Intel Optane DC P5800X PCIe 4 NVME SSD Review – SSD Perfection Via Throughput, IOPS and Latency

ATTO DISK BENCHMARK VER. 3.05

ATTO Disk Benchmark is perhaps one of the oldest benchmarks going and is definitely the main staple for manufacturer performance specifications. ATTO uses RAW or compressible data and, for our benchmarks, we use a set length of 256mb and test both the read and write performance of various transfer sizes ranging from 0.5 to 8192kb. Manufacturers prefer this method of testing as it deals with raw (compressible) data rather than random (includes incompressible data) which, although more realistic, results in lower performance results.

INTEL Z590 TEST BENCH

AMD X570 TEST BENCH

blank

CRYSTAL DISK BENCHMARK VER. 7.0.0 x64

Crystal Disk Benchmark is used to measure read and write performance through sampling of random data which is, for the most part, incompressible. Performance is virtually identical, regardless of data sample so we have included only that using random data samples.

INTEL Z590 TEST BENCH

THROUGHPUT

blank

IOPS

blank

LATENCY

blank

AMD X570 TEST BENCH

THROUGHPUT

blank

IOPS

blank

LATENCY

blank

Intel Optane DC P5800X PCIe 4 NVMe SSD results on the Intel Z590 Test Bench are simply the best we have ever received from a single form factor SSD.  Just under spec listed 7.2GB/s read, 5.6GB/s write, 1.5 million IOPS and a low latency mark that we have never seen before make this one incredible SSD.

24 comments

  1. blank

    For the price I would consider the SSD has at least 16 PCIe lanes and not this bottleneck with 4 lanes.

  2. blank

    If the SSD has sequential write speed of 5.6 GB/s, how comes that in that 425 GB file transfer test on page 5 it shows only 2.45 GB/s? Was it bottlenecked by reading speed of another SSD used as source? This is a faulty test, I think. You make a conclusion that P5800X does not drop its write speed, but in reality you tested that it only maintains 2.45 GB/s speed. You did not test how it maintains its full write speed. Your max temperature test is also faulty, since it is based on the same file transfer test limited to less than half of max write speed. If you transfer the same 425 GB folder at max speed of 5.6 GB/s (using a pcie gen 4 SSD as source) then max temperature could be way higher.

    • blank

      Despite your best efforts at negating the validity of this report, I always welcome the opportunity to assist others. If you have a decent grasp of storage, you will understand that listed specifications with respect to flash storage are just that, listed specifications. They are a unrealistic high that are seldom reached in reality, and do more to sell SSDs and storage than serving any real purpose. Several years ago I did a study of just this and determined that an SSD EVER reaches its real peaks only about .04% of the time. This is exactly the same when, in your scenario listing the ‘sequential write speed’ you are hoping to hit these highs. This is nothing new and I don’t know of a single SSD in the hundreds that have been through our hands that has ever reached its peak (or sequential write high) as you discuss here, during file transfer…not one. In fact, of all the reports I have ever seen in the industry, I have not ever seen even one. Now, there are many factors that are the cause of this, the first being of course that manufacturers determine their listed specifications with of benchmarks such as ATTO that provide a high performance variable through sequential data transfer. As well, different types of data transfers at different speeds depending on size, compression used and other variables, this being front and center in our true testing. Next up…temperature. If you are relying on your first point to validate your second, this point has now been negated but… Temperature is very easy to determine while monitoring a SSD during testing, which is explained very clearly.

      • blank

        I’m not “negating” anything – I just asked if that write speed could be bottlenecked by reading speed of the source disk. Based on your response, I assume it wasn’t? I currently have Samsung 970 Pro, and it reaches 2 GB/s in actual file transfer speed (according to the same Windows popup window, not according to synthetic tests). That’s why I thought that 2.45 GB/s in real file transfer speed for P5800X is kind of low, taking into account how much more advanced it is. Either way, thanks for responding.

      • blank

        Prior to responding, I had a few industry peers provide thought on your input. The received response for the most part was to simply delete or ignore your comment as you state how the testing was faulty, prior to simply asking why such and such result was achieved. I elected to respond to assist. Setting that aside, you have shown exactly what I described with the 970 Pro reaching 2GB/s in actual transfer speed. By your model, I would ask why it doesn’t reach the full 3.5GB/s transfer speed. Hope this helps.

  3. blank

    Les, I’m not real familiar with the Ryzen architecture but wanted to know if it supports any PCI-E lanes direct into the CPU like the newer Intel CPUs do. I’d be curious to know if that generates any better performance specs on SSDs over running them through chipset lanes.

    • blank

      AMD Ryzen CPUs (And more specifically the x570 motherboards) support PCIe lanes from the CPU natively, and others via the chipset. It all stacks up very similarly to Intel’s solutions.

      In general, the x16 graphics slot + 1 x4 NVMe slot generally come natively from the on-die PCIe lanes, and the rest come from the chipset.

      Graphic here from Anandtech’s x570 motherboard roundup, showing the lane breakdowns:

      https://images.anandtech.com/doci/14161/X570.png

      • blank

        So, there are some caveats with that. Yes, there are 4 lanes dedicated to the first NVME / m.2 slot, but they are electrically connected to that M.2 slot. You would have to use an M.2 to U.2 adapter of some sort to plug the drive in to the m.2 slot. It can be done, but PCIe 4.0 is a real stickler for signal integrity and trace length. I have not seen a good u.2 to m.2 adaptation setup that reliably supports PCIe 4.0 speeds. I have some parts on the way that will be tested soon to see if the m.2 socket can be made to work properly with this drive.

        The author is using a PCIe carrier card that adapts the drive directly to a PCIe slot. This is a viable connection method but has some drawbacks for an enthusiast system. Consumer amd be intel systems will bifurcate the 16x PCIe slot that the gfx card uses down to the required two 8x PCIe slots if you plug the carrier card into the motherboards second cpu direct attached 16x slot. While this is fine for the nvme drive, it does create some minor performance issues for the gpu if you have a baller enough GFX card (rtx 3090. . .).

        Plugging this kind of drive into a chipset connected PCIe slot would likely negate the benefit of having a drive that has the performance characteristics that this one has.

  4. blank

    Thanks for your review. I’m interested in random performance and latency for my apps. My suspicions have been confirmed. Cheerz.

  5. blank

    Want the AMD #s up there with Intel #s?
    Simply mod the Intel driver’s .inf file a bit.

    My 800P:
    Q1 R4K read: 200 to 291 MB/s
    Q1 R4K Write: 161 to 217 MB/s

    Screenshot and linked driver in my FarceBook post here:
    https://www.facebook.com/groups/AMDHardwareEnthusiasts/posts/3083610618363854/?__cft__%5B0%5D=AZW0Ohn2H97vVE3CtHDwoLDG7A4Rx-Hol8tVupcZY_LedzCFmhRTw4si_OjFNc1JrhFN-tM1OPZppHhQfvBJQJETKYrhZRUyF-flb2_lGKd6nwvOrKE5w9ICUakxxv_8dSKM4xBxt4hhWMV_JiJ9R_1U&__tn__=%2CO%2CP-R

  6. blank

    Any chance to get AMD vs Intel CPU benchmarks on Linux (e.g. Debian)? AMD usually performs better on Linux than Intel. Would be interesting to see if that carries over to the SSD benchmarks.

  7. blank

    The best thing about Optane is not the durability or the latency, it is that it doesn’t need a cache. You can up the copy test size as much as you want and it will stilll go like hell.

    There is no cache misses with Optane. I am done with Nand, I just want Optane v2 to get cheap enough that my wife won’t kill me when I buy it.

  8. blank

    Beautiful work. I also bought a P5800X 800GB and it did take a bit of time optimizing the performance. I used your result as a reference for all this and it saved me a lot of time.
    BTW, are the drivers generic microsoft drivers or did you install intel drivers? Under my environment, it had a 10-20% performance boost (X570 + 5950X connected to CPU direct PCIE 4.0 x4).

  9. blank

    Hi guys,

    The P5800x drive come with the pci express card adapter 4x?

  10. blank

    Hey man, Thanks for this incredible review. Man those random reads are freaking crazy. It baffles me that the 4kQ1 reads match a SATA SSD’s sequential reads. Since random reads are what result in system responsiveness, did you notice any performance difference between this and a SATA drive like 870 EVO or 970 EVO Plus in everyday tasks like opening Premiere, Photoshop, browsers etc. I did personally did not notice a difference between SATA and NVME with everyday use which I believe could be because of similar random read performance. So how about the Optane vs NVME or SATA in everyday use? Thank you

    • blank

      That is actually a great question which is something I speak of most often. You will never see any difference in any SSD, be it SATA, NVMe, PCIe or what have you, in every day tasks and many software applications. The reason for this is that the upgrade we see in everyday use comes from the incredibly fast access of SSDs which is always very similar at .01-.03ms. I challenge many to tell the difference between fastest and slowest SSD on a regular basis. Cannot be done. In fact, most people don’t even use software that will challenge the SSD… software such as Adobe Acrobat or PhotoLab 5. both of which I use. Now…moving data in large size is a different animal, especially for media files.

  11. blank

    maybe AJA test should add 4k 10bit 422,the reference value is limited for 1080p, no one will pay a lot of money to buy the best hard drive just to process 1080p video

  12. blank

    What is the brand of the U.2 to PCI card you used for those tests? I can’t find it in the review…

  13. blank

    hello, thank you for this amazin review! for me its still not clear wher you connected the p5800x in the amd build, was it the lanes to the chipset or to the cpu directly?
    would be amazing to clarify what pcie slot you used on the motherboard.
    cheers!

Leave a Reply

Your email address will not be published. Required fields are marked *