Intel Woodcrest: the Birth of a New King
by Jason Clark & Ross Whitehead on July 13, 2006 12:05 AM EST- Posted in
- IT Computing
Multiple Load Points
For AnandTech Database Benchmarks, we have always focused on "real world" Benchmarks. To achieve this, we have used real applications with loads such that CPU utilization was 80-90%. Recently we discussed how most Enterprise Database Servers do not average 80-90% CPU utilization, but rather something closer to the 30-60% range. We thought it would make more sense to show performance where it is most likely going to be used, as well as the saturation numbers for the situations where the CPU is maxed. We feel this is consistent with how GPUs are reviewed, and how you might test drive a car. With GPUs, the cards are tested with varying resolutions, and anti-aliasing levels. With a car, you don't just hit the highway and see what the top end is.
We settled on six load points for testing. These load points are consistent across all platforms and are throttled from the client, independent of the platform being measured. We chose these load points as they split the load range into 6 roughly equal parts and allow us to extrapolate data between the points. The last/highest load point is a "saturation plus" load point to verify that we tested up the capability of the CPUs.
For any given load point, there is a defined number of threads. Each test is 20 minutes in duration, which includes an 8 minute warm up period followed by a 12 minute measured period. For a given load point, the client submits requests to the DB server as fast as the DB server will respond. The rate which the client is able to submit requests is measured during the final 12 minutes of the test and averaged to determine the Orders/Minute for Dell and Transactions/Minute for Forums. After much blood, sweat, and almost tears we were able to produce repeatable loads with an average deviation of 1.6%.
For each platform we ran the test 5 times for each load point and then averaged the 5 results. This was repeated for all loads, all tests, on all platforms... that is 300 test executions!!! (We won't even get into the debugging issues we had to deal with prior to the final results.) Thankfully, we managed to automate the process as much as possible when implementing the throttling mechanism for the load points.
For AnandTech Database Benchmarks, we have always focused on "real world" Benchmarks. To achieve this, we have used real applications with loads such that CPU utilization was 80-90%. Recently we discussed how most Enterprise Database Servers do not average 80-90% CPU utilization, but rather something closer to the 30-60% range. We thought it would make more sense to show performance where it is most likely going to be used, as well as the saturation numbers for the situations where the CPU is maxed. We feel this is consistent with how GPUs are reviewed, and how you might test drive a car. With GPUs, the cards are tested with varying resolutions, and anti-aliasing levels. With a car, you don't just hit the highway and see what the top end is.
We settled on six load points for testing. These load points are consistent across all platforms and are throttled from the client, independent of the platform being measured. We chose these load points as they split the load range into 6 roughly equal parts and allow us to extrapolate data between the points. The last/highest load point is a "saturation plus" load point to verify that we tested up the capability of the CPUs.
For any given load point, there is a defined number of threads. Each test is 20 minutes in duration, which includes an 8 minute warm up period followed by a 12 minute measured period. For a given load point, the client submits requests to the DB server as fast as the DB server will respond. The rate which the client is able to submit requests is measured during the final 12 minutes of the test and averaged to determine the Orders/Minute for Dell and Transactions/Minute for Forums. After much blood, sweat, and almost tears we were able to produce repeatable loads with an average deviation of 1.6%.
For each platform we ran the test 5 times for each load point and then averaged the 5 results. This was repeated for all loads, all tests, on all platforms... that is 300 test executions!!! (We won't even get into the debugging issues we had to deal with prior to the final results.) Thankfully, we managed to automate the process as much as possible when implementing the throttling mechanism for the load points.
Facebook
Twitter
RSS
59 Comments
View All Comments
Lonyo - Thursday, July 13, 2006 - link
What's the fastest Opteron dual core CPU you can buy?What's the fastest Woodcrest CPU that will be released?
AMD don't make anything faster than 2.6GHz, so it doesn't really matter what speed Intel have to be at to beat it, they beat it with their top end part. And the Opteron is nearing its end (at 90nm), Woodcrest is new, so it will go faster probably, same as 65nm Opterons will go faster.
Woodcrest is not behind Opteron, it is better per watt, and the high end Woodcrest beats the high end Opteron. Enough said. Whether Intel is clock for clock better or not still doesn't matter. They are better, and if they are not better clock for clock, it doesn't seem to matter because, again, they have higher clocks.
Spoonbender - Thursday, July 13, 2006 - link
"What's the fastest Woodcrest CPU that will be released? "Umm.... None?
Does that mean AMD beats Intel by an infinite margin then?
True, if Intel has a 3ghz part out, and AMD only has 2.6, then it makes sense to compare these two.
But for now, let's just keep in mind that Intel doesn't have a 3ghz part out. They don't have a 2.6GHz part either. We are still comparing an unreleased product to one that has been out for a while.
Cooler - Thursday, July 13, 2006 - link
Their on new egg right now...http://www.newegg.com/Product/ProductList.asp?Subm...">http://www.newegg.com/Product/ProductLi...rchInDes...
xtremejack - Thursday, July 13, 2006 - link
You should note the way the two processors are compared here. Both are dual-CPU systems. Intel's FSB based system architecture means lower system bandwidth than AMD's DirectConnect architecture. The Opteron's have an on-die memory controller and a point-to-point interconnect. I am sure if you put Woodcrest on a Paxville system, you would see significantly worse performance. The 3.0Ghz Woodcrest is probably capable of a bit more performance, but the lower bandwidth FSB does not help it reach its full potential. Also coupled with the fact that FB-DIMMS have more latency than standard DDR2 means the Woodcrest isn't at a serious advantage compared to the Opteron system.Bottom-line system performance for the Woodcrest processor is still 5-20% better than Opteron. But thats way better than being 30% lower during Paxville days.
Now Conroe does not have all these complications that Woodcrest has, thats why you may see better performance advantage, also since it is a single-CPU solution, the system architecture is much simpler.
swtethan - Thursday, July 13, 2006 - link
how many people running servers are going to overclock their system? :Dfitten - Thursday, July 13, 2006 - link
Zero. I'd fire any IT person on the spot if I found out they had overclocked a production server.FesterOZ - Thursday, July 13, 2006 - link
I find this article somewhat surprising in tone. My company is a Fortune 500 and a big Dell shop so we have had access to Woodcrest workstations and servers for testing for a while. We have also tested these vs HP 9300 Athlon based Workstations and vs Sun x4100 servers and HP DL385s. Based on our tests which involve business applications, trading applications, etc., the performance of Woodcrest vs the Athlons is slightly better (about 5-10%). Nothing to really rave about, especially when its the latest Intel designs on 65nm. This actually disappointed our in-house Dell groupies, especially since they were comparing the top of the line new CPU design from Intel vs AMD's older platform. As a result we are moving away from Dell simply because they do not offer choice of CPU's at the moment and into HP's world, with our first purchase being 3 full chassis of AMD blade servers.IMHO, its now a two baron world with a missing king, each with strenghts and weaknesses.
Kiijibari - Thursday, July 13, 2006 - link
>IMHO, its now a two baron world with a missing king, each with strenghts and weaknesses.Yes I back that opinion.
Woodcrest is hampered by its FBDs. While it delivers much better bandwidth, it has worse latencies. Furthermore the 4 MB L2 cache & Core2 prefetch does not help that much in a multithreaded server environment, than in the average desktop application area.
What I want to say is, that the performance difference between Conroe/Athlon64 will be bigger than that between Woodcrest/Opteron.
First "tests"( I was told by an administrator of a huge financal institute) also showed a Woodcrest performace lack with gcc compiled 64bit applications. Are some of your applications 64bit, too ? It would be interesting, to get more 64bit statments. For some reason, there are none from Intel so far ...
cheers
Kiijibari
Spoonbender - Thursday, July 13, 2006 - link
Yes, I've been wondering about 64-bit performance too. Intel hasn't mentioned it with a word, but I hope they've made a decent implementation this time around.duploxxx - Thursday, July 13, 2006 - link
no they didn't, still the same as in the Netburst. some small 64bit testing has been done on XS forums seeing a core architecture gaining 17-18% performance on a 64bit os + program like 64bit cinebench. the opty 940 gained 31-38%