It's not the sequential speeds that matter, but the small random IOs and latencies. HDs can barely provide 1MB/s 4KB random read/write speeds, whereas even the SSD 840 offers up to 250MB/s (depends on the IO size and queue depth, but even in the worst case scenarios SSD 840 should be tens of times faster). Combine that with much lower latency (HD latency is measured in milliseconds, whereas SSD latency is in microseconds) and the overall performance is much, much better.
Yes, reading reviews of SSDs, especially on anandtech.com, taught me that small file-size, random performance is the bottleneck for ordinary, everyday use scenarios, not the sequential performance.
But is is the sequential performance measured by the Blackmagic tool that are usually posted on this forum. So to mix things up, below is the report of
4K random writes, followed by 4K random reads, from
digilloyd's Disk Tester "IOPS random I/O operations" test for my 840 Pro 512 Gb, about a month old, with latest firmware, with an
additional 10 Gb or so of over-provisioning and connected by SATA III in an early-2011 Macbook Pro. Note that the average throughput is 10 Mb/sec write, 8.5 Mb/sec read, in contrast to the ~500 Mb/sec sequential numbers that show up in the Blackmagic tool.
Why do the write speeds fall? Is it because of the on-disk cache? The report says the "File system unified buffer cache" is disabled, but I am not sure whether that includes the RAM cache located in the SSD itself.
I'd love to see some other participants post their results for this or similar tests. Hellhammer, do these results look about right to you?
DiskTester 2.2 64-bit, diglloydTools 2.2.0, 2012-12-15 18:05
OS X 10.6.8, 8 CPU cores, 8192MB memory
Friday, March 1, 2013 2:02:23 PM MT
disktester iops FM_Boot
Capacity GB: 466.07
Free space GB: 261.29
File system signature and id : H+ / HFS Plus
Simultaneous I/O threads: 1
Transfer size per I/O: 4K
Number of transfers (per thread): 131072
Aggregate I/O size: 512MB
File system unified buffer cache: DISABLED (default)
Test type: RANDOM
Code:
Writing ...
MB Write MB/sec Write ops/sec
65 32.8 8404.6
97 24.4 6234.0
123 20.6 5274.7
146 18.3 4681.6
166 16.7 4276.3
186 15.5 3979.2
205 14.7 3750.5
223 14.0 3571.4
240 13.4 3424.0
257 12.9 3299.9
274 12.5 3195.6
290 12.1 3104.5
307 11.8 3024.4
323 11.5 2954.5
339 11.3 2895.4
355 11.1 2841.0
371 10.9 2795.4
387 10.8 2752.9
403 10.6 2715.6
418 10.5 2680.7
434 10.4 2651.0
450 10.2 2622.6
467 10.2 2600.5
483 10.1 2581.4
500 10.0 2563.5
512 10.0 2552.6
Reading ...
MB Read MB/sec Read ops/sec
16 8.5 2169.6
33 8.4 2147.8
50 8.4 2158.2
67 8.4 2158.9
84 8.4 2157.7
101 8.5 2163.8
118 8.5 2166.6
135 8.5 2168.6
152 8.5 2169.6
169 8.5 2167.3
186 8.5 2167.0
202 8.5 2166.0
219 8.5 2166.3
236 8.5 2166.5
253 8.5 2167.0
270 8.5 2168.8
287 8.5 2168.7
304 8.5 2166.7
321 8.5 2166.0
338 8.5 2165.9
354 8.5 2165.2
371 8.5 2166.3
388 8.5 2165.4
405 8.5 2165.2
422 8.5 2164.4
439 8.5 2164.7
456 8.5 2165.6
473 8.5 2165.7
490 8.5 2165.8
507 8.5 2166.4
512 8.5 2165.7
Write 512MB in seconds: 51.3
Read 512MB in seconds: 60.5
Aggregate write rate MB/sec: 10.0
Aggregate read rate MB/sec: 8.5
Write ops/sec: 2553
Read ops/sec: 2166
Command "iops" executed in 114.07 seconds on Friday, March 1, 2013 2:04:17 PM MT