Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
World Community Grid Forums
Category: Community Forum: Hardware Chat Room Thread: xeon relative performance 1366/v1/v2/v3/v4 |
No member browsing this thread |
Thread Status: Active Total posts in this thread: 17
|
Author |
|
thunder7
Senior Cruncher Netherlands Joined: Mar 6, 2013 Post Count: 217 Status: Offline Project Badges: |
Is there any information in relative performance of newer xeons - compared to say a x5650 per core per GHz running as dedicated WCG crunchers?
I'm tempted to buy something new, but I can't find how much faster e5 is, let alone v2/v3/v4. http://boincstats.com/en/stats/15/host/breakdown/cpu/0/8/0 seems to be what I need, but I can't understand it. Average credit/cpu second doesn't put the cpu's you'd expect at the top (say, a 4790K or something like that). I'd like to see a formula like given the same GHz, per core: socket 1366 -> E5 (+10%) -> E5 V2 (+5%) -> E5 V3 (+8%) -> E5 V4 (+4%). |
||
|
OldChap
Veteran Cruncher UK Joined: Jun 5, 2009 Post Count: 978 Status: Offline Project Badges: |
Not sure how helpful this will be for the v3's and v4's but take a look:
----------------------------------------https://docs.google.com/spreadsheets/d/1ll78D...9Xazh6Bvo2xlII/edit#gid=0 Edit: You can of course copy and paste into your own SSheet then manipulate/sort information to your own priorities [Edit 1 times, last edit by OldChap at Jan 8, 2017 5:39:07 PM] |
||
|
Mamajuanauk
Master Cruncher United Kingdom Joined: Dec 15, 2012 Post Count: 1900 Status: Offline Project Badges: |
This has been discussed many times in various forms, most of which I've been an 'Outside Watcher' not participating, just observing. Although OC and I, along with others have batted this about a few times...
----------------------------------------Mainly because I don't think a meaningful comparison can be drawn unless the various CPU's and versions can be monitored running the same science/project. Given the different complexities and lengths of runtime of the different projects work units etc... Personally, I strongly believe the newer processors produce a better output and with the bonus of lower power consumption. A lot of grid crunching comes down to what hardware is available/affordable to the individual. There are so many variables, from the amount of RAM to the HDD/SSD the system is running on, to say nothing of any other tasks/usage... Something our earlier discussions appear to have confirmed is that the more threads a machine runs, the more economical the system is on power. Mainly down to the reduced peripherals, i.e. HDD/PSU/MB/RAM etc. so 2 x 8 thread systems would cost more to run than a single, equivalent single 16 thread machine with similar peripherals. While this is a little 'Non-committal' hopefully it does show the complexities of comparison. It is only my opinion/conclusions from my crunching and discussions as mentioned...
Mamajuanauk is the Name! Crunching is the Game!
|
||
|
Mamajuanauk
Master Cruncher United Kingdom Joined: Dec 15, 2012 Post Count: 1900 Status: Offline Project Badges: |
Just a further thought to add to the complicated algorithm...
----------------------------------------Some projects run better/faster on Windows, others run better/faster on Linux based machines... It goes on and on. Only a true test would answer this question with any accuracy.
Mamajuanauk is the Name! Crunching is the Game!
|
||
|
enels
Senior Cruncher Joined: Apr 25, 2008 Post Count: 286 Status: Offline Project Badges: |
The main thing that seems to matter is the manufacturing technology. At 22nm and below there isn't much performance difference. Signaling the end of Moore's Law. Newer Xeons may use less electricity but it isn't that significant.
----------------------------------------Edit: At 22nm Intel intoduced another ALU and AGU to each core, which provides a significant advantage over 32nm Sandy Bridge. Since then not much has happened to increase Instructions Per Clockcycle. AMD is introducing their 14nm Zen processor soon in 8+ thread versions. It may not be as efficient as Xeons, but they are pricing lots of threads at a relatively inexpensive price compared to Xeons. I've heard a 40% increase in IPC but we shall see. [Edit 1 times, last edit by enels at Jan 8, 2017 7:32:57 PM] |
||
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7219 Status: Offline Project Badges: |
A pretty good site to compare the performance of the various CPU's is :Passmark . They list some prices and have a performance index and value index. Here are their top ten for performance and their top ten for value. Just remember the cautions from enels and Mamajuanauk. The prices on the performance CPU's is WOW!
----------------------------------------Cheers
Sgt. Joe
----------------------------------------*Minnesota Crunchers* [Edit 1 times, last edit by Sgt.Joe at Jan 9, 2017 1:59:35 AM] |
||
|
thunder7
Senior Cruncher Netherlands Joined: Mar 6, 2013 Post Count: 217 Status: Offline Project Badges: |
So the difference between an E5-2699 V4 (22 cores / 2.2 GHz) and an E5-2698 V4 (20 cores / 2.2 GHz) at 17% from 32248 to 37929 is more than the 10% difference in cores?
Strange. And in the google spreadsheat in the second post, a dual E5-2695 ES V4 @ 2.8 GHz differs from a dual E5-2683 ES V3 @ 2.3 GHz by exactly the difference in the clockspeed, running the same OS and the same tasks - which implies v3 to v4 is no upgrade at all. If I compute the google spreadsheet and divide Boinc PPD's mixed units by the number of cores and the clockspeed, to get a PPD per core per GHz - the highest number is from a 2x Xeon E5450 @ 3 GHz, which comes in at 324 PPD/core/GHz (7777 PPD for 8 cores) compared to an E5-2628L V4 @ 1.5 GHz which comes in at 277 PPD/core/GHz (10002 PPD for 24 cores) - both running linux. If I remove the windows machines, I get something like v0 (socket 1366 and earlier) 230 PPD/core/GHz v1 256 PPD/core/GHz v2 265 PPD/core/GHz v3 264 PPD/core/GHz v4 280 PPD/core/GHz I'm no wiser than when I started this.... |
||
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7219 Status: Offline Project Badges: |
Remember, you need to look not only at the efficiency of the CPU, but also at the efficiency of the total machine. Look at the total wattage used by a machine and then do the math. For instance, use of a solid state drive versus a spinning drive will reduce total wattage. Use of more efficient fans, different case designs, different motherboards, more efficient heat sinks or even a more desirable location to take advantage of air currents will reduce total wattage used by the system. It is the total efficiency of the package, not necessarily the efficiency of individual components.
----------------------------------------Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
thunder7
Senior Cruncher Netherlands Joined: Mar 6, 2013 Post Count: 217 Status: Offline Project Badges: |
I used the column BOINC PPD mixed wu's - which leaves out wattage. Still, it appears replacing a quad Opteron 6282 which is doing about 30k PPD is hard (only the very best cpu's are faster).
|
||
|
fuzzydice555
Advanced Cruncher Joined: Mar 25, 2015 Post Count: 89 Status: Offline Project Badges: |
You may not see an increase in absolute performance, but you gain a lot in efficiency.
----------------------------------------771/E5450: 110W - 3300 PPD - 30 Points/Watt 1366/X5650: 164W - 6400 PPD - 39 Points/Watt 2011/2660 dual: 236W - 18000 PPD - 76 Points/Watt 2011v3/2683v3: 152W - 14000 PPD - 92 Points/Watt 2011v3/2628Lv4: 83W - 10000 PPD - 120 Points/Watt This means that for the same money spent on electricity, you get 4x points with the v4 xeons, compared to the old 5400 series. You can get a 12 core 2628Lv4 for 130$ on ebay, 2011 CPUs are even cheaper. 2011/2011v3 boards are pretty expensive though. If you can get a watt meter reading on your quad opterons, watts/PPD will give you your efficiency. [Edit 1 times, last edit by fuzzydice555 at Jan 20, 2017 10:55:28 PM] |
||
|
|