2CPU.com » News » February 2004 » Intel 3.2GHz Xeons Reviewed @ 2CPU.com!

After a long day at work, I still managed to find the motivation and energy necessary to finish my review of Intel's 3.2GHz Xeons. I was able to run these processors through quite a few benchmarks and I even managed to make some comparisons with a pair of AMD Opterons.

If we all sat down and played a game of word association and I suddenly spit out "Xeon" and sat back awaiting responses, what words would you quickly blurt out? Cache? Server? Reliable? Expensive? These are just a few of the responses that I would expect. As a product name, Xeon is as entrenched in the IT community as you can get. Hardware geeks like us and even those systems administrators that just get whatever's cutting edge from IBM without really knowing what's inside the box all know the name Xeon, and know it well.

If you'd like to take a look inside Intel's latest Xeon, read on.

ayt999 Posts: 3026 Joined: 2002-10-28

#26587 Posted on: 02/04/2004 10:30 AM thanks a whole lot for your dedication to 2CPU and all the time you put into it. the P4 architecture is scaling very well and the addition of the L3 cache does bring some performance increases. too bad there wasn't a 3.2Ghz xeon without the L3 cache to compare the effects of just the L3... I just don't like the pricetag. nice review, keep them coming. SETI@home|folding@home|RC5|find-a-drug|DPAD|ZetaGrid|climateprediction.net| communityTSC|BOINC|Predictor@home|seventeen or bust|ECC2-109

EM Posts: 4110 Joined: 2000-09-04

#26588 Posted on: 02/04/2004 10:54 AM Nice an concise. I like it. Too bad I don't like the price. :eek: Political Correctness is a doctrine, fostered by a delusional, illogical, liberal minority and rabidly promoted by an unscrupulous mainstream media, which holds forth the proposition that it is entirely possible to pick up a turd by the clean end. Artificial intelligence is no match for natural stupidity. [URL="http://www.heatware.com/u/4993"]Heat[/URL]

SalemTheCat100 Registered User Posts: 1 Joined: 2004-02-04

#26589 Posted on: 02/04/2004 01:45 PM Why Handicap the Opteron System then use them to justify that the latest Xeon's are the greatest. Come on a pair of 246's with pc2700 ddr. The 248's and ddr 3200 have been out for quite a while. Salem

Forge Registered User Posts: 720 Joined: 2001-05-12

#26590 Posted on: 02/04/2004 02:22 PM Hooz only had PC2700 reg/ECC when he tested the Opterons. We don't have 248s to use, yet (You'd have to have some swing at AMD or Paypal me a boatload of cash to fix that), but I do have 4*512MB of PC3200 reg/ECC courtesy of OCZ. I'll be rerunning all the tests on the correctly pimped out Opterons just as soon as my EPS12V PSU shows up. Have no fear! Registered Linux user 82133 (li.org has a short memory)

Tessel8 Registered User Posts: 583 Joined: 2001-02-05

#26591 Posted on: 02/04/2004 02:30 PM The higher the set-associativity, the higher hit rate you can expect. Forge was kind enough to chip in his "uber-n00b" definition: "The higher the associativity, the more likely that data you're looking for is there." Hi. You might want to double check this statement. Higher associativity increases the chances of capacity misses, which will reduce the chance that you will find the data you are looking for in a cache, and will generate additional bus traffic. Associativity is a trade off, just like other variables in cache design. There is no "one best". Tessel8 - Game System - Misc System - Workstation *Not speaking for anyone but myself :-)

Jim_ Administrator Posts: 3574 Joined: 2000-03-15

#26592 Posted on: 02/04/2004 06:24 PM Originally posted by Tessel8 Hi. You might want to double check this statement. Higher associativity increases the chances of capacity misses, which will reduce the chance that you will find the data you are looking for in a cache, and will generate additional bus traffic. Associativity is a trade off, just like other variables in cache design. There is no "one best". Tessel8: I've done a lot of reading on cache design and I've never read that a higher set-associativity can actually have a negative impact on performance. Can you hook me up with some documentation? Thanks. With respect to the PC2700 used with the Opterons, it's all we had at the time. Forge just received Registered/ECC PC3200 from OCZ last week but until he gets his EPS12V PSU from PCP&C, we'll have to use the benchmarks that we have. I'll update the graphs once we have the updated numbers. [url="http://www.2cpu.com"][size=1]2CPU.com[/url] - Because two are always better than one! [url="http://www.jimkirk.org"]jimkirk.org[/url] - Not a Myth any Longer. Just a Dad.[/size]

Jim_ Administrator Posts: 3574 Joined: 2000-03-15

#26593 Posted on: 02/04/2004 06:30 PM Originally posted by Gandalf It would have been well to see the Linpack results for the Opteron; why weren [url="http://www.2cpu.com"][size=1]2CPU.com[/url] - Because two are always better than one! [url="http://www.jimkirk.org"]jimkirk.org[/url] - Not a Myth any Longer. Just a Dad.[/size]

spikegifted I do what? Posts: 6131 Joined: 2000-07-24

#26594 Posted on: 02/04/2004 08:56 PM Originally posted by Jim_ Can you hook me up with some documentation? Thanks. There are some interesting academic studies, based on simulations.... Cache Simulaton Performance tradeoffs in cache design Understanding CPU Caching and Performance (Ars) 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...

LRSeriesIII Aspiring Rocket Scientist Posts: 1120 Joined: 2002-08-29

#26595 Posted on: 02/05/2004 01:33 AM Very nice review. I look forward to seieng more when Forge gets his PSU. Originally posted by SalemTheCat100 Why Handicap the Opteron System then use them to justify that the latest Xeon's are the greatest. Come on a pair of 246's with pc2700 ddr. The 248's and ddr 3200 have been out for quite a while. Salem I do not think Jim said anywhere that the Xeons are flat out the best dual platform available. The Opterons in the review were not the fastest AMD has to offer, but they still provide a very useful comparison. Personally I find it very annoying when a website or magazine focuses all of their review attention on the fastest hardware available. Sure, it is interesting to see how the flagship products from the various major manufacturers compare to each other, but I am generally more interested in how that performance comparison looks across each manufacturer's entire product line. Opteron 246s and PC2700 RAM should not be treated as though the do not exist simply because 248s and PC3200 are available. After all, most people buy hardware that is at least a notch or two below the top of the line, so what good is a review for those people if it only covers the very best hardware available. Personally I am interested in seeing well done, consistent benchmarks for as many different system configurations as possible, because I like to have an understanding of how and why each variable affects or does not affect the performance of a piece of hardware for each situation. ->Computers ->Folding for team 3074

i_wolf labhair dom as gaelige Posts: 2034 Joined: 2002-11-19

#26596 Posted on: 02/05/2004 08:58 AM nice review Only thing i would have liked would have been a linux comparison where 64bit Opteron running linux compared with 32 bit Xeon and 32 bit mode opteron. Also a comparison as to what speed increases are yielded by compiles with ICC and how well it benefits Opteron. Other than that, nice work and congrats. Kind Regards i_wolf Hung like a donkey. Go like a horse!

Forge labhair dom as gaelige Posts: 720 Joined: 2001-05-12

#26597 Posted on: 02/05/2004 09:03 AM Only thing i would have liked would have been a linux comparison where 64bit Opteron running linux compared with 32 bit Xeon and 32 bit mode opteron. Also a comparison as to what speed increases are yielded by compiles with ICC and how well it benefits Opteron. Alright, alright, I'm WORKIN on it! Everything you mentioned is actually on my list of things to do. Registered Linux user 82133 (li.org has a short memory)

zathrasater Registered User Posts: 339 Joined: 2003-08-24

#26598 Posted on: 02/05/2004 11:27 AM Interesting read, and I look forward to seeing the 248s added to the mix. Nitpick on the history section which I'm sure is just an oversight: Ignoring the 133FSB versions, PIII Xeons came in two families: the 700/900s which (as stated) were available with 1MB or 2MB L2 cache, and the earlier 500/550s which were also available with 512k cache in addition to 1MB and 2MB (the same options as for the PII Xeons).

Jim_ Administrator Posts: 3574 Joined: 2000-03-15

#26599 Posted on: 02/05/2004 07:28 PM Originally posted by zathrasater Nitpick on the history section which I'm sure is just an oversight: Ignoring the 133FSB versions, PIII Xeons came in two families: the 700/900s which (as stated) were available with 1MB or 2MB L2 cache, and the earlier 500/550s which were also available with 512k cache in addition to 1MB and 2MB (the same options as for the PII Xeons). Not so much an oversight as an omission. I was trying to keep that section as brief as possible so I just touched on the highlights. With respect to the 133MHz FSB slot-2 P3 Xeons... I mentioned that line, but didn't really talk about them much since I hardly even consider them Xeons. They only scaled to DP, came with 256KB of L2 cache and were basically just Slot-2 Pentium III's. [url="http://www.2cpu.com"][size=1]2CPU.com[/url] - Because two are always better than one! [url="http://www.jimkirk.org"]jimkirk.org[/url] - Not a Myth any Longer. Just a Dad.[/size]

i_wolf labhair dom as gaelige Posts: 2034 Joined: 2002-11-19

#26600 Posted on: 02/05/2004 09:18 PM Thanks for the replys guys. Can't wait for the review update! Another idea that I had was if it would be possible to show some bang per buck measurements for each of the different apps. Just that a lot of people will buy a machine because it can excel in a particular area that they work in, so this woudld be invaluable for somebody looking for their next mathematic 5 platform or 3dSMax box. I was figuring just a real simple excel chart where you divide the score of the app by the cost of the machine. But if there was one per app it would be really useful and isn't something that I see often done. Not sure how much extra work would be involved though? :S Regards i_wolf Hung like a donkey. Go like a horse!

zathrasater Registered User Posts: 339 Joined: 2003-08-24

#26601 Posted on: 02/06/2004 08:38 AM Originally posted by Jim_ Not so much an oversight as an omission. I was trying to keep that section as brief as possible so I just touched on the highlights. With respect to the 133MHz FSB slot-2 P3 Xeons... I mentioned that line, but didn't really talk about them much since I hardly even consider them Xeons. They only scaled to DP, came with 256KB of L2 cache and were basically just Slot-2 Pentium III's. That's fair - I realize that's far from the point of the article, and I'm sure you're more familiar with the Xeon lineup than I. As for the 133FSB chips I generally dismiss them as well as having no redeeming features. Xeon prices without extra performance or scalability. My comment was only directed at the distinction between the 100FSB PIII Xeons. Now... when can you bench some 248s against these?

Forge Registered User Posts: 720 Joined: 2001-05-12

#26602 Posted on: 02/06/2004 08:43 AM Well, we'd have to get some first. Registered Linux user 82133 (li.org has a short memory)

i_wolf labhair dom as gaelige Posts: 2034 Joined: 2002-11-19

#26603 Posted on: 02/06/2004 06:44 PM I was just thinking, out of interest, do you guys have to fork out of your own pockets for the software and hardware to bench etc... or is it a case that the majority of the time you get sponsored or intel/amd give you hardware to bench?? Hung like a donkey. Go like a horse!

Hooz Administrator Posts: 2388 Joined: 2000-03-29

#26604 Posted on: 02/06/2004 07:29 PM Most of the time we get the products directly from the manufacturers. In the case of AMD, they sent us Opteron 246s and once we get some Opteron 246 reviews under our belt we'll see about getting some 248s. Unless you know of a place where the mystical Opteron tree grows (right next to the PC3200 Registered ECC DDR tree?) and we can pick our own. [size=1][url="http://www.2cpu.com"]2CPU.com[/url] - Because two are always better than one! Are you folding? [url="http://folding.extremeoverclocking.com/team_summary.php?s=&t=3074"]2CPU.com Folding@Home Team[/url] [url="http://www.heatware.com/eval.php?id=19979"]My Heatware[/url][/size]

ambit LoveYourFruit Posts: 1243 Joined: 2000-10-15

#26605 Posted on: 02/06/2004 07:45 PM Originally posted by Hooz Unless you know of a place where the mystical Opteron tree grows (right next to the PC3200 Registered ECC DDR tree?) and we can pick our own. AMD's/Crucial's shipping docks? *winkwink*

2CPU.com » News » February 2004 » Intel 3.2GHz Xeons Reviewed @ 2CPU.com!