· Content
· News
· Articles
· Mailinglists
· Knowledgebase
· Trouble Tickets
· Files
· Glossary
· Links
· Compatibility Lists
· Forums
Welcome to our website
To take full advantage of all features you need to login or register. Registration is completely free and takes only a few seconds.
Xeon vs Opteron: Memory and Cache Considerations
Posted by: Jim on: 08/01/2003 04:43 PM [ Print | 9 comment(s) ]
The Inquirer is once again discussing the merits of Opteron vs. Xeon. Today, their main point of discussion revolves around the memory sub-system.
So, currently, a dual XeonDP design would have 4.3 GB/s of memory bandwidth, while a dual Opteron would top it 10.6 GB/s total memory bandwidth, quite an advantage on paper. A quad XeonMP system would have to contend with a 3.2 GB/s FSB pipe towards its memory, while a quad Opteron would have four dual-lane memory highways, reaching a total of 21.2 GB/s.They see Intel bumping the Xeon's bus speed to 800MHz as a short-term fix. Take a look.
« IBM Launches Opteron-based eServer 325! · Xeon vs Opteron: Memory and Cache Considerations
· SMTP Trusts too Much »
Comment
|
Big B Psychic or Psycho? Posts: 3631 Joined: 2001-07-03 |
#24068 Posted on: 08/01/2003 10:12 PM
Xeon's don't suck by any means, but I'd have to agree that speeding up the FSB and adding extra cache are somewhat of a quick fix---at least with the currently available Xeon chipsets. I'm sure Intel has a higherbandwith chipset, but they're still working out the kinks. Hardware Forums|2CPU|My DVD's|Heatware=BigB Gigabyte NF650SLI|2GB OCZ PC6400|Pentium E 2160|GF7900GS (SLI)...XP SP2 MSI P965 Neo-F|1GB SuperTalent PC2-5300|PentiumD 920|Radeon X1650...XP SP2 |
Comment
|
spikegifted I do what? Posts: 6144 Joined: 2000-07-24 |
#24069 Posted on: 08/02/2003 04:43 AM
I think, somewhere along the line, Intel will integrate the memory controller (or hub or whatever it is called these days) onto the core of the processor... When? I don't know...  spikegifted.net "At the very beginning of the 19th century chemists generally thought that compounds from living organisms were too complicated in structure to be capable of artificial synthesis from non-living things, and that a 'vital force' or vitalism conferred the characteristics of living beings on this form of matter." Well, how wrong were they? So much for dogma... |
Comment
|
ShaneBrooks Unregistered |
#24070 Posted on: 08/02/2003 05:34 AM
I wonder how often the memory bus is actually saturated. |
Comment
|
happyfox Aspiring Duallie Posts: 93 Joined: 2002-11-13 |
#24071 Posted on: 08/02/2003 07:11 AM
In multiprocessor system, the memory bandwidth of Opteron system is widely varied based on the design of a motherboard. In most of ATX sized dual Opteron mobo, all the memory sockets are connected to one processor and the other processor is connected to the memory through the hypertransport link between two processors. In this case, they can not provide such high memory bandwidth mentioned in the article. |
Comment
|
spikegifted I do what? Posts: 6144 Joined: 2000-07-24 |
#24072 Posted on: 08/02/2003 03:53 PM
spikegifted.net "At the very beginning of the 19th century chemists generally thought that compounds from living organisms were too complicated in structure to be capable of artificial synthesis from non-living things, and that a 'vital force' or vitalism conferred the characteristics of living beings on this form of matter." Well, how wrong were they? So much for dogma... |
Comment
|
soliddiy SMP Newbie Posts: 4 Joined: 2003-04-30 |
#24073 Posted on: 08/02/2003 05:01 PM
Opteron has wider bandwidth than XEON, so of course opteron average memory access time should be shorter then XEON even its FSB is 800Mhz, Opteron has wider bus lines, it can fetch more bytes in one memory access than XEON, so it needn't as high freq as XEON on FSB. |
Comment
|
knight0334 Lord Vader Posts: 354 Joined: 2001-08-01 |
#24074 Posted on: 08/03/2003 07:40 AM
from what I'm looking at, each K8 has its own memory bus and dedicated banks to read/write to. Heatware= knight_0334 | eBay= knight0334 Graybeards Outdoors |
Comment
|
Tairc Will work for bandwidth! Posts: 117 Joined: 2000-03-01 |
#24075 Posted on: 08/05/2003 07:51 PM
It may have the /ability/ to be connected such that each core gets it's own bus connection, but due to wiring costs and area costs, its much easier to simply let one of the CPU's go unconnected to memory, and perform all of it's accesses through the hypertransport link to the other CPU (which is connected to memory) |
Comment
|
duraid SMP Qualified Posts: 387 Joined: 2002-03-31 |
#24076 Posted on: 08/07/2003 04:17 AM
...that the 875 chipset gives a P4 lower latency access to memory than opteron, even though opteron's memory controller is _on_ the opteron chip. If you don't need 64 bits, there really is no reason at all to buy opteron. |