Sign in with
Sign up | Sign in
Your question
Closed

Anandtech benches dual Magny Cours and Xeon

Last response: in CPUs
Share
a b à CPUs
March 29, 2010 5:38:05 AM

http://it.anandtech.com/IT/showdoc.aspx?i=3784

I won't cherry pick graphs to post so you're going to have to read it yourself :)  However, I will post these two as I thought they were rather interesting.




Looks like the Magny Cours is getting its butt handed to it, but then we look at the Linux graph and see this:



The Intel system doesn't really do any better (actually slightly worse) but the AMD one does alot better! Of course it's Alpha software at the moment so performance in later versions will likely vary alot. In any case the Linux version wipes the floor with the Windows one as long as you're not on an Intel rig.
a b à CPUs
March 29, 2010 3:09:25 PM

Would like to see more linux tests actually. That said, intel is now irrelevent at 4P and AMD has never been stronger in HPC either.

Considering bulldozer will be a simple drop in, anybody buying a server right now must be looking at AMD and realising intel's pricing schemes are ridiculous in comparison.
March 29, 2010 3:22:49 PM

jennyh said:
Would like to see more linux tests actually. That said, intel is now irrelevent at 4P and AMD has never been stronger in HPC either.

How have you reached the conclusion that Intel is irrelevant at 4P?

a b à CPUs
March 29, 2010 3:43:00 PM

Chad Boga said:
How have you reached the conclusion that Intel is irrelevant at 4P?


I'd imagine chugging a bottle of cheap Scotch helps considerably :p . Not to mention forming your AMD-sanctioned opinions a year or so ago, long before any product release, over on AMDZone :kaola: 

Actually it is a fairly good if incomplete or premature review. Would like to see the perf. vs. power consumption numbers. AMD wins a few but also loses substantially in the increasingly important virtualization VMmark.

Of course, Intel could always pull an AMD and lower prices if they feel threatened - eventually AMD with nearly 700 mm^2 total die size per 12core MC would find their profit margin evaporating long before Intel's 6-core with under 250 mm^2 die size, when you take into account that SOI is something like 20-30% more expensive than strained silicon. In fact I would not be surprised to see Intel cutting deals for HPC-like quantities.
March 29, 2010 4:55:17 PM

fazers_on_stun said:
I'd imagine chugging a bottle of cheap Scotch helps considerably :p . Not to mention forming your AMD-sanctioned opinions a year or so ago, long before any product release, over on AMDZone :kaola: 

Actually it is a fairly good if incomplete or premature review. Would like to see the perf. vs. power consumption numbers. AMD wins a few but also loses substantially in the increasingly important virtualization VMmark.

Of course, Intel could always pull an AMD and lower prices if they feel threatened - eventually AMD with nearly 700 mm^2 total die size per 12core MC would find their profit margin evaporating long before Intel's 6-core with under 250 mm^2 die size, when you take into account that SOI is something like 20-30% more expensive than strained silicon. In fact I would not be surprised to see Intel cutting deals for HPC-like quantities.

Intel have Nehalem-Ex to come out, but it would be really stupid if Intel didn't anticipate the possibility of AMD going the MCM route.
March 29, 2010 5:54:10 PM

So the 12 core beat the 6 core. I am shocked.
a b à CPUs
March 29, 2010 8:46:44 PM

roofus said:
So the 12 core beat the 6 core. I am shocked.


The 45nm 12 core beat the 32nm 6-core with turbo, Hyperthreading yadda yadda.

Anyway, fastest cpu in the world.
a b à CPUs
March 29, 2010 11:08:40 PM

Wow has it really been so long?

Seems like I posted that last one an hour ago, yet no intel fanboy is arguing with it?

The silence is deafening. :) 
a b à CPUs
March 30, 2010 1:07:57 AM

Wow. So now you post some blog that says AMD has the worlds fastest chips because "Although Intel Corp (INTC - Analyst Report) and AMD have continued to battle for supremacy in speed and performance over the last few years with successive launches of their dual core, triple core (from AMD) and quad core microprocessors, AMD’s new chip tops the list so far.

The new chips are based on a design called the Magny-Cours and come with eight and 12 cores, meaning that they are the fastest chips in the world. Intel’s new Xeon 5600 chips announced a couple of weeks ago have six cores."

What a waste of time.

I am not saying that the new 12core chip is not the overall fastest chip either. I have truthfully not looked into in detail yet and dont actually know. Just a brief look through the anand article is all I have seen. But linking to that page as proof that the AMD's are the fastest in the world is flat out stupid. Even for you.

And jenny I think you said something in a other post about the difference between AMD and Intel fans. Here is the actual difference. AMD will defend anything/any product AMD ever makes blindly to death. An Intel fan will do the same. Both are idiots. The people you actually think of as Intel fans are simply not people like you and the other loons at AMDzone that have a love affair with AMD for some reason.
a c 131 à CPUs
March 30, 2010 1:26:06 AM

someguy7 said:
Here is the actual difference. AMD will defend anything/any product AMD ever makes blindly to death. An Intel fan will do the same. Both are idiots.

I think I missed where you told us what the difference was.
a b à CPUs
March 30, 2010 1:38:19 AM

Yes. I did not structure that last statement very well. But my actual point was after that line you quoted. Which is "The people you actually think of as Intel fans are simply not people like you and the other loons at AMDzone that have a love affair with AMD for some reason."

Let me try and clear it up. There is NO difference between fanboys. The difference is the likes of Jenny think that anybody that is not a member of the AMD fan club just has to be a Intel fanboy. That is what I was getting as the difference as it related to her other post.

She stated something like AMD fans stick by AMD through thick and thin while the Intel fanboys dont. That being said. Those people in her eyes that are not sticking by Intel are fanboys that jumped ship. No, they would just be the normal person who is not a loyal fanboy idiot.

Ya dig?
March 30, 2010 3:03:58 AM

someguy7 said:
Yes. I did not structure that last statement very well. But my actual point was after that line you quoted. Which is "The people you actually think of as Intel fans are simply not people like you and the other loons at AMDzone that have a love affair with AMD for some reason."

Let me try and clear it up. There is NO difference between fanboys. The difference is the likes of Jenny think that anybody that is not a member of the AMD fan club just has to be a Intel fanboy. That is what I was getting as the difference as it related to her other post.

She stated something like AMD fans stick by AMD through thick and thin while the Intel fanboys dont. That being said. Those people in her eyes that are not sticking by Intel are fanboys that jumped ship. No, they would just be the normal person who is not a loyal fanboy idiot.

Ya dig?

So true.

Just because JennyH and her ilk at AMDZone have deep seated psychological issues around CPU's, it doesn't mean other people do, and this is something that they never seem to comprehend.
March 30, 2010 3:21:10 AM

Well you all have to admit the massive performance increase in memory bandwidth. The quad channel controller really helped it seems.
March 30, 2010 3:42:05 AM

yannifb said:
Well you all have to admit the massive performance increase in memory bandwidth. The quad channel controller really helped it seems.

Was anyone ever disputing such an obvious thing was on the cards?
March 30, 2010 3:51:35 AM

Chad Boga said:
Was anyone ever disputing such an obvious thing was on the cards?

No, but it seems you only focus on the negatives. And no need to get hostile, i said i liked the bandwidth, not that i killed your cat :D 
March 30, 2010 4:02:56 AM

yannifb said:
No, but it seems you only focus on the negatives. And no need to get hostile, i said i liked the bandwidth, not that i killed your cat :D 

To whatever extent I focus on the negatives it is due to trying to counteract the often insane propaganda of AMD Zealots like Jenny and Keith and other bozo's from AMDZone, who make it difficult to have a rational conversation about computing technology.

However a number of things need to be kept in mind about this launch, Magny Cours loses as often as it wins against Westmere Xeons and as another poster pointed out to me on a different forum, the following things about what AMD has done here(besides avoiding outright annihilation in the server market) :

"AMD is expending 692 mm2 of silicon to get near Intel's
240 mm2 Westmere in performance and has to price it low
to sell. Servers used to be a place AMD could get decent
margins...

Intel wins in performance per core, total performance for most
everything other than STREAM, and software licensing costs.

MC is a sign of how far AMD has fallen in competitiveness
since its Opteron heydays. "

a b à CPUs
March 30, 2010 4:06:29 AM

One thing to keep in mind is that at least one of the benchmarks, although I don't remember which one, only uses up to 16 threads. So while the Xeon system can use all 12 available physical cores, the Magny Cours system is only using 16 of its 24.
a c 127 à CPUs
March 30, 2010 4:18:12 AM



Um this is useless. It only has Dunnington. And its from AMD themselves.
a b à CPUs
March 30, 2010 4:18:24 AM

doesnt the new Xeon come out in the morning?
a b à CPUs
March 30, 2010 1:25:53 PM

Chad Boga said:
"AMD is expending 692 mm2 of silicon to get near Intel's
240 mm2 Westmere in performance and has to price it low
to sell. Servers used to be a place AMD could get decent
margins...

Intel wins in performance per core, total performance for most
everything other than STREAM, and software licensing costs.

MC is a sign of how far AMD has fallen in competitiveness
since its Opteron heydays. "


I laugh every time I see 'performance per core' and in no way does it win in total performance. Buyers don't care if it's 692mm2 or 240mm2 of silicon either.

The fact is MC could have matched or surpassed the Xeons in price, but AMD chose to price it even cheaper. We'll see what intel responds with soon enough.
a b à CPUs
March 30, 2010 1:33:40 PM

yannifb said:
No, but it seems you only focus on the negatives. And no need to get hostile, i said i liked the bandwidth, not that i killed your cat :D 



it needed killed... little b*stard...
March 30, 2010 1:51:59 PM

jennyh said:
I laugh every time I see 'performance per core' and in no way does it win in total performance.

It depends on the application/benchmark. And don't forget some licensing models are done on a per core basis.

Quote:
Buyers don't care if it's 692mm2 or 240mm2 of silicon either.

True, but it continues the trend of AMD's cash cow(i.e. server) getting leaner and leaner.

Quote:
The fact is MC could have matched or surpassed the Xeons in price, but AMD chose to price it even cheaper.

It is no fact whatsoever, unless you feel the people running AMD are stupid beyond belief.
March 30, 2010 3:09:24 PM

that's really cool to hear some intersting here
a b à CPUs
March 30, 2010 3:31:19 PM

jennyh said:
The 45nm 12 core beat the 32nm 6-core with turbo, Hyperthreading yadda yadda.

Anyway, fastest FATTEST cpu in the world.
[/url]

There, fixed it for ya :D .

That financial blogger has as much credibility as you do, it seems, since he equates physical core count, and only core count, with speed :p ..
a b à CPUs
March 30, 2010 4:00:30 PM

Whatever methods he got there by doesn't matter, MC is the fastest chip in the world. Fattest too? Hell yeah but all that matters is intel has no counter and I expect to see this claw back much of the lost market share last year.

Broadpoint maintains buy rating on AMD

Quote:
"...Additionally, further strength could be driven by better-than-expected sales of "Magny-Cours" on compelling performance and power advantages that nearly mirror Intel (with substantially lower price pointsprice points). As such, our server revenue assumption of $690M for FY10 (or ~10% of AMD sales) could prove conservative on ASP mix-shift and higher uptake rates."
March 30, 2010 4:05:35 PM

To show how unconcerned Intel are by AMD, here is their pricing on Becton

http://files.shareholder.com/downloads/INTC/319913976x0...

X7560 24MB 8c 16t 130W 2.26 GHz 6.40 GT/s $3692
X7550 18MB 8c 16t 130W 2.00 GHz 6.40 GT/s $2729
E7540 18MB 6c 12t 105W 2.00 GHz 6.40 GT/s $1980
E7530 18MB 6c 12t 105W 1.86 GHz 5.86 GT/s $1391
E7520 18MB 4c 8t 95W 1.86 GHz 4.80 GT/s $856
L7555 24MB 8c 16t 95W 1.86 GHz 5.86 GT/s $3157
L7545 18MB 6c 12t 95W 1.86 GHz 5.86 GT/s $2087
X7542 18MB 6c 6t 130W 2.66 GHz 5.86 GT/s $1980
a b à CPUs
March 30, 2010 4:13:05 PM

intel drops a million on politicians

Chad they have to keep prices high what with all the politicians that need bribing, and because of the expensive wages they are paying those US engineers when they'd much rather hire on some slave labour from China. :D 
March 30, 2010 4:16:40 PM

jennyh said:
intel drops a million on politicians

Chad they have to keep prices high what with all the politicians that need bribing, and because of the expensive wages they are paying those US engineers when they'd much rather hire on some slave labour from China. :D 

If AMD had answers, Intel wouldn't be able to get away with those prices.
March 30, 2010 4:19:08 PM

jennyh said:
Buyers don't care if it's 692mm2 or 240mm2 of silicon either.


They care about power consumption (which generally scales with the number of transistors under load) and price. And AMD care about margin, which will be much lower on a chip that's nearly three times the price... though, as has been said, the margins in server CPUs are high in any case.

Either way, it's nice to see AMD having some kind of competitive CPU again even if they have to resort to an MCM to achieve it.
a b à CPUs
March 30, 2010 4:44:15 PM

MarkG said:
They care about power consumption (which generally scales with the number of transistors under load) and price. And AMD care about margin, which will be much lower on a chip that's nearly three times the price... though, as has been said, the margins in server CPUs are high in any case.

Either way, it's nice to see AMD having some kind of competitive CPU again even if they have to resort to an MCM to achieve it.


That's what is so great about MC - it is faster than intel at the same power level, even though it has double the cores, 50% more cache and using 45nm instead of 32nm. The only real drawback is the huge die.

MCM just makes sense at the server level - it's intels margins that will be suffering here, not AMD's. Intel either have to drop prices on their 90% market share or they will lose a lot of that share.
March 30, 2010 4:48:04 PM

jennyh said:
That's what is so great about MC - it is faster than intel

Except for all those times that it's not.

What a bizarro world you live in, when getting close to Intel, becomes being faster than Intel.
a b à CPUs
March 30, 2010 4:50:45 PM

jennyh said:
That's what is so great about MC - it is faster than intel at the same power level, even though it has double the cores, 50% more cache and using 45nm instead of 32nm. The only real drawback is the huge die.


No, that's what's so sad about MC - despite years of desperate, nail-biting, teeth gnashing struggle, AMD's premiere CPU loses to Intel's Westmere Xeon by a score of 8 to 5. In other words, Intel wins in 60% more of the benches.

And if we were to average out the winning percentages, Intel with that one win by a whopping 942%, would just flush the MC(M) down the toilet.

Interesting how AMD had to copy Intel's MCM approach - I guess they secretly admire all those expensive US engineers who can come up with truly innovative designs :p ...

Sad, but true...
a b à CPUs
March 30, 2010 5:38:33 PM

fazers_on_stun said:
No, that's what's so sad about MC - despite years of desperate, nail-biting, teeth gnashing struggle, AMD's premiere CPU loses to Intel's Westmere Xeon by a score of 8 to 5. In other words, Intel wins in 60% more of the benches.

And if we were to average out the winning percentages, Intel with that one win by a whopping 942%, would just flush the MC(M) down the toilet.

Interesting how AMD had to copy Intel's MCM approach - I guess they secretly admire all those expensive US engineers who can come up with truly innovative designs :p ...

Sad, but true...


Fazers we've already been over this. Random pointed out that one of the tests only used 16 threads, so you want to add 50% to MC's score and see who really wins that one? :D 

8v5 becomes 7v6

Seems to me you're counting the Blender result twice yes? The way I see it, the fastest overall was the MC at 18s, 50% faster than the 26.4 best effort of the Xeon.

6v6 wow look it's a draw already. :whistle: 


Now had the reviewer actually used the top Opteron priced closest to the Xeon, at least two of them would have changed places.

Let's not forget
Quote:
None of our benchmarks required more than 20 GB of RAM.


Xeon has 24gb ram, MC has 32gb ram hmm that seems fair in your warped pro intel world right? :D 

The reviewer could hardly have made it easier for the Xeon and in the end well, when intel is bankrolling your site you have to do what you can to keep your masters happy, don't you? If the software was available to make full use of both chips, MC would be well ahead. It's the fastest cpu in the world, don't bother arguing we all know it's the case. ;) 
March 30, 2010 7:24:43 PM
a c 127 à CPUs
March 30, 2010 8:25:13 PM

jennyh said:
http://translate.google.com/translate?hl=en&sl=de&tl=en...

Seeing as we're counting benchmarks, here's a real review.

9 to Opteron, 6 to Xeon - that's 2nd best Opteron vs th best Xeon btw. ;) 

http://images.tecchannel.de/images/tecchannel/bdb/392958/881BEE66929A490040B56EE5999FB019_800x600.jpg


Yea its a second best Opty using 24 real cores in a server enviroment that likes more real cores not doing much better than 12 real and 12 fake cores of a Xeon. I think thats what you are failing to see and why a lot of people are unimpressed with MC.

If a 12 core MC barley outperforms a single Xeon by say 5%, then in most cases I bet he companies will go with the single Xeon to save power and money in the end. But still this would be like a Core 2 Duo with HT outperforming a Phenom II X4 in highly multithreaded apps. You see it now?

The die size also bodes bad news. Its large and that means less yield in the end. The arger the die, the less CPU per wafer and the less CPU per wafer means the less they make per CPU and the less they make per CPU means they can't keep the price as low.

You seriously have like mega neon green glasses on don't you? You praise them for producing a 12 core that barley outperforms a 6 core. I am afraid to watch it try to go against Beckton.
a b à CPUs
March 30, 2010 8:43:19 PM

Lol I can't believe how the intel crowd try to turn this into some victory.

I'll just say it in the words of JF from AMD.

Quote:
So, what are the 32nm benefits? I remember some company saying that 32nm would allow for lower power, greater performance and lower price.

When I look at the comparison I see the following:

Intel 32nm platforms have higher power at the wall than AMD
Intel 32nm platforms perform lower than AMD
Intel 32nm platforms are more expensive than AMD

Can someone please explain how Intel has a "32nm advantage"? For all of those that said we were at a disadvantge because we were still on 45nm, can you please print this post out and put it on your wall?
a c 127 à CPUs
March 30, 2010 9:57:18 PM

^Yea.... lets use some AMD marketing guy who would obviously praise Intel. Just like Paul from intel would praise AMD....

You really like using biased crap don't you?

If he is talking server, I would expect a CPU running at 3.33GHz to consume more than a CPU running at 2.3GHz. Also an extra channel of RAM adds too.

Sorry but if it comes from the mouth of AMD, its useless here or anywhere.
a b à CPUs
March 30, 2010 10:10:37 PM

Any of the intel fanboy crap in this thread is useless except for a form of entertainment, namely mine. :D 

Which part of it don't you get? AMD can make 12 cores at the same TDP as intel's 6 cores. On an older, more power hungry process.

I dunno about you jimmy, but if I were a hopeless intel fangirl, I'd be wondering how that is even possible.

AMD can fit 12 cores at 45nm into the same TDP envelope as intel can fit 6 cores at 32nm. And the AMD chip is a lot faster when it gets to stretch its legs. And it draws less power too - at idle and load.

I've been telling the likes of you and Fazers for months now, surely you are beginning to understand that AMD has far, far superior engineers? How else can this happen? They are beating intel with worse tools because they are so much better. This isn't about intel holding back their best - the 5680 is 130w TDP that IS their best and it isn't good enough to beat even the 6174 Opty at 115w.

If more cores are so much better than logical cores, why is intel wasting time with logical cores? Where are intel's 12 core cpu's at 130w TDP?

The answer is, their engineers aren't good enough to make them, even with an entire process nodes advantage. What's your explanation jimmy?
March 30, 2010 10:19:43 PM

jennyh said:
Which part of it don't you get? AMD can make 12 cores at the same TDP as intel's 6 cores. On an older, more power hungry process.


Where are the actual power measurements? I don't see any in that article, only some comments about how AMD can cut power consumption when the CPU is idle... which won't happen very often in a CPU-hungry server.
a c 127 à CPUs
March 30, 2010 10:23:18 PM

Jenny, yet you forget that not too long ago Intel had quad cores at 65nm in the same power envelope that AMD had their Athlon X2s in..... hmmm...

It goes both ways. Says who Intel isn't thinking of a MCM 12 core? And I am pretty sure Intel would have no problem doing it.

Quote:
Nehalem-EX (Beckton) (Octo-core):
up to 8 cores, up to 16 threads
24MB shared L3
45nm
90/105/130W TDP
4 FBD2 interfaces
4 QPI links
Socket-LS (LGA 1567)
44 bits physical address
48 bits virtual address


An old article but still valid

http://www.dvhardware.net/article29763.html

So Intel goes from 6 cores at 130w TDP to 8 cores and a massive 24MB L3 cache at 130w TDP.

Intel can easily do 12 cores at 130w TDP or less.
a b à CPUs
March 30, 2010 10:30:04 PM

MarkG said:
Where are the actual power measurements? I don't see any in that article, only some comments about how AMD can cut power consumption when the CPU is idle... which won't happen very often in a CPU-hungry server.


Maximum energy consumption



Btw - this might even be the opty at a disadvantage as I read JF saying that the test systems they had given out weren't fully optimised for power draw. At least that is why there were no Anand power consumption benchmarks.
a b à CPUs
March 30, 2010 10:32:01 PM

jimmysmitty said:
So Intel goes from 6 cores at 130w TDP to 8 cores and a massive 24MB L3 cache at 130w TDP.

Intel can easily do 12 cores at 130w TDP or less.


Yeah lets see how much it draws at the wall though. That's where to measure it, and as usual it doesn't look great from an intel pov. :) 
March 30, 2010 10:36:47 PM

jennyh said:
Btw - this might even be the opty at a disadvantage as I read JF saying that the test systems they had given out weren't fully optimised for power draw. At least that is why there were no Anand power consumption benchmarks.


Again, I don't really see anything useful in some random graph pulled from some random web site; particularly if it's showing 'maximum power' (whatever that's supposed to mean) rather than performance per watt or some similar metric. My Atom system would only use 57W maximum power running the same benchmark because that's the most it can possibly draw from the wall, but I doubt Google will be buying Atoms to replace their servers any time soon.
a b à CPUs
March 30, 2010 10:44:09 PM

Maximum power should be pretty self explanatory. It's what each system pulls at 100% load.

You wan't minimum?



You want perf/watt?




As this is currently the only website with MC benchmarks that matter (anand's don't), this is what you'll have to go by for now. It's not going to change by much either way, the simple fact of the matter is, MC is faster, cheaper and less power hungry than Westmere.

March 30, 2010 11:38:56 PM

jennyh said:
Maximum power should be pretty self explanatory. It's what each system pulls at 100% load.


But, again, it only matters if your system is running at 100% load. As I said, my Atom would take a small fraction of that power, but those server CPUs would then be idling for six hours while the Atom was still chugging away at 100% load to get the work done.

Quote:
As this is currently the only website with MC benchmarks that matter (anand's don't), this is what you'll have to go by for now. It's not going to change by much either way, the simple fact of the matter is, MC is faster, cheaper and less power hungry than Westmere.


Thanks, that's more useful.
a b à CPUs
March 30, 2010 11:41:15 PM

Well that's the thing yanni.

Bulldozer will be a plug-in to the same G34 socket. AMD recognised 2 years ago that things were going pretty badly, and only fantastic execution kept them alive. MC is a miracle chip designed and executed to absolute perfection, and that's why the intel fanboys can't get their small minds around it.

You can safely buy one of these now. AMD is selling them so cheap they are almost irresistable. I told Chad this was going to happen 2-3 months ago, we made a bet too, and he will use the avatar of my choice for 6 months. ;) 

There is no reason to buy anything except MC for 2P and 4P. Just buy it and upgrade to Bulldozer in a year or two, AMD is committed to the socket, it's cheap as hell and it cannot fail - it will only get better with software.

AMD will put an entire fab behind producing MC for the future too - this isn't some short term stop-gap until Bulldozer, it's the start of Bulldozer and will reap the benefits of Bulldozer for years to come. :) 
March 31, 2010 12:33:55 AM

Round 2: Fight!
!