I am implementing a Solution VI 3 and with 2 dl380 g5 from HP what's the best choice ? 1 Quad core CPU or 2 Dual core CPU ?
Technical :
2 Dual Core CPU (because of INTEL MULTIPLE CORE architecture) is a better choice than 1 Quad core CPU. The difference will be...not perceptible for ESX Server (as long as you've got a strong motherboard for this architecture)
Financial :
1 Quad CPU only need :
\- 1licence for ESX
\- 1licence for VMotion
\- 1licence for VCB
\- etc.
2 Dual CPU need :
\- 2 licence for eSX
\- 2 licence for Vmotion
\- 2 licence for VCB
\- etc.
So
=> If the price of 1licence for ESXVmotionVCB is cheaper than the difference between 2Core and 4CORE, take the 2CORE
=> Else, take the 4CORE
Kingest regards,
Seb.
OK Tanks but the problem is finance but is not on the license but in price off 2 dual core versus 1 quad core , because 1 quad core its more cheaper than 2 dual core the all solution is with 4 dl380g5 , i just need to understand if the performance is not halted
The perfromance wont be halted, but it wont be the same either. While you ahve 2 dual cores, and 4 total cores, you have to slots... to paths for the information to flow to other parts of the server systems (memory, disk etc etc).
Also lets say (and just usign a number here) that you only get 80% of the perfromance on a second core (not a 2x increase over a single core proc for overall perfromance). Then in a dual core 2 socket system you have:
100%80%100%+80% = 360
Oon a quad core (again jsut an example not real numbers)
100%80%80%+80% = 340
Operations that are core heavy (tcp operations etc) like more cores. Operations that are going to use a lot of memory, tons of disk IO etc, I would still be worried about the number of sockets somewhat.
So is there a HUGE perfromance hit? depends, probably not HUGE but it depends on the types of work you will through at this.
Another thign to consider is that I think Vmware only officially supports at least 2 sockets installed even if you could get a quad core single CPU to run.
Ron
OK Ron that help a lot
There was a nice article in the german iX magazine comparing SMP performance of Intel quad-core versus AMD dual-core.
I know that an Intel - AMD comparison is like an apple - pear comparison, but the results are nevertheless interesting.
Quad-cores are good at number-crunching but loose significantly when memory and I/O comes into play.
And in ESX we need memory and I/O.
The issue with the quad-cores seems to be the memory interface.
I was looking at these charts yesterday and found them very interesting:
http://www.tomshardware.com/2006/12/06/quad-core-xeon-clovertown-rolls-into-dp-servers/page7.html
The issue though is if the Intel design on the quad core is poor enough where it will impeded any benefits when the vmkernel is trying to swap the code in & out.
Message was edited by:
kreischl
unfortunately most of these benchmarks are either "number-crunching" benchmarks or desktop related
That is true...but I didn't expect the quad core to be slower in that many tests.
I expected it for a few reasons
The quad core cache architecture is more complicated
The memory / I/O bus is a limiting factor (4 cores competing instead of 2)
The chipsets are newer and (probably) buggier
Most applications aren't SMP aware / aren't able to really use SMP.
To get good results with 4 cores, you have to use highly optimized code and therefore use compiler settings you normally won't use for production.
Umm.. no.
While the cores may require additional pathways, EACH core has it's own pathway to the memory. So the peformance of a quad core is actually FASTER than ANY dual core, even if they are separate.
Just because you have 2 sockets, does not imply 2 SEPARATE paths to memory, bus, and PCI. They still share the path.
Quad Core is much faster than Dual core, even in real world tests, and especially ESX. AMD is faster than Intel for pure memory operations, but that is another story.
But your numbers are off. A core isn't like hyperthreaded processors, not sure where you got that info, a core is a core, period. They are full processors.
That being said, that in this scenario yes a Dual Core (if they are separate sockets) is better than a single CPU quad core, because of the point of failure, but that would be the only reason.
Quad core is more expensive, but not by much. Quad core, even though slower per core, is more efficient. The overall speed you get is far and above that of dual core, 2 processors.
EACH core has it's own pathway to the memory.
this definitely is not true
They share the 3rd level cache - how could they have separate paths?
Xeons are UMA systems.
That's not what this says.. and I have more numbers..
"After Effects comps, the quad core machine beat the dual core in every single test, but wasn't able to best two of the benchmark results of the Dell single-processor Core 2 Extreme machine running at 2.93GHz we tested a couple of months ago. But after those small defeats, the Clovertown machine couldn't be touched again by any of our contenders, especially with the Nightflight After Effects composition, which was devised to show the power of the Macintosh over the PC using extensive vector graphics rendering. The quad core machine crunched through that in 14:23, besting its closest competitor by more than 25%.
"
http://videoediting.digitalmedianet.com/articles/viewarticle.jsp?id=83846
Plus even VM ware has whitepaper that shows that Quad Core is WAY better than Dual Core. Maybe OTHER apps can't take advantage of SMP, but we aren't talking about them, are we? We are talking about VM Ware, which is the \*ONLY* thing that matters.
Hmm.. you need to read up on Quad Core Architecture then.
I meant 2nd level cache
In the Intel specs there's the following sentence
Processors employ Intelî Advanced Smart Cache (shared cache).
With a shared cache it is impossible to have separate mem paths due to cache coherency implications.
Message was edited by:
oreeh
and f you take a look at the Intel datasheets
Four 32-KB Level 1 data caches
Two 4 MB Level 2 caches
http://www.tomshardware.com/2006/03/13/idf_spring_2006/index.html
Shows the intel graphs on all bus/path designs.
And AMD Opteron may share L2 Cache.
In any case, the REAL WORLD testing PROVES that Quad Core is better than Dual Core, even if there is an "apparent" sharing of pathways (which I don't think there is)
So we can argue the finer points of how Intel Archictecture works with respect to ESX, but it doesn't matter, numbers don't lie.
Quad Core is the preferred, and faster method over Dual Core, any day of the week.
I don't think anyone has done 2 Dual core procs vs 1 quad core proc however, but that really wouldn't be a fair test, anymore than comparing DDR2 memory on one bus, versus 2 DDR2 Memory (which isn't really possible because they must be in pairs) but you must compare like processors to like processors.
Quad Core is the preferred, and faster method over Dual Core, any day of the week.
maybe when Intel tests it
quote from kreischl:
http://www.tomshardware.com/2006/12/06/quad-core-xeon-clovertown-rolls-into-dp-servers/page7.html
Message was edited by:
oreeh
Don't trust benchmarks from the manufacturer.
Numbers always lie. As I said above, you can't conclude from SPEC numbers to real life values - especially not with virtualization.
Otherwise we all would run Itanium systems with 32bit applications and wonder where those SPEC results where gone.
RParker, where is that white paper? I'm not having any luck finding it.
Sure " Quad Core beats dual Core" but that was not the question... I mean more cores.... better, more cycles, etc etc. BUT...
The question for this guy was from a perfromance perspective in his ESX box is a SINGLE proc quad core server better for him than a DUAL proc Dual Core server...
The debate / Question from the original poster was Dual Proc Dual core vs Single proc/socket Quad core....
Ron
Message was edited by:
citrix_guy