ANANDTECH: a decent Opteron Review

[mr_gobbledegook]

I found Tomshardware review of the Opteron absolutely appalling !! I admit I myself dont know that much about benchmarking but jesus christ his setup and tests were totally messed up. Even the benchmarks he used dont have any significance to 'real world' performance. The benchmarks he used were purely scienfitic and do not translate into anything. They are just a bunch of figures with no real meaning.

For a proper review I suggest you visit <A HREF="http://www.aceshardware.com" target="_new">Aces Hardware</A> or even better <A HREF="http://www.anandtech.com" target="_new">Anandtech</A>.

(Anandtech uses real world examples of just what Opteron is capable of with benchmarks that actually mean something)

<font color=purple>Ladies and Gentlemen, its...Hammer Time !</font color=purple>

 

[juin]

The best is that evertone benchmark 1.8 GHZ model when it will not avaible until June where new xeon will be there and Madison Super score will be out.Only a 1.6 will have been too slow so they benchmark a a paper lauch.

According to anandtech intel will release a 1MB Xeon DP for next month so before 1.8GHz is release.

[-peep-] french

 

[bikeman]

Why should Anand's review be more reliable then THG's? Ok, the benchmarks themselves are indeed real-world examples, but ...
1) They don't tell us what hardware they use.
2) They use only one type of webserver and DB-server, which might favour one of the two architectures. I really don't know much about it, but it look to me like testing a game-macine by using only for example Return to Castle Wolfenstein. But you'll also have to test other games in order to get a complete overview.

So far my comments, because for the rest, Anad's review is quite open, I think.

Greetz,
Bikeman

<i>Then again, that's just my opinion</i>

 

[reever2]

Opteron 244 will not just be availible in June, what will be availible in June is servers from people like Newisys and Fujitsu that use 2 and 4 way opteron cpus


http://www.pcprogress.com/product.asp?m1=pw&pid=AMDOSA244BOX

 

[imgod2u]

They specifically mentioned that "all hardware used were the same when possible". And as for the review. They used 3 different DB servers. If you look at tomshardware's review, it was two SQL servers with different commands (commands which are rarely used may I add).

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

 

[mr_gobbledegook]

Part 1 of the review only covered 'Enterprise Performance' granted they could have used more than one database platform but I suspect the differences would not really be that significant.

Anandtech have still yet to publish Part 3 & 4 of thier review. This will cover '1U Server Performance' and 'Workstation Performance'. All 4 parts should combine to give one of the best <b>meaningful</b> reviews of Opteron on the Web so far.

<font color=purple>Ladies and Gentlemen, its...Hammer Time !</font color=purple>

 

[eden]

So far, by Anandtech's results, I am more impressed somehow. We see only 32-bit performance here as well.
Yes there is also the lack of Xeon MP 1MB 3.06GHZ, which does impair things.
However in the total pages processed test, it is simply nothing short of amazing. A true web server.

--
This post is brought to you by Eden, on a Via Eden, in the garden of Eden.

 

[imgod2u]

It definitely shows how a low-latency memory system and a flexible interconnect like Hypertransport can really help the server market. THis is actually one of the reasons single-CPU performance matters very little in the server market. SMP scalability is much more important.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

 

[lhgpoobaa]

Also, check out aceshardware... they have also done a very thourough and detailed opteron review.

<b>Now can someone explain how we ever got the idea that baby bunnies lay multicolored eggs made of chocolate in our living rooms?" :lol: </b>
<A HREF="http://members.iinet.net.au/~lhgpoobaa/images" target="_new">http://members.iinet.net.au/~lhgpoobaa/images</A>

 

[juin]

We can allready conclude that with most of software for webserver file server data base server opteron will allwayse be faster that a Xeon Nw core if not Nocona too.

Only few exception appear that Xeon can pull it out.There too much branch prediction on a webserver app for P4 core.

I wonder if 4 CPU Banias from HP can be faster that the 2 way from newsys.

[-peep-] french

 

[eden]

It could be, but again it all depends on HT scaling.

Which brings me to another interesting finding.
Logically, since HT allows excellent scalability and flexibility of added CPUs, that means that once you continue adding CPUs, the lead the Opteron will have will be even more significant. So 2-way only shows a fraction of what Opteron's damage can be.

I am honestly surprised overall at Opteron. This is the only product this year so far that has not disappointed me, out of all the products and technologies out from any hardware company. Well, yeah it disappointed me in Workstation, especially but ESPECIALLY SSE2, when you remember that Opteron's bandwidth is aplenty, at 5.4GB/sec, which makes you just wonder how the heck does the P4 win more with SSE2, not to mention Opteron DOUBLING SSE2 register amount! Sometimes it also felt like the Barton 2500+ was easily playing against the Opteron, when the latter has a damn powerful memory controller. But, I think you're right Juin, lack of driver and compiler optimization easily hurts such a new chip.

But again, for a new core, it most definitely does not do badly. Imagine an Opteron against a similar clocked Pentium 4, such as a 1.8GHZ 533MHZ Xeon underclocked. OUCH!

--
This post is brought to you by Eden, on a Via Eden, in the garden of Eden.

 

[imgod2u]

The P4 with an FSB800 scores better in SSE2. It's not surprising as SSE2 is mainly data-driven. Notice that a P4 with FSB533 (which supposedly brings 4.2 GB/sec) scores similar to an Opteron, but you crank up the FSB, it scores better. This suggests that it is very data-driven which is really just memory bandwidth limited.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

 

[juin]

adding CPU increase lantency on opteron not on Xeon but bandwith dont scale you just need a Quad channel DDR with point to point link like Redwood from rambus can be use having much less lantency that intergrade controleur from AMD and fixe lantency

[-peep-] french

 

[Schmide]

Could you elaborate...Opteron has independent memory controller per CPU and 3 hyper-transport links for inter CPU communication. Why would having any amount of CPUs in a system increase latency? Xeon on the other hand shares memory subsystems and thusly would be affected by the amount of CPUs sharing the resource.

Dichromatic for your viewing plesure...