Don’t Mind SC11…
Becasue AMD and Intel just launched a pair of quad channel memory monsters...
Nov 21, 2011 in analysis, Channel, Chips, Desktop, Efficiency, Humor, Microprocessors, Mobile, Opinion, Reviews, Servers
Intel’s Sandy Bridge-E launch and the SC11 conference dominated the headlines this last week. But AMD also managed to get a few words in with the launch of its Opteron 6200 series. In spite of this Intel managed to soak up most of the spotlight as evidenced by the one sidedness of the reviews this week.
Dustin Sklavos of Anandtech took a look at one of the cheaper Ultrabooks on the market, the Toshiba Portege Z835. Priced at $879 and on sale for $799 the Z835 is a good choice if portability is your greatest concern. But when compared to its lower priced competition like Lenovo’s X120e, which we review earlier this year, the Z835 looks less appealing. The cost difference is between them is about $400, and even if you add in the $100 to $200 for an SSD upgrade, there’s still a $200 dollar price advantage in favor of the X120e. Admittedly the Z835 will no doubt win when it comes to raw performance, but the X120e can make up the difference with its superior keyboard and matte screen. When it comes down to it, Ultrabooks are just not cheap enough yet to be a truly compelling offering in the market place.
Over at the Tech Report Geoff Gasior reviewed Intel’s new Sandy Bridge-E desktop CPU. He noted that although Sandy Bridge-E is the fastest desktop CPU in the world, the launch as a whole feels a little rushed due to the X79’s lack of real improvement over the P67, and beta AHCI and RAID drivers from Intel. He also found that Sandy Bridge-E puts out a considerable amount of heat as evidenced by the chip throttling itself when he raised clocks into the 4.5Ghz range. Of the four motherboards he tested Sandy Bridge-E on, Intel’s DX79SI and ASUS’s P9X79 PRO were his top picks. Sandy Bridge-E is here, but for many it just doesn’t warrant upgrading from their 2500 or 2600K’s, maybe Ivy Bridge will change that.
Patrick Schmid and Achim Roos of Tom’s Hardware found that Intel’s Sandy Bridge-E is actually a less efficient CPU than the original. With similar clock speeds and no GPU many expected that Sandy Bridge-E would improve upon the original’s performance per watt numbers. The fact that Sandy Bridge chips are superior to their upsized, and GPU-less, Sandy Bridge-E brethren came as a bit of a surprise. It seems that the two extra cores and chunk of cache consume more power than their impact on performance merits. Although some have pointed out that the lack of efficiency might be attributable to the quad channel memory controller. Whatever the case, Sandy Bridge-E is another example that efficiency does not scale linearly with performance.
Anandtech’s Johan De Gelas took a brief look at AMD’s new Opteron 6200 series finding that Interlagos is an odd beast. In quite a few of the applications he tested the new Opterons were only marginally ahead of their predecessors. Idle power consumption is lower in applications that are core-gating aware, and slightly higher under full load when compared to the 6100 series. Still, in certain applications the 6200’s can be up to 32 percent faster than the 6100’s and the 6200’s are faster in single threaded workloads to boot. So Interlagos is better than Magny-cours, but it’s not really the game changer that AMD needs in the server market. Johan’s working on a part II, so check back for that next week.
Rounding out our lineup this week we have Guru3D Hilbert Hagedoorn’s review of the nicest X79 motherboard on the market. ASUS’s Rampage IV Extreme sports the same red and black aesthetic found on all of the Republic of Gamers branded motherboards. In addition this board offers a lot of features, many of them being hardcore overclocking specific. For example the Rampage IV has a dual bio switch, a CPU power switch, and error code reader, and even a special device called the OC key which can overlay clock speed and voltage information from the motherboard onto your monitor independent of your OS. With a bit of tweaking the Sandy Bridge-E chip was running at 5GHz and putting up Cinebench numbers that were very similar to AMD Interlagos chip from the Anandtech review. ASUS has once again produced a motherboard that is suited to hardcore overclocking like no other. Records were meant to be broken, and with this board in the hands of enthusiasts it’s only a matter of time.S|A
Amid all the AMD bashing this week, everyone forgot to mention that these new 16 core Opterons are cheaper than the 12 core chips!
I will happily add two AMD Opteron 6272 processors to my development box before paying the same price for just one 6172!
I <3 ASUS KGPE-D16 dual socket G34!
http://www.epinions.com/content_553298202244
It’s obvious that SB-E would consume more than SB if you know how Intel counts consumption of their CPU. When they talk about 85W TDP, they mean only the TDP of the core part, the uncore part is not included in the calculation. So if you add 2 cores, but remove a part that was not counted anyway, then it’s clear that the power can only go up.
None of the Interlagos reviews I have found deal with the view from HPC.
Weave Gotta Move Dees’ Microwave oven, we’ve Gotta Move These Color TVs’. Maybe you Gotta Pistol on your Little finger, maybe gun on Yout Thumb.
Money fior Nut’in & Tricks are FREE. You Play Your Guitar on MYV. Sheks’ For FREE.
Lok At DAT. My MTV. Money Anit for Nut’in. EZ EZ Chicks,Cheques for FREE. DAT Aint Nut’in.
Whew, broke out of ad mode, running for nuth’in. oops. looks like 1H’12 for Ivy, thats Break time. then 2H’12 pci-e from ivy to mains to graphics cards. then started with -=8=- as follow up, looks like 2013 be pinacle. usb3 finalized.
moore, you betcha. moore drashek….
When you take your meds make sure to snap the pills in half.
The Anandtech review was complete crap. 99% of servers run business applications with virtually no floating point math, just integer operations like string manipulation and if/then/else logic, such as databases, web servers, domain controllers, etc… and most don’t run at 100% all day long.
So what does Anandtech do? A floating-point happy Cinnebench(compiled with ICC) festival, with an equally bizarre VMWare benchmark. When someone in the comments section posed the question “Why the f**k don’t you do a normal workload”, Johan responded with “we’d love to, but it’s so hard to scale a normal workload to 16 cores”, which translates in English to “a normal workload doesn’t bring us to the conclusion Intel paid us to come to”.
Well, if anybody needs to scale a workload past 16 cores, they will certainly find a way. If not, that 64 core beast is still well capable of running 64 low-intensity VMs (aka, what virtualization is for), which means that for $25,000 you buy 2 servers can comfortably run 64 different VMs, and have the 2nd machine as a failover. That’s $390 per VM, or only $195 if you chose not to be redundant and only buy one server.
Insofar as the majority of internet servers (not corporate server room) run Linux, Anandtech could have just used Phoronix Test Suite, choosing the server subset of tasks.
IIRC AMD said orignally that the idea with BD was that most server apps are integer & logic code, thus the module design.
HPC codes are normally tuned & fine tuned for the target CPU… that dual 128 bit FPU in each BD module looks pretty neat to me but I suspect code must be tailored for it.
Yeah, but Cray had dibs on Interlagos before they even shipped. Those HPC guys spend big bucks on hardware, I’m sure they’re pounding out finely-tuned code as quickly as possible.
Hey, maybe the HPC market will actually show us what BD-optimized software performance actually looks like, since the rest of the community doesn’t seem to care.
I read those comments. While I don’t have much interest in server benchmarks, I couldn’t believe that they were benching a server chip with workloads that are so obviously atypical in server environments. On top of that, when someone actually pointed it out, Johan just made lame excuses like you said. The person who wrote the initial comment then replied, reiterating why what he said really had nothing to do with real-world server usage (which is the whole point of benchmarking chips in the first place, to get an idea of performance in real-world usage scenarios).
Then another fool comes along and says something like “I don’t understand why some people think that every review should include comparisons to every similar product available” (which, again, is kind of the point of a review), which didn’t even have anything to do with either of the OP’s comments, to which Johan said:
“Frankly I can’t imagine a situation where someone would have trouble to decide between a Westmere-EX and an AMD CPU.”
I don’t necessarily think that they’ve been bought off by Intel, I just think that after Bulldozer finally materialized, the underwhelming feeling took over and nobody bothered to do a serious review.
I’ve been reading Anandtech on and off for about 5 years. If you don’t necessarily think they’ve been bought off by Intel, then you haven’t been reading their articles for the past 5 years.
Their methodology is always this questionable, it always favors Intel, and it changes to help distort every generation to look faster. Here are some examples:
Core2 vs. Athlon X2: Core2 is marginally better in games, you should buy Core2.
Nehalem vs. Phenom II vs. Core2: They all provide acceptable frame rates, but you should buy Nehalem because it was 1% better. They also note that there is some jerkiness in Core2 because it doesn’t have an IMC(where was this insight when you were recommending it a year earlier?).
Nehalem vs. Phenom II: Nehalem is great at video encoding, so 10 out of 20 benchmarks should be video encoding, 5 out of 20 should be worthless synthetics, 3 games, and 2 productivity.
Bulldozer vs. SB: Bulldozer is as good or better at video encoding (except for cack-quality Quicksync), so let’s slowly move away from video encoding, except to occassionally throw in Quicksync’d benchmarks, acting as if the quality is acceptable.
Then there’s the Futuremark debacle, the Cinnebench debacle, and a number of other debacles where ICC was used to compile a majority of their benchmarks, despite the fact that ICC has so little marketshare in real life that it can’t be a coincidence.
The icing on the cake is that whenever there’s a victory within the margin of error, it’s always in Intel’s favor (Tom’s Hardware is also a master at this). If an Intel CPU completes a task in 60.0 seconds, the AMD CPU will complete it in 60.1 seconds. Logic tells us that the results could be reversed if you retest it, but FFS Intel can sweep every close result in an entire article by .1%.
Fortunately for us, people who actually decide and purchase these servers – sometimes take the price into account rather then those Anand and THG reviews.
I was pleasantly surprised when 1&1 turned out to use Opterons for their dedicated VPS. Something not many hosting providers do.
Both Anand & THG are quite guilty of biasing towards Intel (how else would they get their samples earlier?).
I remember well, how much they ridiculed AMD’s X2/4 after the launch of Core 2 family but were quite merciful on Intel when AMD reigned with those X2′s a few years earlier and the giant had only P4 EE gyro ovens to counter them…
Arstechnica did an Interlagos review.. Executive Summary: meh.
Just read it. I have to say the review articel and it’s methods are pretty meh too.
Look at the comments the moderators chose to hide. Something like:
“OMG AMD R SUXXX”
Is perfectly acceptable, but something like this:
“I found your methodology suspect and your writing style unprofessional”
Is “moderated due to substance-free trolling”. That kind of says it all.
Ars Technica didn’t review anything. All they did was talk about everyone else’s benchmarks. They also relied heavily on the faulty assumption that the Anandtech benchmarks actually made sense.
It would appear that NeelyCam probably owns Arse Technica. No news, no original work, just a “fair and balanced” slant on other people’s work that demands no interjection of your own common sense and reason, but just to accept their conclusion.
Arse Technica, ha! I like that.
I don’t normally read that site, I just checked it out to read the supposed review. Obviously if they’re writing crap like that, then I’m not missing anything. I’d be embarrassed to publish an article like that.
Ars used to be a good site. Jon “Hannibal” Stokes did great writing and didn’t show any significant bias. Hannibal is gone. The site was bought by Conde Nast (owner of Wired and Vanity Fair). The comments/forums still have a lot of the old Ars community, but the articles have gone to crap.