So I'm really confused about the supposed difference in speeds and the performance benefits they incur.

For regular SATA cables I see they are all branded with a 6gb/s transfer speed. I assume that means gigaBYTES per second. SATA Express brings that up to 10gb/s.

YET, if we look at the average SSD, they have read/write speeds of 500mb/s (I assume this is megaBYTE), with the top tier ones being in the 1000mb/s (1gb/s) range.


So what is the point of SATA Express? Surely it's absurd overkill considering SSDs are barely over the 1gb/s threshold? Am I missing something?

"Gb", with a lower-case b, is bits. The speed is Gigabits per second.
An SSD is usually rated on MB, with an upper-case B, which would be bytes. The speed is Megabytes per second.

So for an example, this Samsung EVO SSD runs at 540 MBps (MegaBYTES per second). This is equal to 4.32 Gbps (GigaBITS per second).
In that sense, a 3/Gbps SATA port and cable, would not support the full potential of this drive.

Ah thank you for that answer. That explains it. I now have another question. I'm contemplating purchasing this:
https://www.pccasegear.com/index.php?main_page=product_info&products_id=31656&gclid=CJnh4MWmwcgCFYSjvQods1EAVw&gclsrc=aw.ds

It boasts a 1400mb/s read speed and 600mb/s write speed. But if I look at the speed of the M2 bandwith it is 10gb/s = which translates to 1250mb/s.

So what gives? Why make a device for a M2 connector where you're going to be losing the overhead 200mb/s due to the bandwith limitation?

I hate how some people use bits and some use bytes. Same shit with ISPs, "Oh you get 25mb/sec!" Bullshit, in bytes we're lucky to get 3MB/sec. Everyone should be mandated to use bytes. Why not bits? Because I said so.

In general when you're talking about physical properties resulting in maximum speeds of a medium, or in other words bandwidth, like Internet connections, PCI buses, and of course SATA speeds, the figure is usually measured in bits. (e.g. kbs, gbs, etc.)
When you're talking about the amount of data a device can process, there's no way around measuring in Bytes because thats generally the smallest data unit.
And here comes why: The SATA protocol needs 2 control bits for every Byte of data transferred. That means SATA with a bandwidth of 1000 mbs transfers 1000 MBit/s / (8 Bit/Byte + 2 Control bits per Byte) = 1000/10 MByte/s = 100 MB/s.
That means that SATA III (official name: Serial ATA 6,0 Gbit/s) can transfer up to 600 MB/s (physical units!). And that converted to binary units (which are used in Windows) is 600 MB/s / 1.024^2 MB/MiB = 572,2 MiB/s, which is pretty close to most consumer SSDs. That's why SATA 3.2 (or SATA Express) was developed. Apparently it was easier to replicate the SATA protocol over the PCI Express bus than improving the existing SATA yet again. Btw. there's 2 variants: the 8 Gb/s variant uses 1 PCIe lane, while the 16 Gb/s uses 2.

probably marketing reasons. Also your calculation is wrong. You lose more. See my above calculations as to why.
EDIT: Actually I just realized why: The read figure is maximum read speed which you usually do not reach with real world data. So the SSD has some headroom to transfer that "hard to fetch"-data just as fast as ideal data.

see my reply.

Post has been edited 2 time(s), last time on Oct 14 2015, 6:35 am by NudeRaider.

The main reason I'm purchasing this M.2 SSD is for gaming, and I've got the slot just there waiting to be utilised. When I first got an SSD for Battlefield 4 I noticed an immediate difference in boot time on Windows and in BF4 itself. On the old HDD I would load into the round and people would have already taken all the vehicles and driven halfway across the map. After the SSD I was able to load up and choose my vehicle.

This new SSD will have double the read speed, which, presumably, should be double the loading time, which I will enjoy for games such as the Witcher 3, and upcoming Fallout 4 and Battlefront, Deus Ex, and any future titles going into the future.

Just how exactly does SSD make loading time go faster? I hope I don't need the fast write speed as well. Hell, I don't even know what loading IS. I think it has got something to do with the hard drive moving the game files into RAM? Hell. No idea. If anyone knows please educate me.

I have no estimation on this one, but while read speed is a big factor it's far from the only one, so you can expect to go faster, but not linearly by how much faster the SSD is.
Also sequential read speed (which is the figures you're comparing) is not the only factor to consider when you want to read x number of files. For small files IOPS get more and more important which SSDs are more than a magnitude better than HDDs. This is the reason OS loading time more than halfes on SSDs; OSes load many small files. Games on the other hand have a much larger percentage of reading large files which is governed by sequential read speed, which is only about twice as high as HDDs, which in turn means the advantage of playing games from SSDs is not as big as the advantage for the OS or average programs.

By having higher sequential read, and much higher IOPS.

Only a negligible amount of writes is done during loading.

That's a big part, yes. Another significant part is decompressing textures and starting and initializing the game engine, which requires CPU and/or GPU power. There's probably more that I'm not aware of. But this already shows that loading speed is just as much affected by a fast processor and graphics card as it is by the SSD/HDD.

Thanks.

If you already have an ssd that was made in the last 3 years or so, don't expect any speed boost from buying an M2 drive. It might knock a few seconds off at most. Law of diminishing returns, and also typical benchmarking hype - real-world noticeable results are what matter. HW review sites take an objective scientific approach, but very seldom do they actually take a quantitative approach to what the results mean in practice for day to day computing.

About the only component left that is truly worth scaling up is the GPU, but even then if you're running at 1080p most mid-range cards will do a good job in most games.

Actually I've recently purchased the Acer xb270hu.

It's very difficult if not impossible to find graphics cards, even with SLI, that can run the latest games at max settings on 1440p at 144Hz. So, yeah, graphics cards have got a long way to go before I can really start maxing out this monitor.

If you're still on 1080p @ 60Hz, yeah, you should be fine with today's cards.

Do you mean error detecting bits or control bits?

What that actually means in practice for day to day computing would be a qualitative approach, not quantitative.

A lot of it is transferring data from HDD to RAM (some processing, as you said) and to video RAM.

Actually they are used to make sure you have no DC offset caused by transferring more 1's than 0's:
https://en.wikipedia.org/wiki/8b/10b_encoding
This is important to keep the clock in sync and not "charge" the HDD during data transfer.

Correct, I meant to say qualitative but simply used the wrong word. Was typing on my phone so wasn't paying a huge amount of attention.

Just to throw your mind for a loop:

Intel 3D Xpoint is a new technology that looks to be ready to fill into DDR4 slots and for use in PCIe slots. It's as fast as RAM (much faster than nand flash) which means that it could theoretically replace RAM altogether, meaning your computer will turn on virtually instantly, and all of your files/programs will be loaded, since they're already there in the first place.

http://www.intel.com/content/www/us/en/architecture-and-technology/3d-xpoint-unveiled-video.html

This is a pretty cheesy description of how memory works and how Intel 3D Xpoint doesn't work, but what it will do. I still have no clue how the stuff works, which is disappointing considering it should be made fully public, but Intel isn't going to be lying about this stuff. If anyone else did it, I wouldn't believe it.

"Cheaper than DRAM and faster than NAND" is a very deceptive claim. In other words, it's not faster than DRAM, and it's not cheaper than NAND. DRAM is too expensive for a storage solution, and XPoint may very well be as well, even if slightly more affordable. XPoint is claimed to be 1000x faster than NAND, which is still an order of magnitude slower than DRAM. I was hyped for this technology until I realized what they were describing is actually something in an awkward position of not serving any meaningful purpose for consumers.

If costs come down, it may eventually replace SSDs due to the drastic speed improvements, but right now the implication that it is an evolutionary stage for merging volatile and non-volatile memory seems to be deceptive marketing.

According to this article XPoint, like their own product, will be around the same as DRAM speeds but cheaper. From that perspective I'm not as pessimistic as Roy about its usability.

We might see this replace both SSDs and RAM to possibly make office PCs fast and cheap. In any case it should shave off a good part of smart phone costs or of embedded devices.
Today RAM never bottlenecks your PC, so even if the new tech is a bit slower it might still be fast enough. If it suffices for gaming though is another question.
It might also change the way programs are stored on the "hard drive" in a ready-to-use fashion rather than in a read-to-load fashion. So this might have implications on software development as well.

Cool stuff.

Until they release some actual hardware or at least some technical papers, it's just marketing buzz. If anyone can meet these claims, its Intel. But right now they are still just claims.

I do keep hearing contradictory reports about VRAM stacking from this new Intel stuff and DirectX12.

Some say it's coming with DX12, others say its impossible to have VRAM stacking, so I'm not sure who to believe. Obviously if such a thing were possible it would make an SLI setup a very lucrative option. I just find it hard to believe that some software miracle can somehow make hardware start doing crazy shit like that.

HBM already exists; it's not a rumor.