Poor ESXi performance of the R620 vs. R610

Hello Elvis

Has anyone seen or heard of these issues with the R620 unit, or the E5-2650 processors?

No, I am not aware of any issues on the 620 that would relate to your configuration.

I would recommend that you post more specifications of the two systems, and detailed data on what exactly is performing slower. The CPU in the 610 is faster than the 620. If the application is not fully utilizing the 610 CPU then it is not unusual for it to outperform the 620 CPU. The 610 CPU is 6.66GHz faster than the 620's CPU, and there is not big enough of a technological difference between the two to make the new CPU's overcome that speed difference.

If you were to stress the CPU at 100% utilization then the 620 CPU would be able to perform many more processes than the 610, but when not being stressed the 610 should outperform the 620's CPU.

Thanks

Hi Daniel,

Thanks for the quick response. I'll get more specifics regarding what is performing slower, but in a nutshell, it's almost every CPU and Memory test that Passmark runs that is slower on the 620. The R610 is running dual X5550 processors with 64GB of 1055mhz memory.. ESX5i installed on RAID1 15k SAS drives. The R620 is running dual e5-2650 processors with 64GB of 1333mhz memory. ESX5i installed on the RAID SD cards.

                  Passmark ratings for the respective CPUs show the E5 eating the X5550 for breakfast (cpubenchmark.net/cpu_lookup.php).

                 Given the specifications of the E5 units vs the x5550, I just don't see how the E5 unit can be any slower than the X5550 model.. More cache, faster memory speed, higher QPI between processors...and the R610 still bests the E5??? something just doesn't seem to add up. At the minimum, I'd expect the memory access speeds to be on par with the X5550.

Thanks again,

E

Hi,

I am considering to buy an R620. Did you find out what was causing the slow performance? your post made me wonder if it's worth it. you are right, it doesn't make much sense, looking at the specs, you'd expect the R620 to deliver better performance overall.

Thanks

I have a new T620 with two E5-2640 at 2.5GHz.  I am running vSphere 5.0.  the guest OS is Windows 2008 R2.  I also am seeing inexplicable poor performance on the guest OS.  While I do not have the luxury of a second ESXi box at this location to compare with I have run PerformanceTest 7 and gotten results indicating a rate of 77.7 which is rediculous.

Hi,

  Are you running the Dell Optimized build of vSphere? I ended up getting the 2650s swapped with 2670s and performance improved, though I didn't see the blockbuster performance I was expecting.. In the end, it's maybe 10-20% faster in CPU and even in small memory allocation benchmarks against the X5550.. Large Memory benchmark is superior to the X5550.

I am not running the Dell Optimized build.  With the help of Dell support I found that I needed to clear the NVRAM on my T620 server because the motherboard had recently be replaced.  This simple step changed my speed test results from 77.7 to 2660 inside by Windows 2008 R2 Guest VM which is much more in line with what I was expecting from my new T620.  Just for reference, I have an almost identical setup on my oder R610 server and it scores about 1600.  While this is hardly a controlled test, it would seem to imply that a substantial performance boost is to be expected when upgrading from a 610 to a 620 but keep in mind that my r610 was fairly old.

Glad they got you squared away.

We got the same problem here. We have a bunch of R610 servers running on X5660 and X5680 CPUS and our latest server is a R620 running on E5-2640 CPUS. All our servers are running the latest bios/firmwares with the latest ESXi 5.1 and they are part of a cluster. The R620 is so slow compared to our R610 that we cannot run any important virtual machines on the R620 because it's impacting our production environment.

We did a bunch of CPU benchmarks and the R620 always come up between 10 and 50% slower than older X5660 or X5680. We also did some apache response time tests and the R620 virtual machines are around 50% slower. This is crazy. The X5660 are running at 2.80GHz, the X5680 at 3.33GHz and the E5-2640 at 2.50GHz. Sound like this is making a big difference but it's weird that Dell would put slower CPU in their newest server model?

So did the upgrade to E5-2670 fixed your problems?

Hey TT_IT,

      Yup, experienced the same issues I did.. The 2670s did help get the 620s on-par or slightly better in most tests..  You should see similar results with the 2670 CPUs. I'm not running Vsphere 5.1 yet, but can report that the R620s have been running well. One of my units is running the 2667 CPU while the other is running the 2670 unit.

     Thankfully, my rep helped keep the price difference to a minimum, so I'd reach out to your rep and have them make it right!

Try changing your default bios power profile ton the server to "Maximum Performance" instead of performance per watt, or the Dell advanced power profile, and see if your benchmark result improves.

We got a 20% average improvement after upgrading the bios to 1.4.8 and changing the bios power profile to "Maximum Performance". For your information the CPU tests we are doing are "sysbench --test=cpu --cpu-max-prime=20000", "super_pi 20" and  "time echo "scale=5000; a(1)*4"|bc -l". Our sysbench tests are still 20% lower than our old X5660 but it's a 20% improvement over what it was before. Our "super_pi" tests shows a 24% improvement and is now on par with X5660 and X5680. And our "a(1)*4" tests shows big improvement and is now faster than X5660 and on par with X5680.

So we will start doing real tests this week with some virtual machines that we're performing poorly before and we'll decide from there if we keep the R620 or if we swap it with X5660/X5680 CPUs.

TT_IT,

    How did the testing turn out for you?

After changing the "bios power profile" to "maximum performance" instead of "OS contol", the server is almost back to normal. It's still slower than our X5680 CPUs but it now has acceptable performance. So we will keep it in our cluster but if we purchase new server(s) in 2013 I doubt we will purchase the same model.