NVMe vs AHCI

[DWN]

I was excited. Then confused.

 

No real world benefit other than during file copies?

 

List here. Twice the cost for no real world benefit?

 

[AlwaysFSX]

No real benefit? It's several times faster than AHCI SSDs, you will notice just how faster everything is.

[Mattrixx]

I was excited. Then confused.

 

No real world benefit other than during file copies?

 

List here. Twice the cost for no real world benefit?

Much faster, and optimized for the fast speeds and high IOPS of SSDs.

[Captain_WD]

I was excited. Then confused.

 

No real world benefit other than during file copies?

 

List here. Twice the cost for no real world benefit?

 

Hey DWN,

 

New NVMe drives work much faster than the SATA ones. There is a significant speed increase, The PCIe type of connection has a much higher speed limit compared to SATA3 (6Gb/s) one. This is the most basic thing you can compare between SATA SSDs and PCIe SSDs. Everything else only adds on to the benefits that the PCIe has.

 

Captain_WD.

[DirkW]

Well, it depends (as always)...

 

things that load pretty much instantly with a SATA-AHCI-SSD already, will not be improved in any meaningful way. Things you need to wait for, may improve quite a bit.

[DWN]

More confused. Did any of you read the article graphs? Load time for games, Word and Excel, Photoshop, AV scans, etc etc etc. The only significant benefit appears to be file copies.

 

http://www.techspot.com/review/984-intel-ssd-750-series/

 

DirkW has the right understanding, Drive speed is at the point of diminishing returns for 98% of common operations. Things that already seem to load fast, won't load any faster. With a 500MBs SSD everything is already about as fast as it gets. But things that you have to wait for are waiting on other components like network or CPU and so you don't see any direct concrete sizable benefit. Maybe down the road things will take advantage of the increased bandwidth but unless you are using one of the 2% of operations, currently no normal, consumer oriented, regular occurring (daily) PC operation significantly benefits from a NVMe drive connected at full bandwidth over a 500MBs AHCI SSD.  

 

Standard disclaimers apply to video encoders, 3D rendering artists that benefit from everything.  98% of people will see little to no benefit, currently.

 

Why can't video cards see a quadruple increase in performance? With all the pixels we need to push for gaming, that would be more useful.

[Chiobe]

98% of people will see little to no benefit, currently.

98% of the people will have problem maxing out a SSD, let alone this monster.

 

This is a drive, meant for content creators and others who need fast read speeds. Like if you want to do multicam on 4k video.

The write speed is nice, but unless you are writing internally on the disk or too a 2nd one, so will you not get the full write speed.

[Captain_WD]

More confused. Did any of you read the article graphs? Load time for games, Word and Excel, Photoshop, AV scans, etc etc etc. The only significant benefit appears to be file copies.

 

http://www.techspot.com/review/984-intel-ssd-750-series/

 

DirkW has the right understanding, Drive speed is at the point of diminishing returns for 98% of common operations. Things that already seem to load fast, won't load any faster. With a 500MBs SSD everything is already about as fast as it gets. But things that you have to wait for are waiting on other components like network or CPU and so you don't see any direct concrete sizable benefit. Maybe down the road things will take advantage of the increased bandwidth but unless you are using one of the 2% of operations, currently no normal, consumer oriented, regular occurring (daily) PC operation significantly benefits from a NVMe drive connected at full bandwidth over a 500MBs AHCI SSD.

 

Standard disclaimers apply to video encoders, 3D rendering artists that benefit from everything.  98% of people will see little to no benefit, currently.

 

Why can't video cards see a quadruple increase in performance? With all the pixels we need to push for gaming, that would be more useful.

 

Well the improvement and the benefit is already there, it's another matter if people actually need it. Shortening loading times in half when they are 30 seconds is one thing, shortening them in half when they are 5 seconds is another. The improvement and the benefit is there (again cutting in half or better), it's a whole other matter if it's actually needed by the individual. Many people still use simple HDDs for their work and home computers because still many people don't care if their computer loads for under 10 seconds or takes a full minute, as long as they can chat, perform simple office tasks, watch a movie and send an e-mail, but still SSDs became very popular. 

I wouldn't say that other components aren't advancing as fast. 60 fps on 1080p wasn't as easy and inexpensive 4-5 years ago as it is today. You can build a 4K gaming PC for a lot cheaper compared to a few years ago. 

 

PCIe SSDs do have their implementations as well as old 5,400 SATA HDDs. It just depends on what each person really needs or wants.  

 

Captain_WD.

[asquirrel]

I was excited. Then confused.

 

No real world benefit other than during file copies?

 

List here. Twice the cost for no real world benefit?

So here's the rub: These are first generation NVMe drives. You know how bad the original dual core processors were when they first came out? Games had glitches on them, OSs crashed, drivers panicked...it was a mess. Some of those problems were in the dual core silicon itself, and got hammered out with later processor generations, the rest of the issues were with software which thought it handled multi-threading correctly and discovered it didn't.

 

The problem NVMe is hitting with real world performance has to do with how files are read off the disk. Games, word processors...almost all the software you use on a daily basis uses this very, very simple loop to read a file off the disk:

 

read(chunksize) ==> (program waits for 'chunksize' of data from storage) ==> (function call returns, providing 'chunksize of data', program now does something with that) ==> read(chunksize).

 

So...the issue is that the next read cannot be started until after the previous read has finished. If you need to read in a large file (maps or textures for a video game, stored in a container file like an MPQ) then you have to wait *forever* for those single-threaded, synchronous reads to complete one after another. In this case, NVMe offers almost no benefit because, while the protocol is more efficient and does go faster, the vast majority of NVMe's performance benefit comes from parallelization of reads and writes, not making sequential, synchronous reads and writes go faster.

 

So, in the end, it's a software developer's problem. The software needs to be rewritten to take advantage of these drives by doing parallel reads *or* requesting large files as single massive I/O requests that the drive can then chop up into multiple smaller requests and fulfill in parallel. This isn't something the OS or drive manufacturers can really get around for you...sometimes those read calls really do need to occur in sequential order for whatever reason. The drive can't assume it's safe to automatically parallelize reads. Writes, however, are almost always safe to parallelize, that's why any write-intensive task to NVMe drives consistently beat the AHCI drives, but read tasks do not.

 

Let me know if you have more questions. I tried to keep this, rather complex topic, simple.