TM1 Forum

Just wanted to ask what peoples views were on Processor speeds.

It seems to me that the current trend for many processors running at 3GHz does not serve a product like TM1 well. The max speed of a processor seems to be stalled around the 3GHz mark and the hardware industry seems to be focussed on multiple processors in a box rather than the speed.

Obviously these type of servers have their place but I'd much rather have a twin 4GHz than 8 3GHz processors as I think (guess really) my user base would be better served by a server like this.

Anyone else have comments or views?

Cheers,

I guess for TM1 you are right. (Based on the current architecture) May be instead of looking at the frozen speed of modern CPU's, IBM should instead work on changing how TM1 handles multiple threads. The CPU and software industry have been moving this way for some time. Come to think of it where is a version of TM1 that works on Linux? Many businesses are moving their servers over to the likes of Red Hat (Especially Solaris based companies).

I guess it comes down to the CPU architecture. There is a cracking article on wikipedia about CPUs (for uber-geeks only!) but the general issue seems to be greater clocks > greater heat. From 3GHz onwards you are looking at serious cooling solutions and in a server room where rack space is limited i guess that starts to be a factor in CPU choice. Its generally a better solution for most programmes / systems to go multi-threading rather than push the cycles as that reaps better rewards compared to the issues with greater clocks. Its just the way TM1 works, it cannot utilise the architecture as efficiently so it benefits less.

Victim of its own success maybe?

From what I rember it's top do with the number of silicon wafers on one die. The more you have, the more it heats up. As for TM1, as I said above it needs to handle multi-thrweading better. Sitting there watching task manager use all of one cpu while the others are idle can be most distressing. They have managed to solve locking, why not multi-threading next?

this could get horribly technical, but i can see why they haven't

threading is most useful when you have multiple tasks to do that are independant of each other. Unless you have a TM1 model where no cubes have rules or TIs that make them reliant upon each other, or share the same dimensions, i can't see how they can make proper use of threading. It would have to try and calculate 2 cubes until it found a dependancy on something then wait before continuing. How the blazes could they work that out when they can't write automatic feeders?!

Good point. They may have to sort both problems if they are going to utilise the product as they (IBM) have in mind. Here's to holding my breathe. (Not really)

Yes better multi-threading is OK up to a point, but it's only useful for
1. Deliver cached numbers to many users at the same time.
2. Calculating independent set of numbers.

In general terms 2 isn't that common in my experience. We are still left with a lack of raw grunt power these days, especially as we move into 64 bit world with larger models based on more detail.

I'm really disappointed with the speed of my current hardware and all it's doing is adding consolidations no rules or anything.

Can you even buy "fast" processors???
Cheers

It's called £800 of quad core with water cooling. Your boss think your building a gaming rig though!!