Anyone have a quick performance/watt between 6core xeon and 6core amd?

[The-Noid]

I am sure this has probably been posted before. Sorry for the similar request.

Thanks for the help.

 

[classy]

Your question is vague. Is this for a server? Desktop? For what Apps? Home? Work? Do you have a budget? Air Cooling? Water Cooling? Do you want it to glow in the dark? Crossfire? Sli? etc.

 

[JAG87]

What difference does it make to the efficiency of the CPU if it is used for desktop, server, home, work, if it's air or watercooled if it glows, or if you stick 37 GPUs in it.

http://www.anandtech.com/show/3674/amds-sixcore-phenom-ii-x6-1090t-1055t-reviewed/10

Thuban lags behind gulftown when it comes to performance/watt, and you can expect Lisbon to lag behind as well.

 

[The-Noid]

That is what I wanted. Perfect. Need to upgrade the work computers today and we do lots of sims monte carlos.

 

[Chiropteran]

http://www.anandtech.com/show/3674/amds-sixcore-phenom-ii-x6-1090t-1055t-reviewed/10

Thuban lags behind gulftown when it comes to performance/watt, and you can expect Lisbon to lag behind as well.

How do you come to that conclusion based on that graph? Not saying you are wrong, just curious.

I did a quick look at the 1090t and i7 920, the intel uses slightly more load power and does slightly better or worse in the 3d rendering benchmarks, doesn't seem to me like Thuban is lagging behind at all, it is right where it should be, at least compared to the 920. Unless you are focused on idle power. Which is a different sort of number that the OP wasn't asking for.

 

[JAG87]

We're not comparing it to the 920, we're comparing it to the 6 core Gulftown, that's intel's competition for Thuban/Lisbon. We're not comparing an architecture that came out 18 months ago to something that came out 2 weeks ago. 6 vs 6, apples to apples.

And you'll also notice that the Gulftown is clocked at 3.33 while Thuban at 3.2, and clock/clock Gulftown performs better, so it's a total loss on AMD's part. The only thing that saves them is price.

 

[thesmokingman]

These might help?

http://www.anandtech.com/show/2978/amd-s-12-core-magny-cours-opteron-6174-vs-intel-s-6-core-xeon/1

http://it.anandtech.com/show/2919/9

 

[heyheybooboo]

Using Handbrake and a stock 1055T at 100% load I was kicking around 180w (with a Gigabyte 785g AM2+, an SSD and 2 WD 'Green' drives). IIRC (with vcore bumpage around +.05v) at 14x240MHz it was around 220w.

I didn't try stock at 240MHz or attempt to undervolt (which PhIIs love) because I'm moving on up to AM3 and re-running my tests.





--

 

[JFAMD]

Thuban lags behind gulftown when it comes to performance/watt, and you can expect Lisbon to lag behind as well.

How can you make assumptions about Lisbon as the product has not been released yet?

Are you aware of the power bands available? In addition, you shouldn't compare desktop to server parts.

 

[Hacp]

Performance/watt depends on the application.

 

[classy]

What difference does it make to the efficiency of the CPU if it is used for desktop, server, home, work, if it's air or watercooled if it glows, or if you stick 37 GPUs in it.

http://www.anandtech.com/show/3674/amds-sixcore-phenom-ii-x6-1090t-1055t-reviewed/10

Thuban lags behind gulftown when it comes to performance/watt, and you can expect Lisbon to lag behind as well.

The point I was trying to get across was that many variables can affect performance/watt. Especially the app that is being targeted.

 

[JAG87]

How can you make assumptions about Lisbon as the product has not been released yet?

Are you aware of the power bands available? In addition, you shouldn't compare desktop to server parts.

Because desktop and server chips are the same with a different name. People need to get that through their heads. The only difference is server parts are binned to work with lower voltages since they need to go into tiny rackmounts with 40mm fans, and they need to be more reliable so heat is a big issue.

Just like Intel bins their Westmere xeons, AMD is going to bin their Lisbon opterons, they play the same game, just different shirt colours. The power figures that Gulftown and Thuban show now will reflect in the server parts, perhaps just a bit lower for both camps.

The point I was trying to get across was that many variables can affect performance/watt. Especially the app that is being targeted.

Actually there aren't many variables, the only variable is the one you mentioned, and it doesn't have that great of an effect. Just like comparing AMD's Evergreen efficiency to Fermi, although it varies between different games, on average Fermi is far less power efficient. Heck you might even find one game that consumes more power on AMD if you look hard enough. Good chances are the OP is looking for an average unless he runs only one specific application.

 

[JFAMD]

Because desktop and server chips are the same with a different name. People need to get that through their heads. The only difference is server parts are binned to work with lower voltages since they need to go into tiny rackmounts with 40mm fans, and they need to be more reliable so heat is a big issue.

No, lisbon and thuban are different products. You'll see features in lisbon not in thuban and vice versa. They are built on a common architecture, but you can't put a Lisbon in an AM3 and call it thuban, and a thuban in a C32 is not a lisbon.

Most of performance per watt will be driven by the apps as someone correctly pointed out above. You can't make a blanket statment, you really need to know the apps.

 

[aigomorla]

We're not comparing it to the 920, we're comparing it to the 6 core Gulftown, that's intel's competition for Thuban/Lisbon. We're not comparing an architecture that came out 18 months ago to something that came out 2 weeks ago. 6 vs 6, apples to apples.

And you'll also notice that the Gulftown is clocked at 3.33 while Thuban at 3.2, and clock/clock Gulftown performs better, so it's a total loss on AMD's part. The only thing that saves them is price.

i agree with ya however...

that price factor is just too awesome and would completely tip the scales on efficiency... were talking about 1/3 to 1/5 of the price if u look at the higher end hexcore xeons like the X5670.

And that translates to more per dollar Jag, because they are not 1/3 to 1/5th as fast as gulftown, id say about 2/3rds the speed is a good translation.

Yoxxy if u need a lot of spam machines with virtualization, you will get more work done with 2 X6 which will run you about the same as 1 gulftown.
If you need to do a lot of encoding and other things, which are intel HT optimized... then you will get more work done with 1 gulftown because of how HT works.

So Yoxxy it might be better in your sense to build 2 x6 machines for monte carlo if your going to be that heavy on virtualization.

When looking at performance without looking at price.. intel wins.
When looking at effiencency as well as dollar per performance.. im sorry.. but AMD scored on this one..
X6 pricing is just too awesome, and i hope AMD doesnt learn from intel.

 

[classy]

Because desktop and server chips are the same with a different name. People need to get that through their heads. The only difference is server parts are binned to work with lower voltages since they need to go into tiny rackmounts with 40mm fans, and they need to be more reliable so heat is a big issue.

Just like Intel bins their Westmere xeons, AMD is going to bin their Lisbon opterons, they play the same game, just different shirt colours. The power figures that Gulftown and Thuban show now will reflect in the server parts, perhaps just a bit lower for both camps.





Actually there aren't many variables, the only variable is the one you mentioned, and it doesn't have that great of an effect. Just like comparing AMD's Evergreen efficiency to Fermi, although it varies between different games, on average Fermi is far less power efficient. Heck you might even find one game that consumes more power on AMD if you look hard enough. Good chances are the OP is looking for an average unless he runs only one specific application.

No offense but you don't understand performance/watt. I can change simple cooling and increase the performance/watt. I can change hard drives and increase performance/watt. Obviously I was being facetious, but the point I was trying get across was you need to take in account other factors to give a more accurate answer. Or you can get some guy who will answer the question with some slide with 0 other info taken into account......wait....

 

[classy]

That is what I wanted. Perfect. Need to upgrade the work computers today and we do lots of sims monte carlos.

Do you have a certain budget?

 

[JAG87]

No, lisbon and thuban are different products. You'll see features in lisbon not in thuban and vice versa. They are built on a common architecture, but you can't put a Lisbon in an AM3 and call it thuban, and a thuban in a C32 is not a lisbon.

Most of performance per watt will be driven by the apps as someone correctly pointed out above. You can't make a blanket statment, you really need to know the apps.

Xeons also don't fit in LGA1366, and they also have more features than their desktop counterparts.

Thuban is a child of Lisbon just like Bloomfield is a child of Gainestown, and their performance and power consumption will be extremely similar. You can sing your song all day, this is how it is.

i agree with ya however...

that price factor is just too awesome and would completely tip the scales on efficiency... were talking about 1/3 to 1/5 of the price if u look at the higher end hexcore xeons like the X5670.

And that translates to more per dollar Jag, because they are not 1/3 to 1/5th as fast as gulftown, id say about 2/3rds the speed is a good translation.

Yoxxy if u need a lot of spam machines with virtualization, you will get more work done with 2 X6 which will run you about the same as 1 gulftown.
If you need to do a lot of encoding and other things, which are intel HT optimized... then you will get more work done with 1 gulftown because of how HT works.

So Yoxxy it might be better in your sense to build 2 x6 machines for monte carlo if your going to be that heavy on virtualization.

When looking at performance without looking at price.. intel wins.
When looking at effiencency as well as dollar per performance.. im sorry.. but AMD scored on this one..
X6 pricing is just too awesome, and i hope AMD doesnt learn from intel.

Well your looking at the desktop market prices, while he will probably look at xeon/opteron prices. The difference in the server market might not be as much (especially if you don't buy the highest clocked xeon), and we won't know until Lisbon is selling.

I also don't agree with the cost factor because buying hardware is a one time investment, while paying the power bill is a recurring expense, and depending on how long you keep your processors running 24/7, the Intel's might pay for themselves in the end while getting more work done.

No offense but you don't understand performance/watt. I can change simple cooling and increase the performance/watt. I can change hard drives and increase performance/watt. Obviously I was being facetious, but the point I was trying get across was you need to take in account other factors to give a more accurate answer. Or you can get some guy who will answer the question with some slide with 0 other info taken into account......wait....

Umm, shaving off heat is not going to bring you the kind of power saving you speak of. Yes if the silicon runs at lower temperature it will leak less and use less power, but where talking a very small margin. Replacing a 12V HSF fan with a pump + 12V RAD fan will blow that margin away.

And wtf does the other shit have to do with it. Were comparing only one thing, CPU (and it's platform) and assuming the rest of the system is identical. I don't understand why you are bringing the other components into play.

 

[aigomorla]

the Intel's might pay for themselves in the end while getting work done.
.

great point jag.. however i don't think your gonna see much of an end point after u counted the cost of initial from the difference between a Gulftown system vs a X6 system.

Its not half the wattage were talking about, nor is it 1/3

 

[JAG87]

great point jag.. however i don't think your gonna see much of an end point after u counted the cost of initial from the difference between a Gulftown system vs a X6 system.

Its not half the wattage were talking about, nor is it 1/3

That's true. Depending on the price you might have to run them for many years before you compensate for the initial cost.

 

[classy]

There are 4 things you have to look at to give an informed answer.
Performance in the App or Apps-How big of a performance gap
Power Consumption-Actual power being consumed
Cost-Has to be included in relation to the performance. A 15% improvement doesn't warrant in some cases an additional 50% cost.
Desktops or Servers-Desktops may not require being run 24/7 so that will certainly affect power consumption over a server. A desktop also is cheaper and may be an alternative to a server. Of course this all depends on the apps and purpose of the computer.

and 5th being possible Support Costs-To extend warranties on servers cost more than a desktop. If say a hex-core Intel is providing top performance 2-3 years from now, it may in the end be cheaper to pay a little more for Intel over an AMD system right now and not have to do a complete upgrade 2 years from now, thus saving money.

And Jag87 you do know what facetious means right . You did graduate from hi skool, right. And also, they do make 1366 Xeons.

 

[Phynaz]

Most of performance per watt will be driven by the apps as someone correctly pointed out above. You can't make a blanket statment, you really need to know the apps.

AMD never said this when they had a power consumption advantage.

Funny that.

 

[Vesku]

That's true. Depending on the price you might have to run them for many years before you compensate for the initial cost.

Which is not very realistic for those who run enough machines to see decent savings from 10 to 50W power differences. They'll be upgrading every year or so.

 

[JFAMD]

Thuban is a child of Lisbon just like Bloomfield is a child of Gainestown, and their performance and power consumption will be extremely similar. You can sing your song all day, this is how it is.

No, Thuban is a 125W TDP part. I don't know the Lisbon standard power TDP off the top of my head, but it is probably double digit.

And there is an HE (lower TDP) and an EE (significantly lower TDP) that will be available.

The 6 core EE will have a TDP down in the 35W range (which is something that we have said publicly), which puts the performance per watt in a different class. I'd be willing to bet that a 2P Lisbon platform could have a lower power draw than a single socket Thuban.

The two products are targeted at different uses.

 

[hans007]

No, lisbon and thuban are different products. You'll see features in lisbon not in thuban and vice versa. They are built on a common architecture, but you can't put a Lisbon in an AM3 and call it thuban, and a thuban in a C32 is not a lisbon.

Most of performance per watt will be driven by the apps as someone correctly pointed out above. You can't make a blanket statment, you really need to know the apps.

that might night be necessarily true. Most chips with different features in the same generation are still the same.

you reallyd ont think some intel chips have VT-x or VT-d on the die, or hyperthreading or not, or turbo or not right. AMD can do the same thing, like shut off extra HT links so that they can segment their product line. it'll still be the same die, they can just gimp it for a certain market.

sure the pinouts might be different because if you dont hook up that extra HT link or extra ram channels you wont need more pins etc... but in the end its still the same die under there.

I think thuban is basically istanbul in opteron terms. Magny cours is like 2 istanbuls on the same die (thus the quad memory channels, which 2 instanbuls would have since each has dual channels and thus the extra required pins for those 2 memory channels).

From the little i've read socket F = ddr2 registered, socket C32 = dd3 registered. so its like am2 and am3 for servers

So they do need a new socket because the memory is registered and whatever extra pins and changes that requires. i am fairly sure ont he cpu die itself, that is probably something theycan turn on and off depending on what pinouts its hooked up to. it would be pretty inefficient to make 2 seperate dies for that, but if you just change the sockets and disable parts of the die, that woudl be the most efficient way to design a die

but that said, you need a new socket with different pins connected.

 

[JAG87]

There are 4 things you have to look at to give an informed answer.
Performance in the App or Apps-How big of a performance gap
Power Consumption-Actual power being consumed
Cost-Has to be included in relation to the performance. A 15% improvement doesn't warrant in some cases an additional 50% cost.
Desktops or Servers-Desktops may not require being run 24/7 so that will certainly affect power consumption over a server. A desktop also is cheaper and may be an alternative to a server. Of course this all depends on the apps and purpose of the computer.

and 5th being possible Support Costs-To extend warranties on servers cost more than a desktop. If say a hex-core Intel is providing top performance 2-3 years from now, it may in the end be cheaper to pay a little more for Intel over an AMD system right now and not have to do a complete upgrade 2 years from now, thus saving money.

And Jag87 you do know what facetious means right . You did graduate from hi skool, right. And also, they do make 1366 Xeons.

I do know what facetious means wise guy. You're going way too deep into this, all was asked in the OP is what cpu is more power efficient given the same clock speed and load. It's not mine or your responsibility to do all the accounting, that's the buyers responsibility, all we can do is help provide the information asked.

No, Thuban is a 125W TDP part. I don't know the Lisbon standard power TDP off the top of my head, but it is probably double digit.

And there is an HE (lower TDP) and an EE (significantly lower TDP) that will be available.

The 6 core EE will have a TDP down in the 35W range (which is something that we have said publicly), which puts the performance per watt in a different class. I'd be willing to bet that a 2P Lisbon platform could have a lower power draw than a single socket Thuban.

The two products are targeted at different uses.

Yes, sure, the day AMD releases a 35W six core processor that works at more than 1.6Ghz, you call me okay?