Upgrade FROM Intel 750 TO Samsung 950 Pro - Random Read (IOPS) below spec

Screen shot below. The Sequential Read / Write are within spec. The problem is that the Random Read (IOPS) benchmarked at 185,000 out of expected 300,000. The Random Write benchmarked at 93,000 out of 110,000.

http://imgur.com/9m7b04g

*** Background ***

I dropped the new 950 pro on the floor when opening the packaging. The SSD appeared undamaged but I wonder if incident is related to performance numbers.
FYI: The Intel 750 AIC 800 GB is my existing system storage.
Today I received and physically installed the Samsung 950 Pro 500 GB (M.2) on my Asus Sabertooth X99 MB.
Updated BIOS ver from 1802 to 2002
Installed Microsoft hotfix "Update to add native driver support in NVM Express in Windows 7 and Windows Server 2008 R2"
Using Samsung Migration utility, I cloned from source (Intel 750) to destination (Samsung 950 Pro)
Installed Samsung NVMExpress Driver rev 10 (I probably should have installed this BEFORE cloning for faster transfer times, but live and learn).
Installed Samsung Magician utility. Verified that firmware was latest.
Tested using performance benchmark (results above and in imgur image).
Ran Performance Optimization.
Re-tested using performance benchmark (results unchanged).
 
Solution


300k 4k random is the *max* spec, meaning for a mostly empty / largely overprovisioned SSD. Full SSD performance is closer to 200k on the 950 Pro. Your 185k figure is not bad for a mostly full drive, actually. Here is how the 950 Pro performed on our recent (not finalized) steady state 4k random testing:

dist-2.png

perc-1.png

(Source)

Total IOPS for each SSD is listed along the right edge of the above images. Our 950 Pro sample saw 172k in this test, which is lower...
Yes, I previously installed that very same driver from that same Samsung webpage. Unfortunately, the deficiency in my random read/write speeds remain.

Question: I noticed from your signature that you're also in possession of the 950 Pro. Does your random read benchmark reach the full 300,000 IOPS?
 
That's very interesting. I'm now wondering if the SSD is working perfectly but is bottled necked by our CPUS. You're running a i7-4790 with a base and boost clock of 3.6 and 4.0 GHz respectively. My i7-5820k is running with a base and boost clock of 3.3 and 3.6 GHz respectively, which is possibly why your Random Read / Write figures are closer to spec.

When PC Perspective's Allyn Malventano did his performance review on the Samsung 950 Pro back in October, with the identical motherboard and cpu in his test bench, his 5820k was clocked at 4.125 GHz. Also the 16 GB of memory in his rig was clocked to 3333 MHz whereas mine is spec'd at 2400 and overclocked to 2666.

PC Perspective: Samsung 950 PRO 256GB and 512GB M.2 NVMe PCIe SSD Review
http://www.pcper.com/reviews/Storage/Samsung-950-PRO-256GB-and-512GB-M2-NVMe-PCIe-SSD-Review/Internals-Testing-Methodolog
 
while i'm not computer literate enough to state this authoritatively, my cpu is bottlenecking my system

I render a lot of video files as a hobby - I noticed my cpu usage ran 98-100% while rendering. When i went from a 840 Evo SSD (Rapid Mode not enabled) to a xp941, my rendering time on the identical file dropped by 55-60%, from 75 to 31 minutes, which made the xp941 worthwhile.

The xp941 has advertised specs of 1180 MB/s read, 900 write, and is PCIe 2.0 x4 and i had it installed in a PCI 2.0 x4 slot. I figured the 950 PRO might help me bring the rendering time some more, but when i ran the same file, virtually no improvement over the xp941, so i concluded it had to be bottlenecked at the cpu, because of the 98-100% usage. And i tried rendering in a number of different SSD combinations, ie reading the file from the xp941 and rendering (writing) to the 950, and vice versus, reading from a 840 EVO and writing to the 840 (840 has RAPID Mode enabled, and showing 6600 MB/s read, 2500 MB/s write) - and again tried it in reverse, reading from the xp941 and rendering to the 840, and in combo with the 950, both ways. I even tried using one of my storage 5400 HDDs and saw no better rendering times. Another poster in another forum indicated he'd seen the same results.

Conclusion we came to was to reduce rendering time i need to upgrade to one of the new 8 core / 16 thread CPUs,
which i won't be doing for the next year or so.

Hopefully something in that descript tells you something - but you have given me another consideration, ie clock speed

btw, my ram's speed is 1600 MHz
 

malventano

Distinguished
Dec 28, 2015
2
1
18,520


300k 4k random is the *max* spec, meaning for a mostly empty / largely overprovisioned SSD. Full SSD performance is closer to 200k on the 950 Pro. Your 185k figure is not bad for a mostly full drive, actually. Here is how the 950 Pro performed on our recent (not finalized) steady state 4k random testing:

dist-2.png

perc-1.png

(Source)

Total IOPS for each SSD is listed along the right edge of the above images. Our 950 Pro sample saw 172k in this test, which is lower than what you saw on yours, but our preconditioning pass was 4k random writes to equilibrium, which is harder on the controller as it fragments the flash / tables further than your drive there.

Allyn Malventano
Storage Editor, PC Perspective
 
Solution

Piiilabyte

Commendable
May 20, 2016
10
0
1,520


950 Pro
Seq. Read = 2500 MB/s
Seq. Write = 1500 MB/s
Random Read = 300,000 IOPS
Random Write = 110,000 IOPS

Intel 750
Seq. Read = 2100 MB/s
Seq. Write = 800 MB/s
Random Read = 420,000 IOPS
Random Write = 210,000 IOPS

Conclusion:
Seq. Read/Write (especially Write is extremely faster than the Intel 750) winner: Samsung 950 Pro
Random Read/Write winner: Intel 750 (by far).

I would also suggest using RAMDisk, a tool that allows you to use your SSD memory as Virtual RAM, which means random access memory, and this is where the Intel 750 really shines.

My personal choice?
Samsung 950 Pro ALL THE WAY BROOOOOOO!

My Source:
http://ssd.userbenchmark.com/Compare/Samsung-950-NVMe-PCIe-M2-512GB-vs-Intel-750-Series-NVMe-PCIe-12TB/m38554vsm25111

Hope this answers your question.
 

Piiilabyte

Commendable
May 20, 2016
10
0
1,520


You're using a Samsung tool to benchmark Samsung AND Intel? Obviously the tool would favor Samsung, but it could be misleading. Try a 3rd party tool.