Intel Xeon and AMD Opteron Battle Head to Head

[pschmid]

Intel and AMD battle it out in the server market. Is Intel's new Woodcrest able to dethrone the mighty AMD Opteron? We compare two servers and get the facts for you.

 

[zenmaster]

It was a fair article.
They need to include more "server" related items.

Database results, How do the platforms handle 4- GB NICs.
Etc.....

Sorry, I'm not dong video or sound encoding on my servers.
I'm just glad they did not include Call of Duty results.
Too bad they did not add new tests.

 

[ElMoIsEviL]

Good article, but its a moog point. New technology beats old technology. Big news there. :roll:

So Opteron beating Netburst grade chips was a moot point as well? New Technology beating old technology?

I don't think so.

We bench what's available at anytime. But just to be a pain... this benchmark benche's 8th Generation Cores from both AMD and Intel thus is more valid then the Netburst vs Opteron benches.

 

[zenmaster]

Actually I just noted something about the power ratings.
I just reviewed that and noted that about 44w-88w (or even higher) should be added to all of the power numbers of the Xeon.

The only way to judge a processor is in part of a system.

I know that power usage is of extreme importance in many datacenters.
The lower power usage of the Xeon initially had me very excited.

However, this is now not the case if you read the fine print.
Still the new chip is impressive.

Yeah, it may not be news about new architecture beating old, but it does not mean you cant atleast look since it is not always the case.


My issues with the article are lack of true server type of benchmarks and very misleading graphics for power usage.

 

[Scarchunk]

Your a quick learner, exactly! A new technology processor is always going to be better than the old technology.

That is always going to be the case when benchmarking tech. In 07 when the new Opteron is released, it will be benched against whatever Intel has at the moment. Are you going to whine then?

 

[zenmaster]

And as we have seen from Intel, new is not better.

P4 had a hard time ever beating the PIII.

Over time the P4 just became clocked so much faster and had extensions added it won, but in the long run the PIII architecture was shown to be superior and the C2D is based more upon that than the P4.

 

[Doughbuy]

Meh, I don't listen to whiny fanboys. Intel still has a bit of issues with power, but they've come a long way in fixing it. I was also wondering why they released the Dempsey, although i doubt many people would buy it if they could choose between both Opteron's or Woodcrest cores...

Either way, Server market is interesting, but not interesting enough for me to keep up with too much.

 

[idekkers]

2 questions left unanswered :

1. intel is faster in desktop apps.... but what do these have to do with server CPU's ? where are the server benchmarks ?

2. where is the socket F opterons ? they are (thoretically) available for 2-3 month now, they use lower voltage memory...

 

[RichPLS]

Actually I just noted something about the power ratings.

I just reviewed that and noted that about 44w-88w (or even higher) should be added to all of the power numbers of the Xeon.
? ? ? ? ? ?

The only way to judge a processor is in part of a system.

I know that power usage is of extreme importance in many datacenters.
The lower power usage of the Xeon initially had me very excited.

However, this is now not the case if you read the fine print.
Still the new chip is impressive.

Yeah, it may not be news about new architecture beating old, but it does not mean you cant atleast look since it is not always the case.


My issues with the article are lack of true server type of benchmarks and very misleading graphics for power usage.

It must be noted that in the Intel systems, every FB-DIMM consumes nearly 11 W and has its own heat sink for cooling. At over 40 W, the RAM consumes a large amount of power and thus limits the benefits of the lower power usage of the Xeon Woodcrest CPUs.

 

[RichPLS]

plus the Intel system had 16MB ATI graphics, while AMD's had 8MB ATI graphics...

 

[RichPLS]

I would of liked to see the performance of two Xeon 6050's included too.

 

[zenmaster]

That is likely because you don't work with data centers.

Each Server brought in must be tested for peak and average usage.
Cooling costs, Battery Backup costs (No, not little dinky things, the ones that are pickup-size or bigger with lots of them side by sdie.)

When you have 500+ servers lined up and running and running side-by-side this is critical.

The point is not Intel vs AMD. The power usage of the 5100 is much better than the 5000. However, many data centers were switching to Optys because of the power savings. Those savings are much higher than what was shown in the graphics.

In regards to the 5100 series, the 5100s likely have the advantage over the Optys with 4Dimms and trail with 8Dimms.

If I were to take an average of 4 vs 8 dimms and assumed an average of 6dimms for my 500 servers I would be looking at 66w x 500 = 33,000 Watts.

This article is about Servers. Power Usage is all about Data Centers not single servers. The 33,000 Difference is extreme for cooling needs, battery backups, circuits, etc.....

Mind you this is not a critique of the 5100 since all it shows is that the 5100 system could be a little better or worse than the Opty depending on Config. The point is that it does not crush the Opty in Power as the charts would seem to show.

Before the 5100 I would not have considered a 5000 and in fact we switched to 100% optys. Now that 5100s are out, we would consider a move back. Howerver, much of the work we do is very memory and bus intensive. These tests do not cover that.

The information posted was useful, but it is clear that for the most part the testing was done from the perspective of a desktop system than a server system.

 

[zenmaster]

It critique was not in the written text.
It was in the graphics.

The information was correctly included in the text of the article.
However, if I were to post a poll in a new thread of readers of that article that did not read this thread I wager the majority would think that a 5000 Xeon system would be more efficient than an Opty System.

Most people skim over text, look for pictures, and stop.
They do not analyze.

I don't think it was a means of being deceptive.
You measured the CPUs themselves in lieu of the systems.

I can see why this was done so PowerSupplies, Disk Drives and other items would not be included. Perhaps your method to just note was best.
Perhaps an * should have been in the graphic itself since the differential is not minor.
---------------------------------------------------------------------------------
The other issue was with lack of testing the "system" as a whole in regard to real world use of the system.

Example - We have very large "VMware" systems with many OSes running at the same time putting large strains on Membory. We have (8) Gigabyte cards coming out of the systems.

For this type of usage, the Optys were far far superior in performance compared to the 5000s due to bus and memory controller related issues.
The areas in which the Opty excelled over the 5000 were never really tested and it is not seen if the 5100 closes the gaps in those areas.

----------------------------------------------------------------------------------
Again, I do not consider it a bad article but "fair".
My "Fair" may be what others call "good" since I am more conservative in my ratings. I don't "Grade" inflate so everything is an "A" until knocked down. It starts at a "C" with me and if it meets expectations it stays there.
It only moves up or down by exceeding or failing to meet those.

My "fair" rating is based upon the fact it is was indeed well written, but fails to address most of the questions I would have in regards to data center servers in how I see them used.

 

[Doughbuy]

I would think that this article really only compares these two cpu's with no real emphasis on complete systems which might change things quite a bit. All we can really say is that the Woodcrest outperforms the Opteron, but thats only chipwise. Platform wise it may be the other way around.

Comparing chips, its a decent article, but not as informative if anyone is planning to buy a new server off of this information.

 

[Bawz]

That article had nothing to do with workstations or servers. Other than the Autodesk there were no real world applications tested that would be used in environments that workstations or servers are purchased for.

Servers: As mentioned previously, database and web performance? That's what these machines would probably be used for, not Lame or DivX. How about for grins we through in some virtualization bench's if they exist.

Workstations: You would never catch a CAD designer using a workstation with a 16MB video card.

The only thing that the article showed us was that Intel is good at the bench's they have been running since the Duo's were released. It's just my opinion, but I think showing these machines running app's that they are designed for or with configurations that would most commonly be used might help give a better representation of their real performance.

 

[dumbdumb]

There are many different types of 'servers' -- computation server,
file servers, etc. Since the article seems to gear toward the 'CPU'
power, I assume it's OK to ask for more computation related
benchmarks. The benchmarks did include linpack benchmark.
But obviously got it from Intel as it stated 'This software is Intel
optimized.' I'd love to see one that compiled from source using
a generic compiler (such as gcc) and not a hand-tuned version.
Architecture-wise, I just don't see how Intel's single (or double)
bus based structure can scale well when it comes to quad-core
and beyond. Please show us multi-threaded benchmarks to see
how well it scales up (say, in a dual dual-core system). Thanks
for your effort.

 

[AngelDeath82]

When you have the opportunity to do so, I would be interested in seeing the following tested in comparison with these:

Opteron Socket F with ServerWorks chipset
Opteron Socket F with nForce 3600 Professional chipset

Some of the lower-level chips in each series as well to compare if the performance metrics scale based on processor speed or not.

Both 4GB and 8GB RAM configurations

Virtualization benchmarks:
- VMWare Workstation
- VMWare Server
- Microsoft VirtualServer

Database benchmarks:
- Web server (MySQL, MS SQL)
- Application server (MySQL, MS SQL, Pervasive.SQL)

The article was well written, and I appreciate seeing a server-centric review on THG. Props.

 

[dean7]

Well, I think this isn't really aimed at the people buying servers. It's more of an informative article that's meant for the desktop/workstation crowd. Most of the people buying servers don't read this type of article, and just buy whatever CDW/Insight/etc. sells them from HP/Dell/IBM/etc.

The people buying servers don't really care about these specs. They care about things like TCO, heat in data centers (which power consumption affects, but this wasn't a major point in the article), warranty information, support contracts, etc. So, I think this was an article just for us tech enthusiasts who want to know just to know.

 

[gOJDO]

Good article, but its a moog point. New technology beats old technology. Big news there. :roll:

LameNoobMike, go to your dumb forum for trollz and take your BS with you.

 

[Therlian]

Virtualization benchmarks:
- VMWare Workstation
- VMWare Server
- Microsoft VirtualServer

I would also like to see more virtualization benchmarks. Virtualization is a big thing now with companies offering a lot of advantages.

 

[zenmaster]

But if you looked at server Vendors recently both Sell both XEON and Opty lines of servers. The "support contracts" and TCO features of the servers would be about the same.

And yes, the folks buying these boxes do care about the stats.
How many of load-balanced server Xs do I need for my server farm for service Y. How many services can I host of Platform Z?

Sorry to say, but these questions are asked.

Now if you have a 20-use company with a box in the corner with a file shared and a USB printer hanging off it then no, these questions are not asked. But then again what you need is any old desktop PC with a Raid-1 Sata-II setup.

 

[theaxemaster]

Not many people who work with "servers" care about DivX encoding speed. Individuals might care, but they don't buy servers or workstations, now do they?

What I'd like to see is some compiling times, not mp3 encoding. Show me how long it takes to build X or OpenOffice, then I might find benchmarks useful. Show me compiling using x86-64 versus EMT64, these are real world differences. Hell, how about some CFD? Run some FE code and get back to me.

I'm glad for the Linpack test and the memory bandwidth tests, but the rest was useless.

 

[dean7]

Well, I work in IT for a health care organization with over 30,000 employees. We have thousands of servers. We have standard builds that we use, and I highly doubt the people who make the purchasing decisions are looking at the THG reviews. Obviously smaller outfits would be more "agile" and able to respond more quickly to emerging technologies. But, for my industry, we're not going to read some THG review and instantly decide to start buying strictly AMD-based servers.

So, my point was and is: this is more for enthusiasts and possibly for enthusiasts who work for small companies.

 

[zenmaster]

Sorry to hear that your IT department is not capable of making informed decisions.

I only work for very large organizations, some global in nature.
Yes, there are IT standards.
Those standards are based upon technical analysis.

Our servers only go into "Data Centers" and not behind people's desks.

A server does not go into the "Data Centers" without power and lan analysis.

There are often capacity guidelines so we know we can add 10 of model X or 5 or model Y.

I certainly hope nobodies life is lost when your server room goes down because of power-overloads or the fact that a critical server has coffee spilled upon it by a secretary.

 

[adder1971]

"The Woodcrest platform performed very well, exceeding our initial expectations."

Who was suprised by that?? Not me.