IBM, Moore's Law and the POWER 5 chip
Incredible article about IBM's Power processor at The Register.
Snippet:
IBM's recent pSeries benchmark ought to raise a good number of eyebrows. Unix server benchmarks had previously been a game of leapfrog between IBM and HP, with IBM looking increasingly strong in the last round (in 2003) when its p690 server with 32 processors demonstrated a slightly higher transaction rate than HP's 64 processor Itanium 2 server. The latest benchmark from IBM leaves both previous benchmarks in dust. In the TPC-C benchmark, IBM demonstrated over 3m transactions per minute (tpmC) almost three times more than the previous highest result of 1.2tpmC, posting a 37 per cent price performance advantage in the process.
Full article:
http://www.theregister.co.uk/2004/11...r5_moores_law/
Snippet:
IBM's recent pSeries benchmark ought to raise a good number of eyebrows. Unix server benchmarks had previously been a game of leapfrog between IBM and HP, with IBM looking increasingly strong in the last round (in 2003) when its p690 server with 32 processors demonstrated a slightly higher transaction rate than HP's 64 processor Itanium 2 server. The latest benchmark from IBM leaves both previous benchmarks in dust. In the TPC-C benchmark, IBM demonstrated over 3m transactions per minute (tpmC) almost three times more than the previous highest result of 1.2tpmC, posting a 37 per cent price performance advantage in the process.
Full article:
http://www.theregister.co.uk/2004/11...r5_moores_law/
Comments
Originally posted by Brendon
I think that the interesting thing is that either in this article or another that talks about the POWER5 and the POWER5+ chips. The designers have taken curtain routines that may have been implemented in software and done it in hardware. They have even talked about adding to the POWER5+ curtain instructions that deal with requests made by the biotech-ies. This is very much like changing registers or adding registers to address performance issues. The old way of making the chip faster just doesn't cut it anymore and now it will be a combination of speeding up the chip and adding special instruction units that deal with curtain routines much faster because it is done in hardware. I just wonder when something like this will make it into the PowerPC line, which it may already have started, with altivec. Their could be more and that makes things interesting.
Most of the POWER5's capabilities in this area are things which can be done in the OS without having to update the application/server software. Handling the TCP in hardware, for example, means that the processor isn't constantly interrupted to deal with the network while its busy with other things.
Originally posted by Brendon
I think that the interesting thing is that either in this article or another that talks about the POWER5 and the POWER5+ chips. The designers have taken curtain routines that may have been implemented in software and done it in hardware. They have even talked about adding to the POWER5+ curtain instructions that deal with requests made by the biotech-ies. This is very much like changing registers or adding registers to address performance issues. The old way of making the chip faster just doesn't cut it anymore and now it will be a combination of speeding up the chip and adding special instruction units that deal with curtain routines much faster because it is done in hardware. I just wonder when something like this will make it into the PowerPC line, which it may already have started, with altivec. Their could be more and that makes things interesting.
I should have provided a link. I have not been able to find the exact article but this one touches on the subject, called Fast Path. You can find it near the end.
http://news.com.com/IBM+plots+new+Un...74.html?tag=nl
Basically it comes across as fanboy praise of IBM.
Where is the benchmark's results referred to in the article?
How can a journalist for the tech sector f%$@ up the definition of Moore's Law?
Why would said journalist skip over the technical reasons for IBM's supposed technological lead, unless he a) really didn't have a clue how it works, b) had no such info or c) there is no such lead?
This artcle comes across as hype without substance, like a vaporware press release. I look forward to other, much better, more detailed articles on the same subject. Ones with depth.
Originally posted by 1337_5L4Xx0R
Where is the benchmark's results referred to in the article?
How can a journalist for the tech sector f%$@ up the definition of Moore's Law?
Why would said journalist skip over the technical reasons for IBM's supposed technological lead, unless he a) really didn't have a clue how it works, b) had no such info or c) there is no such lead?
This artcle comes across as hype without substance, like a vaporware press release. I look forward to other, much better, more detailed articles on the same subject. Ones with depth.
The results are official and easy to find.
People get Moore's law wrong all the time. Very few people actually appreciate what it is derived from, which is economics.
No idea where the price performance figure comes from. That just appears wrong.
Originally posted by 1337_5L4Xx0R
That's one of the worst articles I've read in a long time.
Basically it comes across as fanboy praise of IBM.
Where is the benchmark's results referred to in the article?
How can a journalist for the tech sector f%$@ up the definition of Moore's Law?
Why would said journalist skip over the technical reasons for IBM's supposed technological lead, unless he a) really didn't have a clue how it works, b) had no such info or c) there is no such lead?
This artcle comes across as hype without substance, like a vaporware press release. I look forward to other, much better, more detailed articles on the same subject. Ones with depth.
The folks at El Reg are hardly IBM fanboys.
IBM is getting praise because they just handed HP their heads on a platter when it comes to server apps. IBM just posted a 300% generational increase in computational power where it is common to only have a 10% increase.
SO IBM JUST OUTPERFORMED EXPECTATIONS BY 30 TIMES.
If you aren't aware of the Power5's latest scores, then you need to read sites other than this one. This has been big news for the last week.