IBM Launches p690

edyavno writes: "IBM just announced the launch of their new high-end Unix server p690. It's based on its new Power 4 chip, and is in the same category as just announced Sun's SunFire 15K. It also includes some mainframe level features and can be used either as a single large server or divided into up to 16 "virtual" servers, running any combination of AIX 5L and Linux. Here's yahoo article, and here it is from IBM itself."

Comments

Overview from IBM's website

[isj]

I grabbed this from IBM's website (http://www-1.ibm.com/servers/eserver/pseries/hard ware/datactr/p690.html):

- Innovative, mainframe-inspired, datacenter-class UNIX
server.
- 8- to 32-way 64 bit SMP server utilizing the first ever
POWER4 dual processor on a chip which uses IBM advanced
silicon-on-insulator (SOI) copper technology.
- Up to 256GB of memory, 160 PCI slots and over 4.6TB of
internal storage.
- Supports up to 16 logical partitions (LPAR), helping to
consolidate workloads, reduce footprints and lower cost of
ownership.
- A dedicated Hardware Management Console that provides a
graphical user interface for configuring and operating the
system including a set of functions for managing LPAR
configurations
- State-of-the-art self-managing capabilities that improve
reliability, availability and serviceability (RAS) and help lower
costs.
- Packaging in a new 24-inch rack with an integrated power
subsystem which accommodates a pSeries 690 system and up
to four I/O drawers.
- AIX clustering and future to attach to SP systems.

It looks very good. I just wonder what you would use 160 PCI slots for?

Re:Overview from IBM's website

[zevans]

It looks very good. I just wonder what you would use 160 PCI slots for?

It is very good - and that list doesn't even include the ability to run Linux in an LPAR, which is also possible.

Don't think of it as 160 slots - think of it as 10 slots per LPAR. Suddenly it looks a lot less.

Zack

Wrong Comparision

[Doctor_D]

IBM's comparing their new server against the wrong Sun server. Here's why:

For unmatched UNIX system performance, the pSeries 690 can scale to a 32-way symmetric multiprocessor (SMP) helping to provide the scalability required to drive a UNIX datacenter.

A Sun Fire 15k contains up to 106 processors (72 with max i/o), a Sun Enterprise 10k contains up to 64 processors, and a Sun Fire 6800 contains 24 processors. Honestly this IBM server should be compared with either the 10k or 6800. It just can't scale as high as either the 10k or the 15k.

LPAR support for up to 16 UNIX or Linux partitions

Humm, first generation unix partitioning from IBM, or 5th generation partitioning from Sun (with help from Cray early on). BTW, a 10k can be in 16 partitions. No it doesn't require a domain to contain 4 processors--that's the max. A single board domain can have 1 i/o card, 1 cpu and some memory--typically a gig. The 15k and 6800 are similar, although the cpu/memory cards are typically maxed. It is *very* rare to find a company who would buy these sorts of systems to not max them out.

AIX 5L offers support for systems with up to 32 processors and 256 GB memory.

Wow, Solaris scales to 106 procesors in a single domain, with at least 1/2TB of memory. Besides, I'd bet there are more apps for Solaris than AIX.

*Note all of the quotes are from IBM's web page regarding the p690.

Re:Wrong Comparision

[eclarkso]

Humm, first generation unix partitioning from IBM, or 5th generation partitioning from Sun (with help from Cray early on). BTW, a 10k can be in 16 partitions. No it doesn't require a domain to contain 4 processors--that's the max. A single board domain can have 1 i/o card, 1 cpu and some memory--typically a gig. The 15k and 6800 are similar, although the cpu/memory cards are typically maxed. It is *very* rare to find a company who would buy these sorts of systems to not max them out.

Without addressing the rest of your post:

It has been well publicized that the p690 contains a number of features adapted from IBM's mainframe legacy. So "1st generation UNIX partitioning," while literally true, discounts 30 years of IBM big iron partitioning experience.

Addressing the rest of your post:

You are leaving out the IBM's performance claims that their processors vastly outperform Sun's. If nothing else, some benchmarks validate IBM's assertions.

Re:Wrong Comparision

[Jeffrey Baker]

The 15K has 106 CPUs, but they are the slowest CPUs on the market in SPEC benchmarks. The UltraSPARC III are slower than the 1300 MHz Power4, slower than the 833 MHz 21264B, slower than the 2000 MHz Pentium 4, slower than the 800 MHz Itanium in FP but faster in integer, and not any faster than the 1200 MHz Athlon MP. The only CPU that is slower is the 552 MHz PA-8600, and the UltraSPARC III barely beats it. The PA-8600 will be replaced imminently by the 750 MHz PA-8700, while the 833 MHz 21264B will soon be replaced by the astonishingly fast 1000 MHz 21264B.

But wait! you say, SPEC numbers aren't everything. Yes the 15K has some seriously inter CPU bandwidth and big-time scalability. Problem for the 15K is that the Power4's inter-CPU bandwidth makes the 15K look like a beowulf cluster running over appletalk. The Power4 has a two cores sitting right next to each other on the same die and can/does have four or more of these double cores wired together in the same package with 128 MB L3 memory bank. The Power4's system bandwidth is 92 GB/s, or 38 times higer than the UltraSPARC III.

It's time to euthanize the poor old UltraSPARC CPU line.

Re:Wrong Comparision

[Tower]

disclaimer: I'm an IBM employee (and have played with systems using the POWER4)

First of all, let me mention that the RS/6k S80 (two releases ago - prior to the p680) outdid the 10k, at reduced cost, with the previous generation of procs. 16 processors outdid 64 in many, many tests (including ones with real-world data movement).

As for partitioning... hmmm... let's think. IBM has been doing logical partitioning in AS/400 for a while, and on the S/390 (now the z-series) for quite some time... a few decades now. A lot of that experience went into this.

Regarding the POWER4:
Scalability: The eServer p690 is able to marshal up to 1,000 processors for high-performancesupercomputing duty, in applications such as Business Intelligence or seismic data interpretation. (think - the big supercomputers are right now POWER3, with several POWER4 systems in devel)

Raw power:
Our POWER4 processor can handle seventeen times more data than the UltraSparc III chip used in Sun's brand-new "Star Cat" top-of-the line F150000. Only 32 IBM processors outperform double the number of Ultra Sparcs - which draw much more power, create more heat and are less efficiently packaged than ours, which use modules developed for the eServer z900 mainframe. (The CPU numbers can be found in SPECmark, and from other benchmarks, including TPC-C, Javamarks, and some other fairly useless comparisons).

Re:Wrong Comparision

[Tower]

Just to back some more of that up:
Power4 (1.3GHz):
SpecInt2000: 783 - 808 (base - peak)
Specfp 2000: 1098 - 1169

UltraSPARC III (900MHz):
SpecInt2000: 438 - 467
Specfp 2000: 427 - 482

That, and the clustering of the cores on the MCMs allowing for massive inter-processor bandwith when all bound together makes quite a powerful machine.

Re:Wrong Comparision

[cartman]

I'm not meaning to flame, but your analysis was so flawed that I'm amazed it got moderated up to level 4. IBM's hardware is vastly superior to anything Sun has to offer.

First, even though Sun's E15k has 106 processors, only 72 of them are even directly connected to the memory fabric. The others are just PCI cards, with staggering latency, consequently they won't help transaction processing performance. The E15k actually has 72 usable processors; the others are there to impress people who measure system performance by "counting processors."

Second, the UltraSparc III is a notoriously weak performer. It can't even execute instructions out of order!! It is quite likely that IBM's POWER4 would outperform it by more than a factor of 2.

Third, although the p690 has only 32 processors, it has 64 cores. Each "processor" has two 4-way CPUs with a very low latency interconnect. The IBM product would be more accurately characterized as a 64-processor machine.

Fourth, the products from other Unix vendors (hp, ibm) always vastly outperform Sun's product with dramatically fewer processors. HP's new 16 processor box gets almost the same tpc rating as Sun's 64 processor E10k. IBM's old p680 with 24 processors almost doubled the performance of the E10k w/ 64 processors. Sun's most recent comments of "we've decided not to use industry-standard benchmarks any more" is likely because they always lose badly.

Sun's real benefit is in software (Solaris is way better than AIX and has more apps) and in the fact that they have one OS and one proc architecture (Solaris/Sparc) across their entire range of computers.

yeah...

[Anonymous Coward]

>absolutely nothing at all to do with the PPC

... apart from the fact that they share the same fucking instruction set, genius.

Re:Differences between PPC G4 and Power 4 ?

[Gill Bates]

Overview of the Power4 processor:

Power4 is the processor that will be used in the next-generation RS/6000 and AS/400 systems (IBM eServer i-series and p-series). It is a high-performance VLSI chip that includes two 64-bit PowerPC microprocessors, connected at high bandwidth to an on-chip memory subsystem consisting of a shared L2-cache memory plus the directory and interface for a large off-chip L3, and with high-speed busses and I/O to enable efficient 8-way systems to be built on a single 4-chip module. The microprocessors will operate at > 1 GHz clock frequency and have processor-L2 cache bandwidths of 100 GB/s. The Power4 chip is divided into 12 units, some of which are being designed by multi-site teams. The Research team focuses on all aspects of VLSI design as well as design tools and methodologies. For the Instruction Fetch and L2 Cache Control Units, the circuit and physical design of the logic circuits (about 2M transistors for each unit) are done in Yorktown, the array designs in Poughkeepsie, and the logic and verification in Austin. Performance exceeding 1GHz is achieved at acceptable power levels using mostly static, custom-designed CMOS circuits for the dataflow. Synthesized logic, implemented using circuit books from a standard cell library, is used for most control circuits. The circuits are designed to be fabricated in IBM's 0.18 CMOS 8S2 Silicon-on-Insulator technology with 7 levels of copper wiring.

Re:ppc java runtime?

[Anonymous Coward]

IBM would be very interested to hear that. You could save them a BUNDLE on all those PowerPC manuals....

POWER-1 was the blueprint for the original PowerPC 601. The 601 basically had the same instruction set, but a few instructions were removed (and handled in software). Some of the original Apple software was actually complied with xlC for the POWER instruction set (perf. wasn't great since a handful of instructions were trapped but it worked). IBM's POWER line of processors continued, but were modified to be PowerPC-compliant to the 64bit PPC spec.

Tom

Re:*Warning* Rumor...

[Paul Boven]

If the past decade, and especially the past few years have thought us anything, it's that it is no longer about what feels right, if it ever was.
And any company but especially a tech company couldn't care less about your feelings of loyalty and 'right', or those of their empolyees, if it impacts their bottom line.
In the capitalist economy, the ever shrinking margins have long since squeezed conscience out of the corporate picture.