I purchased an Intel 750 PCIE NVME in hopes of fast performance. After installation I was entirely underwhelmed. Benchmarks support my lack of enthusiasm. My Benchmarks have something like 60% of the performance shown in reviews and other consumer benchmarks.
Intel 750 PCIe 400gb (4x Gen 3, see attached pic for confirmation)
Asus Rampage Extreme IV (x79 platform)
Not likely relevant:
64gb 2400mhz DDR3
3x GTX 780s SLI
3930K @ 4.75ghz
I have confirmed PCI speeds on my SLI cards using GPU-Z and that Intel SSD toolbox reports Gen 3 with 4x link. Latest drivers (no less than a month old, latest firmware installed). The device is slow. It doesn't even really outperform my Samsung 850 Pro 512gb SATA 3 SSD. What am I doing wrong?
We understant you're experiencing slow performance on your Intel® SSD 750 Series boot drive.
It's always important to keep in mind that benchmarking your SSD while being used as your primary drive will never return accurate results. These tests will be affected by your OS running off the drive, Windows* updates being downloaded, your antivirus background activity, among other factors (the fuller the drive is, the slower it will benchmark, for example).
With this note aside, we would like to confirm that you're using the latest NVMe* drivers for your SSD:
- https://downloadcenter.intel.com/download/26833/Intel-SSD-Data-Center-Family-for-NVMe-Drivers?v=t Intel® NVMe* Drivers Version: 220.127.116.113.
We look forward to hearing back from you.
Unfortunately the drivers you linked to are the drivers I have installed. I understand variance can be expected, but this is egregious. Others are consistently much higher speeds. I didn't pay a premium to go from SATA to PCIe NVME only to barely squeak out a few extra points of sequential read/write. What are the types of things that could be used to diagnose this issue?
I would also like to note that loading Windows is slow compared to videos I have seen from others with the same hardware (motherboard and Intel 750). I have confirmed CSM is disabled, Fast boot enabled, GPT partition and the Win 10 OS was installed using UEFI bootable media. I recently reinstalled to confirm this.
The best way to diagnose performance issues is to install this SSD as a storage drive, format it, and then run benchmarking tests. Although you may skip the formatting part if you wish, as your drive is not full enough to be considered a handicap. You can follow the http://download.intel.com/support/ssdc/hpssd/ssd-750/sb/intel_ssd_750_series_evaluation_guide_332075... evaluation guide for the 750 Series, although CrystalDiskMark* is accurate enough for most ends.
My main concern here is that your write speeds are actually well within spec at 86.7% of the advertised speed. Your reads are definitely slow at 68.5%, but this could be explained by your drive being in use while the tests ran.
1507 * 100 / 2200 = 68.5%
780 * 100 / 900 = 86.67%
As for the numbers you shared from other users, those may be from two 750's being used in a RAID 0, since those results are well above the expected performance numbers of 2200 MB/s sequential reads, and 900 MB/s sequential writes for the 400GB model. Do note that screenshot comes from an empty, secondary drive.
Thanks for your help Carlos. For the sake of argument I will attempt to boot off another drive and use the 750 as storage to see how it stacks up. However I should note that in my original post I meant to draw more focus to my 4k random read/write. I sit around 22 while other score upwards of 40, some even 50. This is what I meant by the egregiously slow speeds. As a significant number of (for example) Windows 10 OS loading is done using these smaller files, this would help explain the slow boot performance. While it is faster than a platter hard drive, its not the 400$ experience I was hoping for.
See attached. Another benchmark. This is the 1.2tb drive, but I wouldn't expect half that performance like I am getting now (and since I purchased it).
We can only provide advice in regards to your sequential results using this tool. To dive into random benchmarks we would need to follow the evaluation guide using IOmeter as linked in my previous reply. As it stands we don't have a good comparison baseline, as larger drives do have much faster random results.
For most purposes, such as gaming, home use, enterprise environments, only the sequential benchmarks are relevant when it comes to SSDs. We only normally have to look at random read/write speeds when it comes to heavier workload server systems. This is usually measured in IOPS, not MB/s like their sequential counterparts.
Although there's a lot of useful information on pages 5 and on, which I can definitely recommend reading if you're so inclined and got the time, to set up the 4KB Random Writes test we can skip ahead to Section 4.3 (page 17). Then the instructions for the 4KB Random Reads test begin on Section 4.4, page 20.
There's also a section on reading the results, but I can gladly help with this if you share your results along with your reply.
Thanks for following up Carlos. I began doing as you instructed and collecting screenshots to support it. I do not have that information in front of me at the very second as I am at work, however the data I collected was terrible. If I am not mistaken, I was getting 1/4th the random read and half the random writes IOPS.
As I discovered this I made up my mind to switch back to my SATA SSD. I used a hard drive cloning software to take my Intel 750 to back to my SSD. You advice suggests the conditions that these test are performed under use the Intel 750 as storage only, with the idea being that any read/writes occurring as a result of the OS using resources or another program using resources could result in a difference in performance.
I would not expect the performance to be so degraded (especially because I don't take those steps to test my other hardware). I just finished the process the other day, I just having gotten around to re-preforming the tests.
I will admit that the procedure provided is convoluted and it would be nice if a script was provided to run IOMeter in the manner suggested by Intel for this test. Not sure if IOMeter allows for such a thing, but just to get the test going takes around 20 mins to half an hour if you are following along with the manual.
Storage benchmarking can get complicated. It's not that you can only achieve these speeds using the drive as storage, but rather that this is the only way to accurately perform the tests.
Imagine being asked to solve arithmetic problems while someone else times your responses. Now imagine doing the same thing, but now you're juggling, holding the stop watch, and writing down the results yourself. You're not taking longer to answer the questions, but to someone who isn't aware of all the overhead, it may seem that way.
Thanks for the IOMeter scripts suggestion, must admit I never considered this. I'm not sure if it's possible, but if it is, it would be wonderful. I'll pass this feedback along.