Top critical review
1.0 out of 5 starsHorrible in all respects
Reviewed in the United States on March 8, 2017
Do not buy this, there's no point. It doesn't behave any better than a SATA SSD and performance is inconsistent. When filling the initial 16GB partition with uncompressable data I got a reasonable 170MB/sec of write bandwidth, but after a few gigabytes the performance suddenly went to hell, dropping to 20 MBytes/sec for several seconds, then going up again, then dropping again, etc until I had finished the write-out. After reading a few reviews I realize this is Intel's SLC front-end cache filling up. Why Intel does something this stupendously stupid I just don't know. NO OTHER NVME BRAND has this sort of write degredation.
After filling up the partition I ran my standard suite of random-read tests. The 600P hardware supports 8 queues, so 16 threads runs a queue depth of 4. I also ran selected tests at 64 threads (queue depth of 8) but performance did not improve, so I didn't include those numbers.
I usually don't bother with a single-thread sequential read test but the part gave such shockingly bad results that I did that too as a one-off, and got around 186MBytes/sec @ 32KB block size. Also ridiculously bad. For example, the same simple sequential read test with a Samsung 960 EVO yields 1200 MBytes/sec. And honestly, this is probably the #1 reason why I'm giving the Intel 600P a failing score. They shouldn't even be selling this piece of junk. The last time I was shocked this badly was with the Intel 750 (the big PCIe card), which completely falls over when I/Os use block sizes 64KB or higher.
Sure, this baby is cheap, only $110 or so. But both the Samsung and Toshiba offerings are only just slightly more expensive and have far, FAR higher performance.
nvme2: <NVME-PCIe> mem 0xc7400000-0xc7403fff irq 34 at device 0.0 on pci3
nvme2: mapped 16 MSIX IRQs
nvme2: NVME Version 1.2 maxqe=256 caps=00100030780100ff
nvme2: Model INTEL_SSDPEKKW256G7 BaseSerial BTPY64430Q5B256D nscount=1
nvme2: Request 64/32 queues, Returns 8/8 queues, rw-sep map (8, 8)
nvme2: Interrupt Coalesce: 100uS / 4 qentries
nvme2: Disk nvme2 ns=1 blksize=512 lbacnt=500118192 cap=238GB serno=BTPY64430Q5B256D-1
randread /dev/nvme1s1b <blksize> 100 16
16 threads - INTEL 600P (16GB area, uncompressable data)
blksize aggregate bw
------ --------------
512 23 MBytes/sec
1024 47 MBytes/sec
2048 93 MBytes/sec
4096 180 MBytes/sec
8192 333 MBytes/sec
16384 570 MBytes/sec
32768 716 MBytes/sec
65536 865 MBytes/sec
131072 930 MBytes/sec
262144 945 MBytes/sec
--
In addition, during a longer multi-threaded read test the 600P hit its temperature limit and auto-throttled. To be fair, this is a good thing.. it hit 75C, flagged it, and throttled to keep it from over-heating (this is the 600P firmware doing this, not the machine or the BIOS). Overheated flash parts can lead to early failures. However, this was a read test and honestly these devices should not over-temp during a read-test. I should note that in all my tests, the only NVMe SSD that got hotter than the Intel was the Toshiba at 80C, and it didn't flag the critical temperature or throttle (insofar as I can tell). Even though Intel is doing the right thing here, I still can't recommend it due to the enormous number of other performance issues with the device.
