Intel Xeon E5-2640v2 vs AMD Abu dhabi 6320

[ServerIT]

Hi everybody.

This is my first topic here.

I'm managing a new IT solution which will process several hundreds of petitions per second. To run this system, we are purchasing several servers and have requested several quotes from hardware suppliers.

At the moment, we have received two interesting quotes which are relatively similar in terms of RAM, hard drive, etc, but the CPUs are quite different. On one side, we have an AMD Abu Dhabi 6320 which is an octocore cpu running at 2.8Ghz while the other quote contains servers with Intel Xeon E5-2640v2 cpus which are also octocore servers running at 2.0Ghz. The Intel servers are also considerably more expensive (about 20%) then the AMD servers

I haven't found any benchmarks which compare these two cpus. In terms of performance, how do the two CPUs compare?

 

[TuxDave]

It probably would help if you can find a benchmark that correlates with your workload. How threaded is it? Is it FP heavy? Can it take advantage of the FMA units? Or is it integer/branch heavy?

If not, there's always this:

http://www.anandtech.com/show/6430/amd-launches-opteron-6300-series-with-piledriver-cores/3

Our Xeon E5-2600 review showed that the 8-core Xeon E5 was between 12% and 40% faster than the 8-module Opteron at more or less the same clocks (Xeon E5 2660 at 2.2GHz versus Opteron 6276 at 2.3GHz). The AMD benchmarks seem to indicate that the new Opteron is 5 to 15% faster at the same clocks, so a 6386SE at 2.8GHz might be able to stay close to the 2.4GHz Xeon 2665, but the higher TDP does not make it very attractive. The 6386SE 2.8GHz might make sense for some HPC people though. If you can recompile your code (and use FMA), AMD claims that a 2.5GHz 6380 is just as fast as a 2.9GHz 2690.

So maybe you can do some math to adjust the statement above from the 6276 to the 6320: adjust for frequency, adjust for the reduced number of cores (note Anand compared 8 cores vs 8 modules aka 16 cores), adjust for the performance boost and see what you get.

 

[ServerIT]

hi TuxDave,

thanks for the reply. The aplication is quite threaded. The solution consists on encodind/decoding information from a specific protocol and storing info into/reading from a database. I'm not sure if these operations are fp or integer.

 

[jhu]

hi TuxDave,

thanks for the reply. The aplication is quite threaded. The solution consists on encodind/decoding information from a specific protocol and storing info into/reading from a database. I'm not sure if these operations are fp or integer.

Sounds like it's fairly I/O intensive too. So the question is, is the current setup CPU bound or I/O bound and will that change with the new machine?

 

[GreenChile]

If you are mainly concerned with performance you should go with the Intel setup. Not only will the Xeon handle 16 threads vs 8 for the Operton, it has much higher IPC so more work is done during each clock cycle. Also the Xeon uses considerably less energy.

A quick search on Spec gives this.
CINT2006 Rate score of 166 for the Opteron 6320.
CINT2006 Rate score of 280 for the Xeon 2640 v2.

CFP2006 Rate score of 147 for the Opteron 6320.
CFP2006 Rate score of 233 for the Xeon 2640 v2.

You get what you pay for.

 

[DaZeeMan]

Didn't have much luck finding specific benchmarks, but here's a side by side comparison of the two...

http://www.cpu-world.com/Compare/986/AMD_Opteron_6300_series_6320_vs_Intel_Xeon_E5-2640_v2.html

I see that the poster above did find some scores though!

 

[ServerIT]

If you are mainly concerned with performance you should go with the Intel setup. Not only will the Xeon handle 16 threads vs 8 for the Operton, it has much higher IPC so more work is done during each clock cycle. Also the Xeon uses considerably less energy.

A quick search on Spec gives this.
CINT2006 Rate score of 166 for the Opteron 6320.
CINT2006 Rate score of 280 for the Xeon 2640 v2.

CFP2006 Rate score of 147 for the Opteron 6320.
CFP2006 Rate score of 233 for the Xeon 2640 v2.

You get what you pay for.

Thanks GreenChile. I hadn't found those scores. Can you post the link where you retrieved them from? I'd also like to compare some other models between eachother.

Cheers

 

[GreenChile]

http://www.spec.org

 

[ServerIT]

http://www.spec.org

Hi GreenChile,

I've searched quite extensively the spec website but I can't seem to find where you found that data. Can you specify better where you found it?

 

[DaZeeMan]

I had one other thought for you. When I was trying to locate some benchmarks for you last night, one of the sites mentioned the licensing cost per core for server CPUs. I'm sure there are some other forum members around here that can share their experiences on this subject. Do your bids show the associated 'core licensing' fees, assuming they apply in your case?

Here's a Microsoft .pdf that talks about core licensing fees, circa 2012:
http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=9&cad=rja&ved=0CGoQFjAI&url=http%3A%2F%2Fdownload.microsoft.com%2Fdownload%2FF%2FF%2F2%2FFF29F6CC-9C5E-4E6D-85C6-F8078B014E9F%2FDetermining_SQL_Server_2012_Core_Licensing_Requirements_at_SA_Renewal.pdf&ei=JerIUqrNLNbooASMs4KoCg&usg=AFQjCNHK47XwNqWcZLazl4rk2tguAyCQ2g&sig2=TY8GokrRrg1fDjQfG79wew&bvm=bv.58187178,d.cGU

Something else to consider at least.

 

[MisterMac]

The EP Xeon will stomp that Abu Dhabi - both in single and multithreadness.
(And in average consumed power and so many other small things, since i imagine you will not 100% loaded with whatever systems you use).

If price is not a concern\"much more" on the Abu Dhabi setup - compare disks, ram, etc - then get the Intel choice.


As others relate - it sounds like your using a custom application written to accept incoming information via sockets.
(Possibly directly via http or thru middlemen servers).


Your problems will most likely be IO related - are you running a newer MSSQL DB or some with similar characteristics ( Newer MSSQL stores all raw stored procs\queries directly in Memory, to preserve disk time access) - then load up on the fastest disk you can find, more memory the merrier and get the Xeon EP 5.

And then start to consider are you going to cluster DBs on several servers or?

 

[Abwx]

If you are mainly concerned with performance you should go with the Intel setup. Not only will the Xeon handle 16 threads vs 8 for the Operton, it has much higher IPC so more work is done during each clock cycle. Also the Xeon uses considerably less energy.

A quick search on Spec gives this.
CINT2006 Rate score of 166 for the Opteron 6320.
CINT2006 Rate score of 280 for the Xeon 2640 v2.

CFP2006 Rate score of 147 for the Opteron 6320.
CFP2006 Rate score of 233 for the Xeon 2640 v2.

You get what you pay for.

This Xeon is a 1000$ CPU while the Opteron is 283$ ,
actualy the real competition is a 16 cores Opteron
at about 880$ and same TDP as the 8 core discussed here,
i guess that given the scores above a 16 cores Opteron would
yield better scores than the Xeon while being still cheaper
for the whole system..

 

[AtenRa]

The thing to consider/find out, is if you are CPU or I/O bound. If you are I/O limited then speding more for faster HDDs or more/faster RAM will be preferable than faster CPU.

You really need to find out what your workload will be.

 

[ServerIT]

The thing to consider/find out, is if you are CPU or I/O bound. If you are I/O limited then speding more for faster HDDs or more/faster RAM will be preferable than faster CPU.

You really need to find out what your workload will be.

Agreed. Therefore I'll share a bit more on the entire solution.

We will have two types of server. The processing nodes which receive petitions and process these and decide what to do with them. This solution is quite CPU intensive.
Then we have the database servers. The processing nodes connect to the database nodes and do the queries on these nodes.

Since licensing was mentioned above, I can share that we will be using a mysql database in the initial months while we prove the concept. Since the solution is very database intensive, once we grow we will be changing parts of the database to a cassandra solution.

The solutions we are looking at are very similar in terms of the remaining specs. In all cases we are looking at 24GB of ram DDR3 (which should suffice since in our tests we were never able to break the 10GB mark). In terms of HDDs, we are also still deciding what to do, but it seems that at least on the database servers, SSDs or high speed HDDs are a must.

Another interesting piece of information is that the difference in price between the server with the AMD and the Intel solution is about 450$ (the remaining specs are very similar). Hence I suppose I'm getting a better deal with the Intel server.

Other offers include an Intel XEON E3-1230v2 CPU and an Intel XEON E5-2620v2 CPU. That's why I asked Greenchile for access to the other benchmarks to see how those servers compare too.

 

[GreenChile]

This Xeon is a 1000$ CPU while the Opteron is 283$ ,
actualy the real competition is a 16 cores Opteron
at about 880$ and same TDP as the 8 core discussed here,
i guess that given the scores above a 16 cores Opteron would
yield better scores than the Xeon while being still cheaper
for the whole system..

I think you got a few things wrong. Let me see if I can help.

Actual price of the Xeon E5-2640 V2 is $885.
link

Price of closest priced Opteron 6378 is $867
link

CINT2006 Rate score for a single Opteron 6378 is 272.
CINT2006 Rate score for a single Xeon 2640 v2 is 280.

CFP2006 Rate score for a single Opteron 6378 is 208.
CFP2006 Rate score for a single Xeon 2640 v2 is 233.

And TDP is still 20W higher for the Opteron which makes performance per watt suffer even more. This is a good example of why Intel holds 95% of the server market share.