Blade 2500 versus w2100

I would ask, which FDTD software package do you intend to use and which operating system do you plan to run? There is such a wide variety of FDTD software products ( both open source and commercial ) available for three operating systems ( UNIX, Linux and Windows ), it has to be one of the most important factors when choosing your hardware. In other words if you are planning to run a commercial or open source application on Solaris, the best choice would be the Blade 2500. If you are planning to run an open source application on both Solaris and Linux switching between the two environments, a better choice would be the JWS 2100z. You mention Solaris and Linux, but which is the primary OS?

On the other hand something you may want to consider if price is an issue is the quality of the machines. Many PC systems that I have experience with have very limited life span. For example I have run an in house comparison between the costs of running Dell Precision 530 workstation and Sun Blade 1000 workstation over a period of a year. Both machines were manufactured in 2001 and were subject to the same environment and upgrades over the one year test. Upgrades included: CPU, Memory and disks and there was a small amount of maintenance as well ( cleaning and testing ). At the end of the year the Blade 1000 was solid and functioning perfectly without any problems, but the Dell machine was literally falling apart and had no further upgrade options ( it was at it's limit @ 2GB RAM, the Blade had room for 6GB more!! ). Furthermore when I took account of the time involved with maintaining availability of the operating systems used ( Solaris and Linux ), Solaris proved exceptionally stable with little changes to the recommended patch cluster over the year. Red Hat Linux on the other hand was a nightmare, I implemented a similar patch strategy to how I was patching Solaris - non automated, with most of the patches made available by Red Hat the documentation was very very limited, greatly increasing the administrative overhead ( researching etc. ). Granted this test was carried out on used hardware, but the history of both machines was quite similar. If you look at factory configured workstations eg. Blade 2500 A39UXB2-C1AB-D2GDN @ 7100 and JWS W2100z A59-NGB2-9D-2GBDQ @ 4550the price isn't that much of a difference if you go on the life expectancy of each system. I would put a SPARC based workstation as providing availability over a 5 year period, but I would expect that a PC based system would give limited availability over a 3 year period. This would give a monthly figure for the hardware purchase @ 118.33 /seat on the SPARC and 126.37 /seat on the AMD machine. Also if you were not running Solaris on the AMD machine you would have to take into account the additional costs of a commercially supported Linux distribution or Microsoft Windows. The memory gap between SPARC and x86 machines has closed in recent years, but 32 bit applications cannot address more than 4GB RAM, ruling Windows out as an option all together. I hope this doesn't seem too much like a sales pitch or journalist bulls4!t - just some information I thought may be of interest to this thread!

mlennon at 2007-7-5 14:07:31 >

The software packages we will use frequently include Comsol Multiphysics (FEMLAB), HFSS, XFDTD and ABAQUS. We will also probably want to have programs like ADS, MATLAB, Mathematica, etc. installed.

I don't know about all of the packages, but I believe most use primarily single-thread processing and would not necessarily make use of dual processors (which doesn't mean I wouldn't want them for multi-application processing), but it is recommended that a faster single processor is better than two slower processors, for example. In this case, it seems like the performance benchmarks (of which I have very little knowledge) for integer and floating point operations (most programs use a bit of both) are better for the w2100 workstation:

CINT:

w2100 - 1 chip=16.7, 2 chips=32.5

2500- 1 chip=8.5,2 chips=17.1

CFP:

w2100 - 2 chips=35.2

2500- 1chip=13.9, 2 chips=27.8

It seems like I can have at least Solaris on either workstation, so OS reliability is equal. As for reliability of systems, I hesitate to put too much stock in Dell vs. Sun comparison since I am not even considering Dell at this point (mostly for that reason). What do you make of the numbers above? Is there any fundamental difference in the memory type or speed between the systems? Much of the processing is matrix storage and inversion, so LOTS of data needs to be moved to and from memory.

I guess one question that I can hopefully answer through the links provided (thank you!) is whether the SPARC processors are really going to offer any performance advantage. I know folks who run simulations on an older 1GB Dell station that take 5-6 days each. Even if we can reduce that to 4-5 days it would be a great help, since this computer will likely be shared between several people.

Thanks for all your advice. I'll take any opinions I can get at this point!

me_cb_24 at 2007-7-5 14:07:31 >

Here is a follow up on some of the memory features of both machines:

<b>W2100z</b>

Memory bandwidth of up to 6.4 GB/sec. per processor (with PC3200 modules)

Registered ECC DDR1-400 SDRAM support

Up to 16 GB (8 GB per CPU) in 8 DIMM slots (4 slots per CPU)

512 MB to 2 GB DIMMs support

<b>Blade 2500</b>

Memory bandwidth of up to 8.4 GB/sec. ( with PC2100 modules )

Registered ECC DDR1-133/266 SDRAM support

Up to 16 GB (8 GB per CPU) in 8 DIMM slots (4 slots per CPU)

512 MB to 2 GB DIMMs support

The 2100z has a feature that allows one CPU to access memory residing with the other CPU, though it reduces memory throughput it will allow you to use the maximum amount of memory with only one CPU installed. The Blade 2500 can only access a maximum of 8GB of memory with one CPU installed. I'll come back with more on the benchmarks, I have some good information available, but I don't have time to read up on it tonight.

mlennon at 2007-7-5 14:07:31 >