Intel "Harpertown" Xeon vs. AMD "Barcelona" Opteron
by Jason Clark & Ross Whitehead on September 18, 2007 5:00 PM EST- Posted in
- IT Computing
AMD Quad-Core Opteron (Barcelona)
The new quad-core Opteron from AMD is the first true (x86) quad-core processor that features one die with four cores. Barcelona is based on a 65nm fabrication process, and rumor has it late next year AMD will move to a 45nm process. More than just four cores on one die, the new quad-core Opteron features several micro-architectural changes from the previous dual-core Socket-F platform. Below are the main highlights of quad-core Opteron.
Independent Dynamic Core Technology
AMD has always been strong with their performance-per-watt numbers, and with this new technology AMD can alter the frequency of each core individually. This should allow AMD to more tightly control their power levels thus decreasing overall power consumption and lowering TCO.
AMD CoolCore Technology
This unique technology evaluates which parts of the die (cores/memory or both) are required by the application, and can cut power to unused transistor areas to reduce power consumption and lower heat generation.
SSE128
The previous generation Opteron had to use two clock cycles to execute 128-bit SSE operations. Barcelona can now execute 128-bit SSE operations in a single clock cycle.
2MB L3 Shared Cache
In order to keep up with multi-threaded applications, AMD added 2MB of L3 cache that all four cores share. The cache breakdown for the new Barcelona is as follows: 64KB L1 cache (64K each for data and instructions) and 512KB L2 cache per core, and then the 2MB L3 shared cache.
Drop-in to existing Socket-F
Barcelona is based on the previous generation Socket-F platform, allowing most modern Socket-F servers (with a BIOS update) to drop in a quad-core Barcelona part. We can attest to this fact, as our S3992 Tyan Board received a BIOS upgrade and was off to the races with Barcelona.
The new quad-core Opteron from AMD is the first true (x86) quad-core processor that features one die with four cores. Barcelona is based on a 65nm fabrication process, and rumor has it late next year AMD will move to a 45nm process. More than just four cores on one die, the new quad-core Opteron features several micro-architectural changes from the previous dual-core Socket-F platform. Below are the main highlights of quad-core Opteron.
Independent Dynamic Core Technology
AMD has always been strong with their performance-per-watt numbers, and with this new technology AMD can alter the frequency of each core individually. This should allow AMD to more tightly control their power levels thus decreasing overall power consumption and lowering TCO.
AMD CoolCore Technology
This unique technology evaluates which parts of the die (cores/memory or both) are required by the application, and can cut power to unused transistor areas to reduce power consumption and lower heat generation.
SSE128
The previous generation Opteron had to use two clock cycles to execute 128-bit SSE operations. Barcelona can now execute 128-bit SSE operations in a single clock cycle.
2MB L3 Shared Cache
In order to keep up with multi-threaded applications, AMD added 2MB of L3 cache that all four cores share. The cache breakdown for the new Barcelona is as follows: 64KB L1 cache (64K each for data and instructions) and 512KB L2 cache per core, and then the 2MB L3 shared cache.
Drop-in to existing Socket-F
Barcelona is based on the previous generation Socket-F platform, allowing most modern Socket-F servers (with a BIOS update) to drop in a quad-core Barcelona part. We can attest to this fact, as our S3992 Tyan Board received a BIOS upgrade and was off to the races with Barcelona.
Facebook
Twitter
RSS
77 Comments
View All Comments
Proteusza - Thursday, September 20, 2007 - link
In fact on the front page of Anandtech there is an Intel Resource Centre link.And the URL looks like it makes sure that Intel knows who the referrer is - typical advertising.
JarredWalton - Thursday, September 20, 2007 - link
Yes, Intel pays for an Intel Resource Center page - it includes all of our Intel-related articles and some other information. It's pretty clear that the page is sponsored by Intel. I have no idea how much they pay, however. Don't like that area? Then don't click on it - it still isn't Intel influenced articles as far as the AnandTech articles are concerned.As for the RAM config, you seem to want us to intentionally handicap Intel just for your own benefit. Eight Registered ECC DIMMs came in the AMD config, and they are single sided DIMMs - meaning, a 2GB double sided DIMM would only consume marginally less power. The Intel setup came with 2GB DIMMs... obviously Intel knows that you pay a power penalty for every FB-DIMM, and you also pay a latency penalty. Ideally, we would have 4x2GB DIMMs on the AMD setup, because any business serious about the platforms is likely running 2GB DIMMs these days.
Taking it to the extreme, obviously running 16 2GB FB-DIMMs uses a lot more power than 16 2GB DDR2-667 DIMMs. I'm not sure how many businesses actually use that approach, though - not many in my experience. As always, we are testing specific facets of performance (and power and whatever else you care to name). Is there more to it? Of course. Is Intel always better or AMD always better? Of course not.
If I were in charge of a server purchase for a large company right now, I'd be looking at my specific needs to determine the best overall platform. For most companies, that's relatively low loads so the Opteron is perfectly acceptable. That means it's going to come down to features like manageability and support rather than performance. Very likely, I'd be looking at slightly more mature hardware anyway - bleeding edge and servers aren't generally a good mix.
Justin Case - Thursday, September 20, 2007 - link
Yes, Intel pays for an Intel Resource Center page - it includes all of our Intel-related articles and some other information.All of them? Interesting, I can only seem to find articles about Core and Core 2, nothing about, oh, say, Prescott or Paxville. In fact, I can't find any article in your "Resource Center" that is even remotely disfarvourable to Intel. Must be a temporary glitch. Or maybe those articles aren't deemed "resourceful" enough.
Also, it's interesting to see an anandtech.com address in my browser's location bar, the Anandtech banner at the top of an Anandtech page, and "This site is presented by Intel" below it. Well, at least "we have been warned".
I guess I'm just used to seeing manufacturer propaganda on the manufacturers' website or inside clearly identified ad boxes, not integrated into supposedly impartial hardware review sites. I know, that's so 20th century of me.
it still isn't Intel influenced articles as far as the AnandTech articles are concerned.
Of course not. And I'm sure that when large corporations make donations to political candidates, they're not expecting to influence their future decisions the least bit. They only do it so they can get a bit of exposure by appearing in the list of contributors. I'm sure Intel would be just as likely to sponsor your "resource center" if your articles pointed out the weaknesses in their products (such as, oh, I don't know, actually using all the FB-DIMM slots on their servers, instead of leaving 75% of them empty).
Proteusza - Thursday, September 20, 2007 - link
Wow I hadnt actually looked at the Intel Resource Centre before, I thought it was a site on Intel's server that Anandtech simply linked to.Now I know better and shant be visiting this site anymore.
Its a shame that you threw away your journalistic integrity for money.
Justin Case - Thursday, September 20, 2007 - link
There wouldn't be anything wrong with an ad banner linking to Intel's (or any other) site. But when a hardware review site allows its own server (and logo, and page template) to be used for advertising, well... says a lot.P.S. - Interesting how any posts that point out the differences in the test systems (and their consequences) get instantly voted down. Must be just some "regular users" who consider that objective facts are "off-topic" here, eh?
FrankThoughts - Thursday, September 20, 2007 - link
I'm not sure which of your nonsensical posts to respond to, so I'll just pick this one. A few points to think about if you're actually capable of such an act.1) How much do you pay AnandTech for their articles? Yes that's right, they're ad supported! Guess that means everything they publish is lies and paid for, eh? Or else, it's just a source of revenue, like it's always been. Looking at the "Intel Resource Center" I see pretty much every article that mentions Intel (CPU, chipset, tradeshow, motherboard, whatever) going back about a year. Why are there no Netburst articles? Probably because NetBurst hasn't been worth discussing for over a year, and Anandtech hasn't reviewed any in that time. Did Intel do this on purpose? Maybe - but wouldn't you if you were in their position? Now, I don't see any omission of articles in the past year where Intel got less-than-glowing commentary, and I see some links on the right to some Intel site stuff. It's not a big deal... and ifyou don't like it DON'T CLICK THE LINK! Moron....
2) Memory configs. Ever tested any memory? Apparently not, because you clearly don't know Jack or Squat about the topic. Let's see, try doing a power draw test with 4 x 512MB DIMMs and 2 x 1GB DIMMs. Or 4 x 1GB vs 2 x 2GB. Tell me how much of a power difference there is, because I've looked at it and I see less than a 2W difference. So in terms of power draw the AMD is penalized a few watts at most. Or if you prefer, Get off your stupid soap box and get the hell out if you can't contribute anything useful! Don't worry, we won't miss you.
3) Flaws with the article's methodology. For one, the only thing I'm really sure this testing shows is how these two servers perform in an AS3AP test. It doesn't tell me how it will work in the servers my datacenter uses. You know what? Short of getting the hardware and testing it I doubt anything will show that information. Different apps, drives, networks, and who knows what else yield different results. So this is just a rough estimate, and anyone that takes it as more than that is already a fool.
The RAM configurations are also somewhat questionable, depending on the use. Some places will only use 4x1GB RAM; others will load all the slots with 2GB modules. (That should SERIOUSLY hurt the FB-DIMM setup!) I wonder why they didn't test that way? Oh that's right: they probably don't have 16 2GB FB-DIMMs available and Intel didn't want to help them out. That's only $2000-$2500 depending on the brand (or $6000+ if you get it straight from Dell! But at least then you know the RAM works properly and Dell supports the setup.) Why don't you send them the memory they need? While you're at it, can you get them some realistic benchmarks that will stress that much RAM? Yeah, didn't think so.
There are flaws in the article, true. There are flaws in every article out there. You don't honestly think the latest reviews showing performance in one specific area of a few games is the same as testing every game, right? Or that SLI and Crossfire work properly in new titles most of the time? Or that quad core on the desktop matters at all when it comes to gaming... or anything outside of video encoding and 3D rendering and a few other specific tests?
Now, a bunch of your posts just got one point higher because I commented instead of downrating. Why rate them down? How about because you're being an arrogant prick and a fanboy, complaining about stuff that is largely out of the hands of the reviewers? I'm done. Feel free to miss the point entirely and complain some more.
Justin Case - Thursday, September 20, 2007 - link
1.1) I (and all visitors to this website) pay Anandtech through the ads they have on their site. The various companies that advertise here only do so because of us. And the reason why people visit this site is to read (what they believe to be) honest and bias-free articles and hardware reviews, not editorial advertising. If Anandtech thinks it can survive on Intel's sponsorship alone, that's fine. But eventually even Intel will stop sponsoring a site that has no credibility (great as Johan's articles are, I don't think he can churn them out fast enough to make people come here on a daily basis).1.2) I wasn't the one who said "The Intel Resource Center includes all our Intel-related articles". That statement was made by a member of Anandtech's staff.
1.3) You might as well say that if some site decided to write "AMD is great!" or "Sony is the best!" at the end of every paragraph, people should just "ignore it" and trust that the rest of the site was not biased in any way. There is a difference between clearly labelled banner ads and accepting money to turn your entire site into a big advert for company X or Z. "Moron".
2) Don't try to spin it. The point is that the Intel system was tested with half as many memory sticks as the AMD one, and (more importantly) with 75% of its memory banks empty. I find it quite telling that, when asked about this, the Anandtech employee posting above wrote that was because "Intel knows that you pay a power penalty for every FB-DIMM". So, because "Intel knows" that, Anandtech's system comparison is tweaked to make it less obvious to the readers...? o_O
3.1) Yes, I'm sure poor Anandtech didn't have the resources to buy or borrow another four FB-DIMMs for this review (so they could use at least half the slots in the board). It's not as if they have contact with any manufacturers or retailers that would be happy to send them the RAM in exchange for a mention and a link in the article, eh...? They did manage to get a 12-drive 15k SAS array, though. I bet they give those away on street corners.
3.2) Learn some maths... "moron". 1 GB FB-DIMMs cost $70 each. In other words, it would have cost them $840 to fill the remaining (12) of the slots in the board, $560 to match the configuration in the Opteron system (8x1) or $280 to test power consumption with half the board's slots loaded. This is assuming they didn't have any more FB-DIMMs available and couldn't get / borrow them for free. All of these values are a far cry from your suggested "$6000" or even "$2500".
JarredWalton - Friday, September 21, 2007 - link
Don't read too much into my comments, Justin - I don't live anywhere near Jason/Ross. Or Anand, Derek, Wes, Gary, or Johan for that matter. In fact, other than editing I have pretty much nothing to do with the other articles. They are merely my opinion, and I'm speaking realistically: why would Intel send a system specifically equipped to handicap it? They won't. Should Jason and Ross go out of our way to do so? To what purpose? The two are relatively comparable, inasmuch as we can get similar hardware. And don't underestimate the difficulty of getting new hardware - especially very expensive server hardware. Beyond that, readers need to read between the lines a little - take their own needs into account. If an IT guy is looking at getting a new server and installing 32GB (or 64GB) of RAM, I hope they have more sense than to look at 8GB server configurations and assume everything will be the same, only with "more RAM".I still don't get your whining about the sponsored Intel section - it's just a "site view" ad for Intel as far as I can see, with content from AnandTech and Intel that may be of interest. If you're looking for a quick collection of Intel information, I'd assume that's useful. It's on the home page on a smallish image, and at the end of any Intel specific (i.e. a new Intel CPU) articles. Heck, I wish AMD had a sponsored view as well. :) At any rate, we have huge ads from Gigabyte, Kingston, OCZ, Crucial, and many others splashed around. What makes the Intel "sponsored" version of the site different (which clearly states "This site sponsored by Intel")? Personally, I don't click the ads and I don't click on the little Intel Resource link either (except to see what it was). I'd assume 99.99% of you are the same. Hey - we're also "sponsored" by Verizon and T-Mobile I guess. What's that mean for our iPhone articles?
We have a separation of editorial and advertising staff for a reason - other than looking at the ads, I have no idea who is supporting us. I don't even know how much an ad costs on the site. I do know that at the end of the month, I get a pay check, and for that I'm grateful. There's also a fine line between tact and flaming that needs to be walked, particularly when writing an article. We've harped on FB-DIMM in the past, we ripped on Intel in the NetBurst era, and we've had ups and downs with pretty much every manufacturer out there.
The fact is, we appreciate good technology and products, and right now Intel has the upper hand in most areas. Barcelona isn't bad, but it's not K8 vs. NetBurst by any stretch. Still, if Intel got rid of FB-DIMMs - or at least made them optional for now - and got an integrated memory controller into their systems yesterday, you wouldn't see me complaining. I guess we need to wait for Nehalem on those area.
Take care,
Jarred
Proteusza - Friday, September 21, 2007 - link
I understand that AMD and Intel probably submitted test demos, and you were either contractually unable to modify their configuration or felt it wouldnt be good science.But, then you should have said "Take these power consumption and performance per watt metrics with a bit of salt because in real life no one would handicap a server by putting in 8x1 sticks rather than 4x2."
You've said yourself that a) putting in more memory sticks, irrespective of size increases power consumption, b) anyone needing such an AMD server wouldnt bother with 8x1.
So I dont know why you are baffled when we cry foul - here we have a badly configured AMD server vs a well configured Intel server. Which do you think is going to draw more power? Its like saying, "Our AMD server came with two HD2900 XT cards in CrossFire, so in our SQL Server tests its performance per watt metric is extremely bad." Duh.
Tell your readers you dont think its a fair comparison then. I dont care if AMD screwed up the config, or you couldnt be bothered about correctness. You've said yourself it would be a handicap for the Intel server to have the same memory configuration as the AMD server. Why not make sure the readers know that?
Again this has nothing to do with fanboyism. It has everything to do with poor benchmarking, and as I see it, benchmarketing.
All we're saying is that the configuration between the AMD and Intel systems should as close as possible for the test results to have meaning. You obviously dont agree.
JarredWalton - Friday, September 21, 2007 - link
You still missed the point: the configs are close. They're close in every area except the motherboard and CPU, which you're not going to be able to change.2x4GB DIMMs (on AMD) uses very nearly the same amount of power as 8x1GB DIMMs. It's not an issue there. On the other hand, 8x1GB FB-DIMMs gets a nice 5-7W penalty per FB-DIMM, because of the AMB. Regular DDR2, the power draw is determined by the number of memory blanks on the PCB. FB-DIMMs is that, plus another ~5W for the AMB. If they ran the AMD system with 4x2GB, I expect it would use within 4W of the same power draw.
FB-DIMMs are bad for power consumption. How do power requirements change with 32GB loads? I can guess that AMD will have an advantage, but without actual testing I'm not going to publish a guess. At the same time, I don't think everyone out there runs 16 DIMMs in a server. I think most companies buying a new server would buy 2GB DIMMs (or FB-DIMMs), but if a company knows that they really only need 8GB of RAM they're not going to install twice that much let alone four times as much memory.
What would be ideal? I'd like to see scaling numbers showing performance and power with 2x2GB up through 16x2GB on both systems. I'd also like to see more benchmarks, and benchmarks that can leverage the availability of more RAM. I'd like to see a truly repeatable benchmark showing how the servers behave in virtualized environments (which is were these high performance quad-core CPUs are truly important). There are always more tests people would like to see, but realistically no one can provide tests of everything that might be done.
And you know what? I haven't a clue as to how to do even the tests Jason and Ross are running, let alone something like virtualized environment testing, because I'm not involved with anything like that. :)