Interesting review but the comments and conclusions were awfully kind to AMD when a cold look at basically every bench result showed the Xeons well ahead in performance.(It would have been called domination if AMD were winning those tests I suspect...)
Still,they have managed to improve the performance a decent amount so that's a good thing....it would however been pretty ugly with a dual CPU Nehalem setup to bench against these new Shanghais.

This review is a joke.
Authors praise Xeon L5430 for its better power consumption even thou Opteron X4 and other Xeon racks uses 16GB of memory and Xeon L5430 uses 6GB

I'd just like to say two things:

1) that I enjoyed the review and I think Scott does great work.

2) Regarding Page8 and the APS.NET performance via the "home brewed" XML benchmarker, I had a concern. Now, I don't want to sound like how *certain* TR members have sounded in the past (and present), but isn't there some sort of issue with the way that a program is compiled that favors Genuine(tm) Intel Processors during runtime? Could the Shanghai be getting "the run around" just because some specific SSE set isn't detected? Or is that whole conspiracy pure humbug?

Very curious how this will translate into Phenom II productivity.

Well, make me a Phenom with one of these cores and make it cheaper than a Q6600 and I'm in. Gotta replace that Athlon X2 in my 780G mobo with something someday.

Pretty sad that it looks like Q6600 is still generally going to beat this core per clock. It's really close though. Only took ~2 years!!

A very nice, informative read, and Thank You for it, up to the point where the benchmarking starts.

With GPUs easily taking over the number crunching task on the PC shows how little these CPUs are actually prepared for it. As nice as the benchmarks provided by your friends are is the work they do not more real-worldly than compiling code for instance. Multi-CPU x86 servers still do a lot of database work and act as file servers, too, and this goes somewhat missing.
What I am missing, too, is true SMP folding. Amber and Tinker are single core folding clients and an SMP Linux client with the A2 core produces more than 2200 PPD. On a single Barcelona CPU one can get as much as 4200 PPD by running three of these clients in parallel, because a single client cannot utilize a quad core to more than 75%, at best. Therefore would I rather like to see what two SMP clients per CPU produce instead of these Tinker and Amber work loads.

The Conclusions are rather fair.

One thing that made me chuckle was the excerpt of the SiSoft FAQ about the Mandelbrot renderer, "...It is a real-life benchmark rather than a synthetic benchmark, ..." - and I thought: just yesterday did I see one of them walking across a street!!!

Nice review. A real shame that virtualisation wasn't tested, because that is what is important in many server installations now.

In addition it is a shame that >2 sockets wasn't tested. This is where AMD excel, and this market is growing. With >2S for Nehalem so long away, and requiring an entire platform evaluation within many companies (or they could stick with the tried and tested Socket F) this would be a good test for AMD.

Also when you have a benchmark available in C, C++ and Java, why rewrite it in C#, and then complain about problems with the C# implementation as it stands? Why not present the C and Java benches as well?

I will agree with the other poster here that said that a lot of the number crunching applications have moved or will move to GPUs, and thus these benchmarks will be less and less relevant in the future compared to other processing tests. Still, right now they're relevant and AMD didn't do many improvements here sadly - Shanghai being a non-core enhancement mostly.

As far as I can see it, Shanghai in a 2P situation has improved AMDs position in the market greatly compared to Barcelona, especially where it matters. Some of the benchmarks show excellent scaling for Shanghai, but sadly stop at 8 threads on 2S, 4S benches would be very interesting. Indeed considering Intel's lack of movement in 4S and above I guess that Shanghai is just cementing being the only choice here. Shame that no-one wants to benchmark it... (maybe AMD should send some systems out, tsk, their marketing people suck)

Good review (and testing), Scott. Thanks for all the hard work.

My oh my... Why has practicaly every site neglected to do any virtualization testing (beside the fact that Shanghai, and Barcelona for that matter, will soundly outperform any Intel Xeon including Nehalem)? I keep hearing this time constraint excuse. Well drop some of those useless server tests like Cinebench, and unbelievably, F@H.

One of the, if not the most important considerations of modern day servers IS virtualization and coincidentally, is where Shanghai excels. There's been alot of effort put into improving virtualization performance to give the industry what it wants and needs, yet has gone almost completely unnoticed or tested. Why this and other sites ignore the feature that the industry is begging for is beyond me (well, I quite postive I know why that is, but that can wait for another time). Might as well thrown in a few SuperPi tests as well.

I'd really like to pester for some info on the settings, material used, etc, etc for the x264 benchmark. I'm routinely taking 1080i and 1080p source (OTA HDTV and my Blurays respectively) down to 720p, and your first pass results seem low and your second pass results seem high compared to my usual workloads. I'm guessing it's all boiled down to you have roughly 2X the number of cores (for the second pass results) along with either slower storage or just different arguments making the first pass slow (I've noticed that I'm rarely using all my cores during pass #1, much less using them all fully).

Of course, I'm on a much newer build of the core, so this may all be meaningless.

WOW.. AMD hate day... well the 8000 series are much better.. 4socket and 8 socket.. AMD scales like a Be-atch...they are still close clock for clock , if you will, to Intel at the 2 socket so it is much better then they were.... But , it would be nice to see more server crucial test..VM for one..

The last sentence in the conclusion really spells out what I was thinking as I was reading all the way through this. It's a bit like watching a couple of minor league / Jr Varsity teams (Serie B for you eruos) fight it out, knowing the real world-class players are sitting out the round. It's like a golf tournament without Tiger Woods, or a swim meet without Michael Phelps. When Westmere and then Beckton arrive I suspect all of these chips will amount to footnotes.

http://www.techradar.com/news/computing-components/processors/first...

Still, it's a good thorough review (as always) and AMD has a decent (however short term) sales proposition to OEMs who are already committed, since they can offer measurably better performance in the same power envelope (or less) without any major re-validation challenges.

Sweet. Thanks for the review. It appears Shanghai's a substantial improvement over Barcelona, at least meeting the Xeons on clock-for-clock performance and even exceeding it in some cases (like SPECjbb2005 and MyriMatch). That and the lower system power consumption might make this the best current platform for Java application serving.

It's too bad you guys didn't have the time to do virtualization testing. I look forward to a followup article on virtualization performance when you do get it worked out.

Given a single-socket Core i7 beat the leaderboard in many of these benchmarks, I'm salivating just thinking about what your results with the Xeon 5500 series will be.

I'd have liked to see some of the less extreme processors in the 5400 board. The 5400 MCH has been around for over a year now, and there are many servers that come with it. Its snoop filter and higher memory bandwidth would make it a more interesting platform for comparison than the 5000 MCH used with the E5450. (The 5100 MCH also remains interesting for low-power smaller-spec use of course.)

I believe Steve plans to write up a blog post about this benchmark and release the source code. If any of our readers have suggestions for improvement, he'll be taking them at that time.

*anxiously waiting!*

Also, Damage, setting thread affinity should be no issue.

I'm a bit curious about the real world consumer desktop user performance
of the Phenom II .
I wish that TR reviewed 3d max 2009 insted of Cinebanch, Photoshop etc.
Real world applications not "my friends" programs, or abstract bench that nobody uses (mass use).

I'm not interested in server to much , not at all. But unfortunately for me , the server with its Cloud will eat us up,so the future is server base CPU (e.g i7).

One more wish real gaming performance at 1600 rez. Is too much?

Well, I did go back to the TR review of i7 and checked the benchmark results myself. It does not look pretty fro AMD. The 1 socket 965 system outperformes the 2 socket 2384 system in every benchmark. Even when considering the outrageous price of $1000 for the i7-965 plus let's say $300 for a motherboard plus some juicy DDR3, it's still much lower then the price of 2x2384s and the corresponding dual socket board. And the performance is higher. Pitty about the 12GB RAM limit in those systems due to the current lack of non-ECC non-Registered 4GB DDR3 modules.

AMDs opportunity is in the commercial application server space. The SPEC benchmark illustrates this but if you look at recent benchmarks then Shanghai really shines in the majority of commercial applications, and most importantly in virtualisation. A 16 core Shanghai outperforms a 24 core Dunnington in VMmark with the same number of tiles and just as well for Intel that power consumption isn't shown.

Factor in that Nehalem doesn't hit 4P for nearly a year, and that customers like standards then all is certainly not lost for AMD in the segment they are concentrating on.

I'd also add that when comparing like systems then Shanghai is more power efficient than the L5430. A good way is to use HP's online BladeSystem sizer. All power and fan components are identical and the blades really are just CPU and memory. Anything over 2 DIMMs and Opteron is much more efficient, especially at loads most likely to be seen in real life (40 to 60%)

I agree with the suggestion that it would have been nice to see the results from let's say an i7-920 there (to roughly match the 2384 in clock speed). Even if that is not a server system. I'm curious how it compares to these dual-socket systems. Based on the SPEC results it should be pretty close.

If you compare it to AMD's previous thing, then this processor is quite brilliant. Adding that to the fact that they can be overclocked to at least 3.2 GHz (usually well beyond that) on air might make it a more tempting enthusiast's platform than what Barcelona was - after all, the difference between, say, a 3.4 GHz intel and a 3.4 GHz AMD can not be earthshattering (with AMD's processors going up to 2.7 GHz only in the review, it's hard to guesstimate), and you can potentially save money by picking the latter. And as soon as they finish fleshing out the chip's accessories (HT), it might become a no-brainer for some server owners.

At least the margins are so high in this market that AMD has lots of room to sell on price. Looks like they'll keep Intel honest in most server markets. I'm very interested to see how things go in the desktop market.

Small nitpick: Shanghai isn't named after the city but after its F1 track (it's the theme used by AMD).

Good review. At least AMD will be able to keep pace in the small server space, one of the few markets where it has consistently maintained a strong presence.

So much for "waiting for AMD's 45nm CPUs", they'd get crushed by Nehalem.

It seems in certain environments the Shanghai processors could do amazingly well, in others just average. Not really sure what to gather from all of this, it just seems AMD really needs these parts to respond well with clock speed ramp up.

I know this is to show dual sockets or better but it would have been nice to see the Core 7i chips in there considering they do so well in workstation situations even if they are single socket.

booooooooo