Intel Xeon E5-2430L Review and Benchmarks with Dual Processors

1. Can anyone tell me why you’d get 2 of these instead of one 1650?

   They cost 2.5 times as much, EACH! So 5 times the total price for just about the same performance.
   The motherboards are more expensive – at best it’s about the same.
   The power consumption is higher. Though the full load power consumption is lower, this is a rare workload in most server environments.

   And in case you have per socket or per core licensed software running on your box, you’re even worse off.

   I’m not bashing the CPU’s, I genuine just do not see the market for them.

2. spazoid –

   less power consumption even w/2 cpu. also you get 12rdimm slots for memory instead of 8. you are right on p/ core license but for memchached – litespeed – nginx or other open source web apps these work fine.

3. Yes, less power consumption at 100% load (which it will almost never hit in the real world, and even if it does, it will probably be for a very short amount of time).

   BUT!
   The amount of CPU’s, cores, cache sizes etc. really doesn’t matter – performance for your usecase does. And since Patrick is good at showing performance in a variety of applications, we can conclude that a 1650 and 2x 2430L’s are about the same performance.

   And yes, it does have more memory slots, but I doubt most of us will be using more than 256 GB RAM anyway.

4. If your use case requires availability of more than 12 threads at one time then a dual processor system is likely to cost less than two single processor setups with some sort of load balancing.

   It was also use half the power based on Patricks test results, potentially half the rack space and only one set of hard drives.

   Two cpu systems also have more ram slots so whilst both systems can achieve 256GB ram, the dual processor system can utilise more cheaper lower capacity ram sticks to get to that level than the single cpu system.

5. I’ve never heard of an application that requires a certain amount of cores or threads – can you name one?

   The 1650 setup does not use double the power of the 2430L setup. It uses 16% more power under full load, and and 19w LESS (about 30%) at idle. Using more, lower density DIMMs will make the power statistics even worse for the dual CPU setup, although you might save a few bucks. But with the price of memory these days, it’ll be a fraction of the total system cost.

6. You are missing the point.

   Servers in multiuser enterprise environments can often benefit from multithreading and it is not always about computing power but the ablility to handle multiple tasks at the same time.

   If you must have an example then MS Exchange server sepcifies a minimum number of cores per role along with recommended and max cores. Microsofts article on capacity planning for MS Exchange 2010 servers is here -> http://technet.microsoft.com/en-us/library/dd346699(v=exchg.141).aspx

   Financial messaging systems dealing with multiple time critical message flows in parallel. Parallel batch processing systems. Virtualisation hosts with multiple VMs all fighting for cores or threads at the same time. These can all benefit from more cores and threads.

   If there is a requirement for more cores or threads and especially if the vendor recommends turning off hyperthreading like MS does for production Exchange servers then the 1650s 6 cores start to look a little low compared to a dual processor system taking the same rack space and able to give a 30% power saving (your figures) at load.

   Whilst the processor may be more expensive per unit, when you start pricing a dual processor system compared to two single processor systems the cost differences become far more pronounced. Add to the fact that, just as an example, HPs current dealer pricelist, where I am, currently has 16GB RIMMS going for twice the price of 8GB RIMMs (S$304 vs S$654 SRP and sometimes much more depending on server model). Putting in 96GB sees a S$300 cost saving per server with dual processor which, on entry level base units can be a 5%-10% cost reduction. Start scaling this up over 10 servers and then try to explain to stakeholder how the company should not care about saving 3k on hardware purchases.

   The price ratios between the sizes are also comparable at a dealer pricing level as well and not just at a retail level.

7. More cores is not always better. Faster frequency = lower latency. Often, processing 2 tasks at twice the speed sequentially is better than processing 2 tasks in parallel.

   Your total workload will be done in the same amount of time, but your first task will be done in half. This lowers latency, which a lot of applications love. Your “multiple time critical message flows” for example. A faster clocked CPU will give you a lower latency on average.

   The technet article has no mention of clockspeeds, which tells me that they’re referring to sockets/cores/CPU’s running at the same frequency. No matter what, it’s a silly article, that can be used for.. about nothing, really. When your users complain about Outlook being slow, you can’t throw a TechNet article in the face and say “but I have 24 cores, and MS says it should be good enough”. A core is a poor measurement for performance. Architecture, frequency, cache sizes, memory latencies – all these matter just as much in high performance applications.

   I think you’ve misread my 30% power saving comment.

   I have no idea what kind of premium HP puts on their memory modules, but the company I work for pay A LOT less for memory than that. We’re talking sub $100 dollars for 16GB modules.

8. Your initial comment was “They cost 2.5 times as much, EACH! So 5 times the total price for just about the same performance.”

   You then asked “Can anyone tell me why you’d get 2 of these instead of one 1650?”

   I have shown you why but you are now going on about processors being twice the speed compared to two processors. The E5-1650 is rated at 3.2/3.8MHz and a singel E5-2430L is rated at 2/2.5MHz.

   If you read further on in the Technet capacity guide I linked to you will find a section on CPU capacity which then goes on to provide formulas for how to work out the suggested speed required per core.

   Yep, my mistake. You seem to like using values and percentages when it suits your point but only percentages when it does not. At load the dual system used 30W less power than a single E5-1650 compared to 16W more than the E1650 at idle. The dual E3-2430L setup therefore saves just under 50% more power at load than the E5-1650 saves at idle. The dual cpu setup would be more cost effective power wise if put in an environment where there system would be under heavy load at anything over 1/3 of the time.

   The point about memory was not the exact pricing but the ratio of cost. 2x 8GB is cheaper than a 16GB (4x4GB even cheaper still). Therefore having more ram slots means you can get to a target memory requirement cheaper. I would rather save a few bucks than spend it when not required.

   Now the E5-1650 has faster memory capacity but less physical memory accessible. It also has less PCIe lanes available (40 vs 2x 24).

   So to answer your question of “Can anyone tell me why you’d get 2 of these instead of one 1650?” I would suggest;
   1. People with environments with more high workloads than idle time who wish to save costs on power bills.
   2. People who have a requirement for more than 40 PCIe lanes available.
   3. People who benefit from the ability to parallel process rather than sequentially process.
   4. People who want to save money by using more lower capacity ram.

   Oh and the 2.5x the price per CPU is a bizarre comment. The Intel list price for the E5-2430L is US$662. List price for the E5-1650 is US$583. Checking pricing on Amazon shows even less differences in price. Price differences of boards can be around US$40, offset by the cheaper ram.

   Would have loved to have seen the results in a simulated multiuser database environment with high load.

   RB

9. Not real results.
   Overstated.
   I tested my e3-1225 v2 processor
   The results are different.

This site uses Akismet to reduce spam. Learn how your comment data is processed.