Ghz isn't everything; IPC (instructions per cycle) is far more important. Remember 3Ghz P4s being blown away by 2Ghz Athalons? Anyways, yes...the state of CPU speed has hit a wall due to voltage/heat limitations of current technologies used; the only answer at this time is "more cores are better than one (right?)". Note that Bulldozer traded IPC for speed and got nowhere fast doing so.

Hummmmm.....with that kind of cache I'll try one on FSX just for giggles....., better, let send a copy of FSX to IBM and ask them if they would be nice enough to put their machine to the test with FSXMark11 using PCI-e 3.0 of course...

 

Alain, FSX will not run on an IBM Power CPU, it is a Windows application compiled for x86 ISA. Power CPU-based systems can neither run x86 apps nor Windows.

Oops! I stand corrected.......did not think about that....shame on me.

Oops! I stand corrected.......did not think about that....shame on me.

 

Fantastic chips though, very powerful. Also very power hungry, hot, and expensive but that goes with the territory.

I don't think that is too much headroom left in the 22nm builds, so can we go smaller? IMHO you would also need a new memory controller to be able to work with higher frequency RAM (think of the heat on say 4000MHz RAM), remove the PCIe bottle neck, and upgrade the OS. But it could happen.

PeterH

4000MHz RAM

 

:shok:

I don't think that is too much headroom left in the 22nm builds, so can we go smaller? IMHO you would also need a new memory controller to be able to work with higher frequency RAM (think of the heat on say 4000MHz RAM), remove the PCIe bottle neck, and upgrade the OS. But it could happen.

PeterH

 

I read an interview with a physicist who said that we may see Moore's Law (That computer processing power doubles every 18 months) stop holding true, because with 22nm we are reaching the limits of what silicon can handle. If we work with silicon wafers that are much thinner, they will not stand up well to voltage. Perhaps the heat issues we're seeing with IB are examples of what happens when silicon starts getting too thin.

RR

Good point. The die size of IB is much smaller than SB and even though the heat output may be lower it is concentrated in a smaller area and so cooling becomes quite important. It will be interesting to see what the next 'tick' brings.

Regards

pH

Alain, FSX will not run on an IBM Power CPU, it is a Windows application compiled for x86 ISA. Power CPU-based systems can neither run x86 apps nor Windows.

 

No emulator available?

 

No emulator available?

 

Doubtful. Even if there were, you lose immense amounts of performance through emulation so it would be rather pointless.

5.2GHz out of the box....(snip)

 

http://hothardware.c...nterprise-Chip/

In this linked article, read the comment below it "By Wagan8r on Sep 17, 2010"

 

This comment echos my thoughts exactly...Ghz smoke and mirrors. I've yet to find any benchmark results using google for a z196; seems IBM does not want cold, hard GFLOP numbers for this chip published.

It's not an HPC-oriented CPU anyway, so GFLOPs are not the focus. IBM designs Power chips to win TPC benchmarks and secure server sales based on that.

I read an interview with a physicist who said that we may see Moore's Law (That computer processing power doubles every 18 months) stop holding true, because with 22nm we are reaching the limits of what silicon can handle. If we work with silicon wafers that are much thinner, they will not stand up well to voltage. Perhaps the heat issues we're seeing with IB are examples of what happens when silicon starts getting too thin.

 

I believe Moore's Law deals with the number of components in an integrated circuit, and doesn't address "processing power".

 

I also suspect the problem of dealing with high frequency signalling is as much a problem as heat and voltage issues.

 

scott s.

.

I believe Moore's Law deals with the number of components in an integrated circuit, and doesn't address "processing power".

 

You're correct. Intel VP David House stated that the doubling of transistor density would result in a doubling of processing power (which is measured in terms of floating point operations per second, or flops), creating a common misperception of what Moore's law is. Moore stated transistor density would double every 2 years, and House figured that the transistors would be faster, leading to his statement that processing power would double every 18 months.

 

Here's a link to the article I mentioned.

I remember reading something about graphene based chips. They were able to run at speeds of 500 GHz, however, you can't turn off the transistors. So you might be able to use it for the "on" state (one), however, you need something for the "off" state as well (zero), otherwise, the power consumption goes through the roof. I'm not an engineer, just what I've heard about them.