Random Performance – Iometer (IOPS/latency), YAPT (random)
We are trying something different here. Folks tend to not like to click through pages and pages of benchmarks, so I'm going to weed out those that show little to no delta across different units (PCMark). I'm also going to group results performance trait tested. Here are the random access results:
Iometer:
Iometer is an I/O subsystem measurement and characterization tool for single and clustered systems. It was originally developed by the Intel Corporation and announced at the Intel Developers Forum (IDF) on February 17, 1998 – since then it got wide spread within the industry. Intel later discontinued work on Iometer and passed it onto the Open Source Development Lab (OSDL). In November 2001, code was dropped on SourceForge.net. Since the relaunch in February 2003, the project is driven by an international group of individuals who are continuesly improving, porting and extend the product.
Iometer – IOPS
The cap at ~200k in this test is actually a limit of our Iometer configuration (pegging a CPU core). Back when we tested the SSD 750, I figured that would be the only SSD to reach that limit on a single IOPS-driving thread. I was mistaken. While it doesn't beat the SSD 750 at any single metric, the NVMe SM951 is certainly nipping at the 750's heels. Another interesting point of note was that the AHCI SM951 actually beat out its NVMe counterpart in a few tests at very low queue depths. This may have been partially due to there being more flash dies available (the AHCI part was 2x the capacity), combined with a higher write speed rating on the AHCI part. Our Web Server test is pure reads, and we see both variants track more as we might expect (NVMe > AHCI). This is more apparent and clearly visible in the latency results below.
Iometer – Average Transaction Time
For SSD reviews, HDD results are removed here as they throw the scale too far to tell any meaningful difference in the results. Queue depth has been reduced to 8 to further clarify the results (especially as typical consumer workloads rarely exceed QD=8). Some notes for interpreting results:
- Times measured at QD=1 can double as a value of seek time (in HDD terms, that is).
- A 'flatter' line means that drive will scale better and ramp up its IOPS when hit with multiple requests simultaneously, especially if that line falls lower than competing units.
Some great low latencies seen here. Not as low as the SSD 750, but still great for such a small form factor product.
YAPT (random)
YAPT (yet another performance test) is a benchmark recommended by a pair of drive manufacturers and was incredibly difficult to locate as it hasn't been updated or used in quite some time. That doesn't make it irrelevant by any means though, as the benchmark is quite useful. It creates a test file of about 100 MB in size and runs both random and sequential read and write tests with it while changing the data I/O size in the process. The misaligned nature of this test exposes the read-modify-write performance of SSDs and Advanced Format HDDs.
This test has no regard for 4k alignment, and it brings many SSDs to their knees rather quickly. The SSD 750 is heavily optimized for 4k aligned writes, which explains those inconsistent results. The SM951's don't seem to care in the least about that little issue, and turn in results that are more in line with their maximum sequential write ratings. Note that the NVMe SM951 is rated at a lower write speed than the AHCI version.
Hey Allyn,
I approach most of
Hey Allyn,
I approach most of this from a gamer’s perspective. Of course, I wouldn’t expect night and day difference with a very high end SSD vs a normal SSD. But what about for something like running around Skyrim with a butt-ton of mods? Open world games, tons of things to load on the fly.
It could still be CPU bottlenecked though. Tom’s Hardware did an article a long time ago about SSD load in gaming. The guy used some sort of trace-based analysis tool from Intel to check if the reads from the SSD during game startup, level loading, and playtime are sequential or random, what size, and what queue depth. It’s very interesting and I think many gamers would like to see such an article.
I’m looking at all the graphs and frankly it doesn’t mean much to me. I don’t run file servers, I load a ton of maps.
Thanks
Great review. I was hoping
Great review. I was hoping you could help out by comparing my workflow to which above benchmark best applies to me.
My apps use up to 29GBs of RAM where 1000s of 64k buffers are used as targets for various streams of audio stored on SSDs.
When I press a key on an 88 note keyboard/synth it goes 1st to the 64k buffer in RAM then a stream of audio follows.
Obviuosly random applies to the 64k RAM buffers and read to the streaming audio files.
Maybe the Workstation benchmark….?
Thanks again for a great source of comparisons on SSDs.
The X99 Sabertooth allow one
The X99 Sabertooth allow one to conceal their M.2 SSD completely under the Thermal Armor. Is that recommended given the heat output for SM951 NVMe?
HAPPY HAPPY JOY JOY!!!
HAPPY HAPPY JOY JOY!!!
Have you heard about a 1TB
Have you heard about a 1TB version of the SM951 being release soon?
I’m using the SM951 ACHI in a
I’m using the SM951 ACHI in a m.2 to PCI x4 card that has a heatsink on it and is plugged into a x4 PCI-E slot on my x99 motherboard. I love it but I’m wondering if I managed to get my hands on one of the new NVME, would it fit into the same heatsink slot (the pins looks the same) or would I have to use the M.2 slot on the motherboard? I’m assuming either would work and just change to NVME in the BIOS.
Second, I have two Samsung 850 pros running in RAID 0 as my applications drive. Would I still be able to keep this in RAID while using NVME on the Asus x99 motherboard?
Hello, I’ve just been
Hello, I’ve just been comparing a 512GB 951 NVMe variant that I purchased yesterday with an existing 512GB 951 AHCI. Apparently it’s a sample rather than a production unit but I’m seeing fantastic read speeds but horrific write speeds. In my case I’m using with an Asus Z97i-plus with the latest BIOS. The board identifies the 951 and allows me to install windows (8.1 all latest updates)… so far so good. Unfortunately when I run speed tests against the NVMe variant I get 10 times slower write speeds compared to the AHCI 951.
CrystalDiskMark: AHCI variant (connected to PCIe 3.0 bus)
Seq Q32T1 – 1172MB/s read | 1043MB/s write
4k Q32T1 – 398MB/s read | 289MB/s write
Seq – 1052MB/s read | 900MB/s write
4k – 35MB/s read | 128MB/s write
CrystalDiskMark: NVMe variant (connected to PCIe 3.0 bus)
Seq Q32T1 – 2264MB/s read | 501MB/s write
4k Q32T1 – 563 MB/s read | 21 MB/s write
Seq – 1299 MB/s read | 170 MB/s write
4k – 54 MB/s read | 0.98 MB/s write
Blindingly fast read but horrifically slow write speeds.
I’ve also tested using the Z97i-plus’s M.2 slot. I see reduced read speeds due to the limited, 10Gbps, speed of the M.2 on this board but the same horrific write speeds.
Is there something that I might be doing wrong? Could this be a BIOS problem? A Windows NVMe driver problem?
That’s odd, but I believe
That's odd, but I believe Kristian from Anandtech had a similar issue with one of his samples. It was an actual defect I believe and they had to swap out his sample, IIRC.