Pci-e 2.0 vs 3.0 vs sata -ahci vs nvme

I would really love some discussion and serious input on these matters.
I am getting increasingly frustrated reading reviews and comments in general where very poor logic is used. For example, Some hard facts on the random iops limitations of sata 3 and pcie 2.0 and 3.0 ? Anyone? It seems a lot of the time you hear logic like "if the pcie 3.0 drive has a sequential read of 1500MB/s, then running it on the pcie 2.0 interface everything must be half the speed, half the performance" Right? Clearly this reasoning is not good enough. For a general halving of performance in my scenario, clearly the ssd would have to be in the process of saturating the pcie 3.0 interface before you could see a dip down to half speed/performance of pcie 2.0 with the same drive. At least for quite a few benchmarks. Taking into account higher efficiency of newest pcie interface, still it doesn't add up.
So I really want to know. How far can you push the pcie 2.0 interface ?(4x the bandwidth of sata 3 if you think x4 lanes) What can nvme bring to the table, in numbers? People are getting tired of sequentials, sequentials, sequentials. What's in it for us with small size random and latency? I don't really see most new m.2 pcie ssd's saturating anything near the pcie 2.0 x 4 lanes. Or am I wrong ? 4 x the raw bandwidth of sata 3 + nvme beating out ahci, and still we can't beat max 100 000 iops of sata/ahci unless going to pcie 3.0 ?

 

The information is there if you look for it. Details matter.

As for specific performance improvements of one type of connection to another; it has always been the case that maximum/theoretical speeds of a connection type does not define the specifications of the component in question - except for those maximum values. There is no guarantee it will reach them either.

Not all components push the limits of the interface they connect to. Not even the cutting edge versions we might assume should (the current version of the Samsung 951 is an example... need to wait for the full NVMe implementation of that model to actually see an improvement in real world usage).

Even the best NVMe SSD currently offered (Intel 750 1.2TB) does not blow me away on sheer 'scores'. Software, operating systems and possibly even file systems at the kernel level will need to be re-written to actually take advantage of the benefits NVMe offers. Make no mistake, in a productivity setting, the Intel 750 1.2TB SSD will blow away anything I have right now - but you're asking about 'numbers' and no, I'm not impressed there.

Sequential speeds matter and Random 4K R/W is also important. But actual productivity is much more than '4 corner' testing as has been pounded into everyone's head for so long now. Testing real world workloads is what the old 'scores' will morph into. Whether manufacturers like it or not.

See:
http://forum.notebookreview.com/thr...the-steady-70-30-read-write-mix-tests.774360/


Worrying about 'numbers' and why they're reachable or not is what most manufacturers and most reviewers want us to worry about today. Easy to run a 20 second benchmark and give us a 'score'. Or test the drive with synthetic data and say 'look! It hit's the interfaces' maximum'. That is not what performance is about.

Real performance is about getting things done faster. Not booting up an O/S (boring; 5 - 10 seconds now on almost any SSD). Not launching programs (same thing; any SSD is within a second or two of each other). But actually completing a task faster. Like Windows Updates, for example. Or, my workloads (for me). In other words, things that are repetitive, time consuming and maybe even critical to how much one earns each day.

Both the booting and the launching of programs is not something that is done enough times each day to be a meaningful aspect of 'performance'. Even HDD are still good enough for that.

Hitting those high IOPS means having a workload with a constantly and consistently high queue depth. This is not a workstation class type of workload. And therefore, those 'numbers' are meaningless from the word go.

 

I am looking at details, believe you me Thank you for the response! I agree with most of what you are saying, but people have different workloads and different appreciation of latency, speed and time. That extra second bothers me. Latency is important and can be noticed. Running several different Vm's with a lot of taxing apps, servers and databases, IS different from browsing the internet and watching movies. A lot of small reads and writes CAN bring down sata 3. And like you say, how fast tasks finish. Pcie 2.0x4 and nvme in itself could improve a lot on sata 3/ahci. So far I am disappointed in the presentation of the m.2 pcie products in particular. Though reading your linked article, there is still no way to tell. I guess that's why I'm trying to force the issue...