AMD Processors
Decrease font size
Increase font size
Topic Title: Which is better for Floating point performance?
Topic Summary:
Created On: 02/16/2007 05:35 AM
Status: Read Only
Linear : Threading : Single : Branch
Search Topic Search Topic
Topic Tools Topic Tools
View similar topics View similar topics
View topic in raw text format. Print this topic.
 02/16/2007 05:35 AM
User is offline View Users Profile Print this message

Author Icon
Shielder
Senior Member

Posts: 303
Joined: 10/07/2003

Sorry, I have to ask the question. I'm going to be building a system that will be doing very intensive FORTRAN calculations (specifically, point-kernel calculations and also some Monte Carlo code calcs) that will involve both integer and FP calculations. I intend (although I may not be able to due to code considerations) to build the system on 64-bit Linux, probably RH EL 5 or Fedora.

The codes can be memory hogs, so I'll be using at least 4GB of memory. Which processor architecture would you recommend? I would prefer to go with the Opteron, but we need raw performance, and from what I've seen, the Core architecture seems to be faster in most of the benchmarks.

I've not seen any integer or FP benchies though, so I've come here for some guidance.

Hope you can help.

TIA.

Andy

P.S. I prefer AMD processors, but I will go with the best, regardless of the brand (I'd even get a cyrix chip if it was the best!

-------------------------
Athlon XP 2500+ (stock) on an Asus A7N8X with 1Gig Crucial PC3200, 30GB Maxtor and 60GB Seagate Drives (both ATA100), Asus 9560 GeForce FX5600 graphics card, on board sound and dual booting with WinXP Pro and SuSe 9.2 Pro with a 400W Q-tec PSU. Using Firefox and Thunderbird for internet and emails and OpenOffice.org for work.
 03/02/2007 11:22 PM
User is offline View Users Profile Print this message

Author Icon
MU_Engineer
Dr. Mu

Posts: 1837
Joined: 08/26/2006

quote:

Originally posted by: Shielder
Sorry, I have to ask the question. I'm going to be building a system that will be doing very intensive FORTRAN calculations (specifically, point-kernel calculations and also some Monte Carlo code calcs) that will involve both integer and FP calculations. I intend (although I may not be able to due to code considerations) to build the system on 64-bit Linux, probably RH EL 5 or Fedora.

The codes can be memory hogs, so I'll be using at least 4GB of memory. Which processor architecture would you recommend? I would prefer to go with the Opteron, but we need raw performance, and from what I've seen, the Core architecture seems to be faster in most of the benchmarks.

I've not seen any integer or FP benchies though, so I've come here for some guidance.

Hope you can help.

TIA.

Andy

P.S. I prefer AMD processors, but I will go with the best, regardless of the brand (I'd even get a cyrix chip if it was the best!



Here are the strong points of each CPU:

Dual-core Opteron 12xx/22xx/88xx
1. Will scale better in dual- and quad/eight-socket situations than Intel chips will, especially quads and eights.
2. Better performance in non-SSE floating-point calculations. But those are not all that common unless you specifically compile code not to use SSE.
3. Less expensive to set up a dual Opteron box than a dual Xeon box due to all parts (CPUs, motherboards, memory) costing less. Also, the Opterons will use less power.
4. Dual Opterons perform better than dual dual-core/quad-core Xeons and the Xeon 33xx single-socket quad-core CPUs in highly memory-intensive applications due to greater system I/O bandwidth.

Xeon 3100/5100 dual cores and 3300/5300 quad-cores
1. Much better SSE floating-point and integer performance per clock speed.
2. Higher core density per socket as Intel has quad-core chips. Also, a single Xeon 33xx chip + desktop motherboard is less expensive than two Opterons and a DP workstation motherboard.
3. Overall better performance per clock speed, and Intel's clock speeds exceed those of the Opterons. So maximum attainable performance will likely be higher with Xeons over Opterons.

It's basically a question as to how multithreaded and memory I/O intensive your application is. If it just has a large working set and not a whole lot of memory I/O (like a large fluid dynamics project I helped to design a system for) then system bandwidth is not that important and the Xeons will run faster than the Opterons will. If the program is also not multithreaded, the Xeons will be faster. The Opterons will only run better for a program that's very multithreaded and slams the memory I/O heavily.

However, AMD's upcoming K10-based Opterons (codenamed "Barcelona") are supposed to outperform Intel's Xeons clock-for-clock by a significant margin. These chips are quad-core and are supposed to ship in May or June and are compatible with current socket AM2 or Socket 1207/F motherboards. If you have a very multithreaded application and you don't need a new machine RIGHT NOW, then wait for the Barcelona Opterons (numbered 226x) as they will be the fastest processors for your usage.

-------------------------
Statistics
112018 users are registered to the AMD Processors forum.
There are currently 0 users logged in.

FuseTalk Hosting Executive Plan v3.2 - © 1999-2014 FuseTalk Inc. All rights reserved.



Contact AMD Terms and Conditions ©2007 Advanced Micro Devices, Inc. Privacy Trademark information