SUN Micro frags Intel with X4600-M2
Opteron 40% faster than Woodcrest. One Opteron server replaces 50 Intel Xeon server. Direct Connect Architecture more efficient than Intel's FSB.... read it yourself.
Oh, boy, I missed this:
"Additionally, under full load, the 2-socket Sun Fire servers demonstrate better scalability and higher performance than similarly sized Intel "Woodcrest"-based servers, with the ability to consolidate up to 22 legacy servers using VMware."
posted by Sharikou, Ph. D. @ 10/17/2006 07:05:00 AM
17 Comments:
"Additionally, under full load, the 2-socket Sun Fire servers demonstrate better scalability and higher performance than similarly sized Intel "Woodcrest"-based servers, with the ability to consolidate up to 22 legacy servers using VMware."
This is what I have been shouting from beginning !!! on this blog on Scientia's on tomsharware forums and everywhere!!!! Conroe behaves poorly under high load and scales badly! it's an inherent defect of the latency hiding efect of FSB using lots of cache ! it only works under light load and light memory pressure with limited thread numbers working at the same time! Conroe IS FBS bottlenecked!
That is purely marketing hype as you can see it is a press release from Sun not from an independent source.
They mention nothing about the specs of the claimed 50 X86 servers. Also please remember that it is a 16 way server
As many people pointed out already, don't compare specs without considering the OS. Try to find a system with both Woody and Solaris, then draw your conclusion.
Sharikou said...
"Opteron 40% faster than Woodcrest."
Lets not look at Integer performance, only FP.
"Direct Connect Architecture more efficient than Intel's FSB"
Hasn't that already been established?
It's a press release from Sun Microsystems. In other words, its an ad for a new server. I believe that there are several IDEPENDENT benchmarks on the net which have the Woodcrest easily beating the Opteron. Hey, that reminds me. Weren't you going to get some Woodcrest vs. Opteron benchies??
I could not actually find the Sun Opteron 2220SE result on Spec, so that is a bad start, looks like it has been submitted and not verified yet .. shows the marketing hype.
As mentioned by previous poster they did not mention Spec INT of course ;)
Also, Opteron 2220 on pricewatch is 40% more expensive than Xeon 5160
Given all that this is nothing to get too excited about.
Hmm talking about ads. I think Intel is really desperate. A recent copy of Forbes had at least 5 or 6 full page ads sporting Intel. (2 may be issued by HP) and 2 or 3 by Intel doing it alone. The funny one to be read was the Intel vPro ad....where they talk and talk about security. Interesting part is they do not mention anything remotely about what OS to use or support etc. (of course it has to Windows OS...) But then such an Ad should fall under the legal spot light of Microsoft., since Intel vPro is claming to do what ever is the purview of the OS. I am quite unsure who Intel marketing is trying to fool this time..Definitely not the IT managers..who don’t buy such crap..
Ha, Ha, Intel desperate because it puts up an ad? They can put 6 pages of ads because they can afford to, not because they're desperate. Do you think that if Amd could afford to do this that they wouldn't? Unfortunaly for Amd they can't, so they are left with a single billboard on a highway and ads on web sites.
They can put 6 pages of ads because they can afford to, not because they're desperate.
One press release by SUN is better than 100 pages of Intel ad. SUN's PR simply denounces Woodcrest as crap.
" Anonymous said...
Ha, Ha, Intel desperate because it puts up an ad? They can put 6 pages of ads because they can afford to, not because they're desperate. "
Are you out of your mind ? How come I did not see 6 Full page size ads from Intel on the Forbes magazine in the last 4 years ??
"it only works under light load and light memory pressure with limited thread numbers working at the same time! Conroe IS FBS bottlenecked!"
I think you should define the term 'load' more precisely. Does it mean high data processing load or high data access load? Core 2 actually performs well in the former. For example, if you transcode an avi to mp4 with the highest quality setting, Core 2 will run noticeably faster than X2. However, if you do so with medium-good quality setting the relative speediness of Core 2 reduces significantly.
OTOH, for things like webserving or database access (realistic ones, not those whose results all fit into a 4MB cache), like you said, Core 2's bottleneck comes form the FSB under high data access load. Woodcrest as a server chip couldn't get enough bandwidth with DDR2, so Intel had to use FB-DIMM; but then the high latency/price/power becomes issues, and we all know that FB-DIMM isn't THE future of memory anymore.
The point is, Core 2 itself is a good core; the shared L2 cache is a good design, too. What is bad is Intel's refusal to embrace open standards. How hard would it be for Intel to adopt HT, or to license ccHT from AMD, and make Core 2 Duo/Quad + HT the hell of a good chip? I guess this is too difficult for Intel and its fanboys to figure out.
Who cares about ads. What I want to know is when sharikou will have those woodcrest vs. opteron benchmarks that he promised us. I know for a fact that he will be unbiased when he gets them. One thing we can all say about sharikou is that he is the Fox news channel of the blogging community. "Fair and balance." That should be your trademark sharikou :)
I wouldn’t be laughing too loud at Sun’s claims. The wizards at Sun are revered by the entire server industry. Intel boys might consider taking a pass on this one.
Intel is advising a great Q4 which may mean Q3 financial report tonight may suck.
Have you noticed all the Intel server TV ads? I wonder why? The average TV viewer doesn’t purchase servers.
If this sharikou character is a doctor then I'm the King of the United States. Now who in here thinks that this guy is a doctor? Dude if you are not getting paid in some way by Amd then you must have a really sad life. Get some fresh air. How about women. Do you like women? Football? You have to remember that life is short and what do you want to be remembered for? "Here lies Dr. Sharikou...the man who single handedly took down the biggest semiconductor manufacturer in the world." Do you think that you'll get a thanks from Amd? Doubt it. They will be too busy building a monopoly. I can see it now...another clown like you wasting his life supporting Via so that they will save the world from the evil Amd empire.
Let me just take a moment to point out that the fake Doctor has no clothes. Here's why. Sharikou has no problem pointing to a press release (not an unbiased source) and concluding that his favorite brand "frags" Conroe. However, when the overwhelming majority of the press concludes after independent benchmarking that Core 2 Duo destroyed even the mightiest AMD FX2 and Opteron 1P platforms, Sharikou says the benchmarks are just synthetic and can't be trusted or that Intel paid reviewers to generate good press.
So yet again, Sharikou lives by a double standard, shows himself incapable of making scientific conclusions and is basically a shill for AMD. I think that about sums him up.
"edward said...
Woodcrest as a server chip couldn't get enough bandwidth with DDR2, so Intel had to use FB-DIMM; but then the high latency/price/power becomes issues, and we all know that FB-DIMM isn't THE future of memory anymore."
Actually, FBDIMMs today are just DDR2s with AMB (a mini memory controller) onboard.
Q:Why does Intel need FBDIMM?
A: you can't stack that many DDR2 DIMMs off a single memory controller for signal integrity.
FBDIMMs use the AMB to "cascade" the memory controller signal from the main memory controller to all the other DIMMs further down the chain.
This allows you to stack more (about double)
But it results in higher latencies for the DIMMs furthest away, cos the memory request is passed down the chain until it gets there.
Q:Why doesn't AMD need FBDIMMs?
A: AMD uses integrated memory controllers.
Each socket drives its own banks of DIMMs. With 2 sockets, you match Intel's FBDIMM capacity without the need the latency and power penalties.
"Actually, FBDIMMs today are just DDR2s with AMB (a mini memory controller) onboard."
And DDR2 today is JUST SDRAM with doubly-pumped SDRAM with double BUS frequency. ;-)
"Q:Why does Intel need FBDIMM?
A: you can't stack that many DDR2 DIMMs off a single memory controller for signal integrity."
You can't stack that many DDR2 DIMMs off a single memory controller at high speed and maintain signal integrity.
Integrated memory controller solves the bandwidth scalability problem.
Post a Comment
<< Home