Sign in with
Sign up | Sign in
Your question

IBM Power6 hits 4.5Ghz Stock!!

Last response: in CPUs
Share
May 22, 2007 3:18:43 PM

while AMD and Intel and counting cores IBM stuck to two but supercharged it to 4.5Ghz stock and changed the architecture. while Intels C2D can get to 4.2 with high end water cooling, i say this IBM proc would hit 7 or 8Ghz if properly over clocked. heres a link of more detail http://www.realworldtech.com/page.cfm?ArticleID=RWT1016...
May 22, 2007 3:25:17 PM

Looks like a dual core CPU, not a single core.

It's an impressive looking CPU, but it's for a totally different segment, compared to AMD and Intel CPUs (except in high end servers, which I think this CPU is designed for - maybe workstations).
May 22, 2007 3:37:03 PM

Sure would be nice to have a X6800 at 4.5Ghz with stock cooling 8)
Related resources
May 22, 2007 3:58:22 PM

Besides being a non-x86 processor, how specific to workstation/server is it? It wasn't that long ago that Apple was making PCs based on the Power5. Barring some major cost penalty since fewer of these chips are sold, what could keep them back from being an excellent choice for a Linux workstation? Are new versions of Mac OS still coming out for Power processors?

Unless your business is tied to Windows (which, granted, covers most businesses I know of), I don't see why these processors shouldn't be a hit. I would like to see how the Power6 stands up to Barcelona or Penryn.

There are a lot of 'if"s in the above paragraphs. I don't really know what besides the OS choice would be effected by using a Power processor. If I'm planning to use Linux anyway, is there some drawback? Would all my programs still work?
May 22, 2007 4:16:17 PM

P4s hit 8 GHz, so?
May 22, 2007 4:23:34 PM

I'm pretty sure any P4 hitting 8 GHz was with LN2 or phase-change cooling. IBM is claiming 4.5 GHz at the same TDP as the previous generation's at half the frequency. One hopes they didn't sacrifice too much on the IPC side in order to hit those frequencies.

I'd take a Core 2 at 4 GHz over a Pentium D at 8 GHz any day.
May 22, 2007 4:24:39 PM

I think someone needs to read the title again :roll:
May 22, 2007 4:40:00 PM

Ya, but what good is a PPC processor to anyone in the normal sect of computing?

x86 or bust I say.
May 22, 2007 5:20:53 PM

Power6 may replace IBM's CELL processor's in the PS3 and could switch Apple back to RISC CPU's. One can only guess as a RISC can do anything a CISC CPU can do all be it without code changes in emulation.
May 22, 2007 5:53:16 PM

i think someone in Apple is going to get a pink slip this Friday, if Macs had Power6 processors; that would mean a total different ballgame for Apple. The Power6 gets 4.5 at the highest voltage at this new architecture, if they just cut the voltage by 2 then that would mean a 2.25Ghz server proc in a mac book, or mini.

where would one have to go to buy an IBM Processor?
May 22, 2007 6:33:08 PM

I don't think it's quite as simple as half the voltage - stable at half the frequency. Still, IBM should be able to spin this chip into a decent mobile part if they wanted.

Apple wasn't dumb to go with Intel chips... Most of what they sell are laptops, and Intel has had the best laptop CPUs since Apple went with them. IBM wasn't delivering with a mobiel Power5, so Apple's hand was forced.

Besides, if you think the Merom chips are impressive, then the Penryn should be that much better. High-K/metal gate should do good things for power consumption. 15W 1.6 GHz Penryn, anyone?
May 22, 2007 7:27:27 PM

What I was saying is that the GHz come from reducing IPC compared to POWER5.
Same with the Prescott that would lag behind a C2D, even at 8 GHz.

And there is no clock speed headroom, since it had to be slowed down to 4.7 GHz from 6 GHz that IBM initially calimed and even so it consumes well over 300W (and dissipates quite a bit).
May 22, 2007 7:41:18 PM

You do realize that the competition for this chip is MP Opterons and Itaniums?
May 22, 2007 7:51:07 PM

They ~doubled GHz and ~doubled performance in same thermal envelope. You can have everything twice as fast with no increase in power or have the same performance with 1/2 thermal envelope.
May 22, 2007 8:32:33 PM

Quote:
What I was saying is that the GHz come from reducing IPC compared to POWER5.
Same with the Prescott that would lag behind a C2D, even at 8 GHz.


wrong the pipeline is the same as for power 5
May 22, 2007 9:23:29 PM

Quote:
Ya, but what good is a PPC processor to anyone in the normal sect of computing?


Hey, you can run OS7 on it. Nice Finder! :lol: 
May 23, 2007 12:43:29 AM

My gigahertz are bigger than your gigahertz. I win, right?
May 23, 2007 1:21:47 PM

Quote:
i think someone in Apple is going to get a pink slip this Friday, if Macs had Power6 processors; that would mean a total different ballgame for Apple. The Power6 gets 4.5 at the highest voltage at this new architecture, if they just cut the voltage by 2 then that would mean a 2.25Ghz server proc in a mac book, or mini.

where would one have to go to buy an IBM Processor?

I think walmart has PS3's which has IBM's CELL processor.
May 23, 2007 3:54:30 PM

My CPU runs at 4.5GHz on air cooling just fine, does that make me elite, or is there more to CPU's than just MHz....

Sun's Niagara II chip is running at onlt 1.4GHz, but will wipe the floor with the power 6.
May 23, 2007 4:04:05 PM

Quote:
Sun's Niagara II chip is running at onlt 1.4GHz, but will wipe the floor with the power 6.


In server applications, no doubt, what's the piepline length on the Niagara? 6?

I guess there's really not that much hope of seeing a Power6-based desktop any time soon, but I'd still hold that the Niagara chip is another class alltogether.
May 23, 2007 6:05:42 PM

Quote:
Besides being a non-x86 processor, how specific to workstation/server is it?


It is intended for servers. That said, you could use it in something else, However....

Quote:
Barring some major cost penalty since fewer of these chips are sold, what could keep them back from being an excellent choice for a Linux workstation?


I've not seen numbers on these, but when I last went to an IBM presentation, I had difficulty convinvincing myself that, for all their technical excellence, the IBM server products built around the previous generation of power products were cost-competitive (against Xeons, etc), so my strong expectation is that the price of a system built around one of these will be uncompetitive.

And you've probably got used to the idea that there is an enormous choice of motherboards with an enormous choice of hardware and overclocking facilities that you can choose from. There won't be as soon as you enter the Power world. And as for economical motherboards, or even just slightly uneconomical motherboards, you probably need to forget it.

At that point, the cheapest Power server (1U, lowest ram, lowest/no hard disc) was about £2000 - £2500. Bear in mind that you could get superficially similar x86 servers from HP and Sun for about half that (and they probably had the higher spec in terms of memory and disc space) and you can see why my interest was limited.

Quote:

Unless your business is tied to Windows (which, granted, covers most businesses I know of), I don't see why these processors shouldn't be a hit. I would like to see how the Power6 stands up to Barcelona or Penryn.


...against Xeon would be a more interesting (or, at least, more likely to be constructive) comparison (because of price).

Quote:

There are a lot of 'if"s in the above paragraphs. I don't really know what besides the OS choice would be effected by using a Power processor. If I'm planning to use Linux anyway, is there some drawback? Would all my programs still work?


No. Anything you can recompile yourself could, theoretically, be made to work, but if you have binary only programs (flash, video drivers, potentially anything not 'free') then you only have the chance of the software supplier compiling a version for you. I'm guessing this is a 'probably not', but YMMV.

and, from DaSickNinja
Quote:
You do realize that the competition for this chip is MP Opterons and Itaniums?


errr, and Xeons, as mentioned. I'm surprised there isn't more interest in Xeons. The latest X3000 and X5000 series parts are not unreasonably priced (compared with FX and Extreme Edition parts you could even argue that Xeons are cheap, but what isn't compared to them? Oh yeah, Lamborghinis :roll: ), the motherboards are a bit expensive, as is ram, but overall they can be competitive with the high-end 'traditional' gamer systems, provided 'loads of cores' is an advantage, which it isn't for some applications. Think you might have difficulty getting an SLI Xeon 8 core motherboard, though :D 
May 23, 2007 6:18:55 PM

Good points, all around.
May 25, 2007 8:21:16 PM

And, in the unconventional position of replying to myself:

Yeah, but.

Yeah, but Intel has just released review samples of 'V8'. Now, as I understand it, V8 isn't an actual product that actual punters can buy, but it is quite closely related to an actual product that will be released, or that you can put together off the shelves.

2x4 core Xeons (5365s which are fairly high end quad core Xeons at 3 GHz on a S5000VXN motherboard, 4G ram). So that'll be 8 cores rumbling away at the 3 GHz, then. I think the Xeons that you can buy today (5355s) are more like 2.8 GHz or slightly smaller cache or something. How could you put up with the low performance :(  ? (...but the real value is a bit lower down the price list, with prices comparable to an E6600). But I've seen everything, including MoBo, RAM, Xeons and PSU priced at £2400, and I'll be slightly surprised if the IBM 1u rig comes in quite that low and you could really save money on the Xeon chips without hiting the multi-threaded performance in the real world.

Yeah, but Amiga. Now bundle this may well be vapourware. And most of you will be thinking that Amiga got out-evolved in the '90s and, having become extinct, wouldn't be heard of again, but the latest rumour of a high-end Amiga from the ever-phoenix-like ruins of the Amiga organisation (and whoever owns it this week) revolved around PA semis PWRficient PA6T-1682M. So that'll be the two Power PC cores clocking away at 3 GHz (seems like the number of the day). And there is some confusion whether that's a single dual core chip or two dual core chips, due to ambiguity in the little information available.

So this would be running software that used to run on something like a very single core 68020 at 50 MHz (at best) on 2 or four cores of multi-issue Power PC goodness. And the software wouldn't have the 'bloat' that has clung to some other systems in the last decade. I'm not expecting it to be slow. Well, not slow in operation, although it might take forever to get released. And that same hardware would run Linux, theoretically, but that's only interesting if someone drives the volume up so the price comes down.

These are interesting times, even if everything that has come out of AMD recently is bullstuff, and I really hope it isn't.
May 25, 2007 9:17:06 PM

Quote:
Power6 may replace IBM's CELL processor's in the PS3 and could switch Apple back to RISC CPU's. One can only guess as a RISC can do anything a CISC CPU can do all be it without code changes in emulation.


So Sony spends like a billion to help design the cell processor, just to get it replaced within 1-2 years of the PS3 being out? I know the cell has other uses and potential in other products........but man that sucks.
May 25, 2007 10:20:03 PM

How likely is the switch from Cell to Power6? These chips are so different how could the PS3 possibly work with the Power6? Also, what's the processing power of the Cell when programs are actually able to take advantage of it?

I agree that the Cell wouldn't be a good DT CPU, but if the applications are written for it, it's very powerful. I would be suprised to hear that the Power6 is more powerful. Rather, I think the Power6 is more of a true general CPU.
May 25, 2007 10:54:59 PM

which fucktard sed P4s hit 8ghz,

becuase they hit 7.2

the architecture is out of date, and they overheat, and a 4ghz c2d will kill it anyway

PPC is an extremely powerful architecture, that is designed for extreme "performance" multitasking, while coupled with enough memory and a good OS (os/400), can handle over 500,000 email users. no slowdown. apple may have made the wrong choice, hahaha. well technically but not financially.

yes, after reading a post somewhere else, it is twice as fast as the PPC5, but half the power consuption, and they've sped it up

8ghz with a proper overclock, nearly 200%

thats like taking a 2ghz and putting it to 4 (very difficult), and only some can do it. so yes its probably possible, but unlikely (and no real point unless your running out of commitment. However in AS/400s (the machines which PPCs are usually in), they just put more CPUs in. Or, some can come with locked cores, (i.e. only 3 of 8 running,) and if your running out, you ring up IBM, pay some money, they give you a code and bam, another 5 cores

clever eh?
May 26, 2007 9:04:33 AM

oh yeah i see, thnx tacos.

except, is it a % oveclock or do you reckon that there is a limit to have many GHz you can stack on top?

I.e. you can get a 16ghz chip (say) to 32ghz? or will it just go to like 20ghz and stop there?
May 26, 2007 1:40:13 PM

i see.

does that mean then that unless AMD change their material for their cores, that their chips will always be bad overclockers?

i know the new C2Ds are awesome at it.

if AMD use SOI, what do Intel use?

thanks for schooling me taco-boy :D 
May 26, 2007 1:41:57 PM

oh yeah,

in the lab, IBM have got PPC6 to 6ghz.

speedy

and I think that may be theri architectural limit, seeing as IBM do infact know what their doing.
May 26, 2007 2:02:56 PM

yeah i see

65 is way better than 90, and I think next C2Ds are made on 45nm process. v small indeed,

so in theory should a 45nm chip OC better than a 65 one better than a 90one?

i mean pentium 4s are 90 but they get to 7ghz
May 26, 2007 9:25:53 PM

Quote:
For onething, it's highly unlikely chips will ever go above 10ghz, they might go a bit higher, but they will primarily just get more efficient


People have had the ALU's in the P4 running at 16GHz http://www.nordichardware.com/news,5505.html so 10GHz is easy if you want to.. Infact, the ALUs on my PC run at 9GHz day in day out :p 

(FYI the ALU's run twice the core - google double pumped alu)
May 26, 2007 11:42:50 PM

Quote:
yeah i see

65 is way better than 90, and I think next C2Ds are made on 45nm process. v small indeed,

so in theory should a 45nm chip OC better than a 65 one better than a 90one?

i mean pentium 4s are 90 but they get to 7ghz

Nanometer change is done for 2 reasons. One is to allow for more design space and the second is half the distance electric has to travel to get from point a to point b. The shorter the distance electrice has to travel the less voltage is needed to be readable at point b. This has many factors such as leakage and how high heat areas change at the lower nanometer.

Less voltage the cooler the CPU stays at a given GHz and allows for higher clocks. P4 design had much less structures than C2D's and thus needed more GHz to do the same job. This is the trade off and human error plays a role and thus revisions are made.

Now moving C2D to 45nm would give higher clocks but may not perform as well as a core designed for that nanometer. Just as netburst became inefficienct so did K8 and the costly R&D process starts over.
May 27, 2007 6:24:27 PM

The efficiency of the chip doesnt go down per say it just doesnt take advantage of extra space like a core made for a given nanometer. Think of it like have a big truck with a 4 cyclinder with the room for a V8. Think of this truck pulling a load. The higher the clock cycle the larger the load. At some point your better of with the V8.

Increasing the clock is important to how much cache is needed so the design doesnt starve for data. The faster the cache the less cache is then needed to make up for slower RAM. Clock increases will always be needed else the cache will end up taking much of the space. Note how cache has went up since the move to dual and quad core.

Now just like P4 and K8 new core designs live up to 3 process changes before major changes are needed.

High K or Low K isnt design but the ingredents used. hafnium silicates is the next big thing in reducing leakage and has much to do with the choice of high k or low k. Hafnium is used in nuclear plants to adsorb heat and energy on the rodes. Its found in zirconium used to make fake diamonds.
http://www-03.ibm.com/press/us/en/photos.wss?topic=407
http://domino.watson.ibm.com/library/CyberDig.nsf/398c9...

High K or Low K is used to lower leakage which lowers the voltage needed. Think if we could use Gold instead of copper. Voltage travels though super conductive materials like gold with almost no loss. The CPU could be the size of your mobo have thousands of cores. This is where the CPU is headed atleast until we get to optic CPU's.
May 27, 2007 9:13:21 PM

Quote:
High K or Low K is used to lower leakage which lowers the voltage needed. Think if we could use Gold instead of copper. Voltage travels though super conductive materials like gold with almost no loss. The CPU could be the size of your mobo have thousands of cores. This is where the CPU is headed atleast until we get to optic CPU's.


It is a common misconception that gold has low resistance, well it does have a fairly low resistance of 2.44 × 10-8 ohm-m but copper has a much lower resistance of 1.72 × 10-8 ohm-m and is thus much more suitable for use inside electrical equipment. The main reason for seeing gold on the contacts of some higher end equipment is actually its complete resistance to any form of oxidation, thus no thin film of oxide builds up over the contact reducing the current flow over the very outer later, plus as I stated above, it's still a fairly good conductor.

But, if you do want to take things to the extreme, silver is the best conductive metal, with a resistance lower then copper at 1.59 × 10-8 ohm-m, this would be the ideal metal to use in chips if it were not for the cost issue.
May 28, 2007 3:15:20 PM

Your both correct and incorrect as gold in its natural state isnt superconductive but at absolute zero a compound of gold and indium is both superconductive and magnetic.
Quote:
In recent years, many discoveries regarding the novel nature of superconductivity have been made. In 1997 researchers found that at a temperature very near absolute zero an alloy of gold and indium was both a superconductor and a natural magnet.

http://www.superconductors.org/History.htm

Quote:
But, if you do want to take things to the extreme, silver is the best conductive metal, with a resistance lower then copper at 1.59 × 10-8 ohm-m, this would be the ideal metal to use in chips if it were not for the cost issue.

Silver isnt the best as high heat superconductive materials would be but i see your point as its better than copper. The problem with silver is cost and is not as plyable as copper. Copper is easy molded. On the other hand silver as a heat sink would offer better cooling.
http://www.overclockers.com/articles452/
Weigh of a heat sink could be a problem so only a small amount could be used like current copper and aluminum heatsinks.
May 28, 2007 4:24:49 PM

yeah. ibm... i remember when they used to make consumer products... hardy har har.
!