quote:The first problem with the review is credibility. The author, Johan De Galas, is the same person who did the worst server comparison I've ever seen: http://www.anandtech.com/IT/showdoc.aspx?i=2772&am...">http://www.anandtech.com/IT/showdoc.aspx?i=2772&am....
This latest review has a lot of comments and odd arrangements that suggest bias.
TDP numbers are mentioned without stating that Intel's are not the same as AMD's nor mentioning that for Intel you have to include both the Northbridge memory controller and the higher power draw of FBDIMM.
At the bottom of this page we do see the biggest reason for doing a pro-Intel server review:
The lucrative 4 socket market was and is still dominated by the 8xx Opteron, which managed to capture up to 50% of the market share
Displaying comparison graphs directly from Intel. Where are the graphs from AMD? This makes me wonder if Intel co-authored this review.
Trying to downplay AMD's huge gains in 4-way servers by suggesting that AMD only gained share because Paxville was so bad.
Giving Intel praise for improvements to a poor design:
Thanks to the shared and inclusive nature of the L3-cache coherency traffic between the four CPUs is significantly reduced.
Yes, it is reduced but FSB cache coherency along with MESI is inferior in every way to AMD's cache coherency.
We start off by using a lower Opteron machine:
The HP ProLiant DL585 available in the labs was not the recently introduced DL585G2 which features DDR2, the new AMD Opteron socket F
Criticizing a superior design with a trivial (and incorrect) comparison to a worse design:
The quad dual core configuration generates more cache coherency traffic, as the 8 cores of the Opteron have to keep 8 L2 caches coherent while the Xeon MP has to keep track of 4 L3 caches.
This is false. Cache coherency between L2 pairs on the same chip do not use HyperTransport. The chip to chip coherency would be the same as Intel if AMD used MESI and had the same size cache. However, since AMD uses MOESI there is an improvement. And, since AMD's cache is much smaller there is again much less cc traffic. Finally, this still completely misses the point that 100% of Intel's cache coherency traffic still travels over the FSB while AMD's does not. This is about as biased or technologically ignorant of a comparison that anyone can make.
Finally at the bottom of the server overview page (where it is less likely to be seen) the author admits that Intel's TDP is not the same as AMD's and admits that FBDIMM draws more power. However, he fails to give any actual numbers to see what the real comparison is. In other words, on page 1 the stated numbers favor Intel, however the correction on page 4 contains no actual numbers to see who is really ahead in power draw. The author is either trying to favor Intel or is incredibly unprofessional.
At the top of this page the author tries to suggest that Anandtech's poor reviews are due to a lack of support from the manufacturers.
In case you're wondering why we chose to use the fastest Xeon DP, the second fastest Xeon MP, and the second fastest Opteron, the reason is simple: those were the CPUs that were made available to us.
Sorry but the biases in this review are not related to processor speed. Also, the 2.4Ghz Opteron is not the second fastest; the Opteron 854 runs at 2.8Ghz. 2.4Ghz is actually the third fastest Opteron.
The AMD server has slower 333Mhz DIMMs while Intel gets 400 and 533 Mhz.
The Woodcrest SpecInt rate numbers are quite good for dual core. However, it isn't clear how much the numbers might be boosted by large cache. Nor does it make any sense to include these with quad numbers since Woodcrest might not scale 100%. However, we do see another common graph cheat where IBM's Power numbers were included to prevent AMD from having the top spot in the chart. Also, the AMD number is low at 160. It should be about 176. It should also be noted that Spec is in a bad position at the moment. No new numbers are being added to update the old 2000 database and there are not yet enough 2006 numbers for a good comparison. I have to wonder if Anandtech will still be quoting Spec numbers in the second half of 2007.
Curiously, the Xeon name is shown prominently in the graph whereas the Opteron machine is simply referred to as an HP.
The SpecJbb test shows what a big difference proper compiling makes. In the second graph the quad opteron is almost identical to the quad xeon. However, the author tries to downplay this:
This is good and bad news for AMD: it means that the Opteron 880 can compete with the more expensive Xeon MP, but it also means that the Opteron requires more "manual" optimization than the Xeon MP. The Xeon MP performs at the same level with 4 instances as it does with one.
There is no mention of the fact that Xeon's I/O requires performance stealing software patches to match what Opteron does natively. Nor is there any mention of the fact that two Intel Xeons have to share the same bus which cuts performance in half when the memory bandwidth is saturated.
And, at the bottom of the page the author makes sure to mention higher numbers for Intel that are not in the graph. Apparently, a tie is not acceptable. The new numbers suggest a whopping 27% lead for Intel. However, if we divide the Intel numbers by the difference in DIMM speed Intel's lead drops to a much smaller 6%.
Although Opteron does well overall and demolishes the P4 Xeon this test is suspect. The graph increases faster from 4 to 8 threads than it did from 2 to 4. This pretty much goes against every theory of processor operation. The test code has a problem of some kind. A normal graph would show either the same rate of increase or more commonly a slight dropoff.
The MySQL numbers are a waste of time. First of all the test is improperly configured:
We optimized for a server with 4GB of RAM.
Why optimize for 4GB's when the servers have 8 and 16 GB's of memory? It would be rare to find someone using a database server with only 4 GBs. Remember that the hardware can handle 64 GB's of memory with the slow DDR 333 that they are using in these tests.
We also see an anomalous increase in the Woodcrest graph between 50 and 100. This is not normal and again shows some problem with the test code.
The server comparison is ridiculous. Why compare an older chipset and processor to Intel's newest? A socket F comparison would be much more professional and unbiased. Remember that the recent RAS features for Opteron (which are important in a high level server) came with Revision F.
We see grudging admissions of unrealistic cache based results:
In applications where the large L3 cache doesn't play a big role, the relatively poor server performance of the "NetBurst" architecture becomes visible again
And, a grudging admission that this server chip is still inferior to Opteron:
In a nutshell, the new Xeon MP will have a hard time convincing people who are leaning towards an Opteron server or want the best performance/watt.
Very true. Now, of course, the author has to try to salvage the review by making a completely false claim:
But on the other hand, the decent performance and superior RAS features will keep the customers who desire high availability in the Intel camp
Or they could just buy a real Opteron system that does have these RAS features. This one server does not represent the entire Opteron server market.
quote: Still, our previous conclusion stands: clock for clock, the Opteron is quite a bit better at this than the Xeon "Core" architecture (Xeon 5160) and a lot better than the Xeon "NetBurst" architecture (Xeon MP 7130).
quote:the fact that the Xeon 5160 cannot scale past its 4-thread peformance at all
quote:One thread of OpenSSL Signing per core is optimal
quote:but I never expected Johan to "tow the party line" like this.
quote: Those benchmarks will be presented in our Clovertown - Intel's new quad core server CPU - review.
quote:Well, we did mentione it at our price comparison. From a performance point of view, the G2 is within 2% of the DL585 given a similar configuration
quote:The HP DL585 also has a few shortcomings: it does not offer any PCIe expansion slots, the SCSI controller is an old SCSI 160 model, and there are no USB ports on the front of the machine
quote:There has been a relentless assault without any mercy on the Server CPU market...