2 Duo needs around 65 watts, at least 40 watts for the similar clocked low power designs using Intel NGMA.

With this design one could run 9 cores for every 2 cores in an Intel Xeon 5100/5300 series server, or Core 2 Duo desktop. Even 6 cores for every 2 in the mobile and lower power server arena (eg: Blade servers, or where cooling and/or performance per watt is more important).

They've covered blades, low power devices, huge core count servers, desktops, laptops, decent single threaded performance, floating point performance (for scientific community), even routers + embedded hardware, and pretty much everything else with just one chip.

If a CEO / CIO with half a clue is reading this please start mass manufacturing and get economy of scale going, this chip was built to cover the market end to end.

Look at the Power5 and G5 designs, examine them for say 30 minutes, then disect this P.A. Semi design, noting it scales both ways. Wikipedia has heaps of links, but knowledge alone doesn't make a company.

I can safely assume most people know the difference between (the various) Pentium 4 NetBurst designs and Athlon 64 / K8 designs, this stuff isn't all that different.

It is a real kick in the ribs that IBM are only doing heavy-iron, but I can see this becoming the next 'big' Open Source style platform, which means eventually I'll become commercial since it'll be far more cost effective (which is why Novell bought out SUSE Linux, instead of re-writting NetWare to support x64 and scale beyond 64 GB).

Thanks to .NET and Java the future is psuedo-code anyway, not binaries (which I don't agree with, but not for this thread I guess), and such 'psuedo-code' would compile to give absolute world-class performance on a platform designed from this P.A. Semi stuff, scaled to 32 - 256+ cores.

Just 1 server cabinet would provide all the processing capacity that large (50,000 staff) enterprise would ever need.

Most software in the world is sub-standard quality anyway... just not just migrate to a totally new platform, re-using old binaries and source code to get there, and start over afresh ?

The open source community already have Linux on PlayStation 3, and likely on Xbox 360 too, both those platforms DESIGNED FROM THE GROUND UP FOR GAMING, use PowerPC processors (althought the SPEs in the Cell are overkill and wasteful). - Yes gamers that is right, not Athlon 64, not Core 2 Duo, and not x86/x64.

But a console is such a small device, with such a consumerish power (usage and number crunching) level.


Combine this tech (which is already based on not so old IBM designs) with IBMs new tech to use a DDR style system within CPUs to double performance while keeping power usage very similar and you've got alien grade technology (relative to everything else on the market).


PS: Doom on PowerPC ran with 4 times the pixel count compared to Doom on x86 for an equal priced machine. This is back when games did not have a GPU to offload 3D rendering too. Surely there are people here who recall this ?
 

Bozzkins

Distinguished
Jan 2, 2007
5
0
18,510
It depends what their business model is. If it is to make chips then they are largely constrained by their ability to get fab space. If its really to license out their IP to third parties then its a lower startup cost but the potential rewards are lower because they are only getting a license fee (and presumably paying IBM a license fee for the original PPC rights too).
 

Harrisson

Distinguished
Jan 3, 2007
506
0
18,990
Looks good, but I still remember how Transmeta was pretty much killed by power efficient Intel and AMD cpu's. At least Transmeta made very universal cpu's, what cant be said about P.A. Semi, who starting in the very niche market already. We still dont know their performance, and while they seem to be amazing for embedded markets (high concurrency, and probably new AMD vision heads in the same way), also blade servers (SUN Niagara anyone?) but I doubt we will see P.A. Semi in desktop mainstream, actualy I'm pretty sure we wont any time soon.

Btw, Sun Niagara 2 with 8 cores consume 72 watts of power, P.A. 8 cores x 5 to 13 watts = 40 to 104 watts. Pretty much the same. If P.A. can deliver much higher performance and get Tier1 wins, they are into lucrative market, otherwise they are already behind some big names, like SUN.
 

TabrisDarkPeace

Distinguished
Jan 11, 2006
1,378
0
19,280
These chips can do alot more than the over-specialised Sun SPARC IMHO.

Performance per Watt would beat Sun hands down, even in 'tasks' that Sun are considered world leaders in (which isn't standard floating point or SIMD).
 

mpjesse

Splendid
Here's the problem:

Thus far, all start ups relating to power efficiency have been monumental failures.

IF they manage to patent a technology that's absolutely outstanding, they might have a future in licensing. But I wouldn't count on them being a serious contender in any market as far as a chip goes.

When you get right down to it, power efficiency isn't at the top of the list of concerns. Though, it's certainly getting there.
 

Harrisson

Distinguished
Jan 3, 2007
506
0
18,990
These chips can do alot more than the over-specialised Sun SPARC IMHO.

Performance per Watt would beat Sun hands down, even in 'tasks' that Sun are considered world leaders in (which isn't standard floating point or SIMD).
I think so too, but its a wild guess as long as we dont have performance figures. From other side - performance isnt everything, so many faster cpu's died to lower performance but cheap cpus made by market leaders. Alpha was killed by Intel/HP just becaused it smoked Itanium out of the water, and was even cheaper, but Intel thought he could make more money with Itanium, which is sinking down ever since :wink:
 

TabrisDarkPeace

Distinguished
Jan 11, 2006
1,378
0
19,280
The guys working for P.A. Semi worked on the Alpha, they also worked on the Opteron and the Itanium (which only has a real future at 45nm to 22nm).

The problem is a lack of well designed & well coded software - that is 5 to 6 years away at our current rate.

http://en.wikipedia.org/wiki/IA-64
http://en.wikipedia.org/wiki/Itanium_2

Considering the Itanium 2 is still only on 90nm, it performs quite well. A jump to 65nm moer than doubles the potential transistor count (at the same material size, could always make larger dies) and will raise the clock speed by 20 to 40% at a similar power consumption.
 

Harrisson

Distinguished
Jan 3, 2007
506
0
18,990
The problem is a lack of well designed & well coded software - that is 5 to 6 years away at our current rate.

Considering the Itanium 2 is still only on 90nm, it performs quite well. A jump to 65nm moer than doubles the potential transistor count (at the same material size, could always make larger dies) and will raise the clock speed by 20 to 40% at a similar power consumption.
Itanium is a dead horse, and at the time it was developed, Alpha was much better performance-wise. Itanium does well in specific tasks, but even behemot as Intel couldnt make market to accept it. Maybe Intel will make it profitable one day, but I very very much doubt it, so far Intel lost billions of dollars.

In any case, its offtopic. I will eagerly wait for P.A. new announcements, benchmarks, but I dont hold my breath it will become big player in cpu field in any area.