SSDs slower than RAM

[life24]

Hi,
The following statement is true?

The SSD is still thousands of times slower than RAM.

 

[ctk1981]

Modern ram is faster, don't know for sure if its on a magnitude of thousands.

 

[Ancalagon44]

Yes it is true.

 

[life24]

Yes it is true.

What is measured? How to calculation?

 

[Ancalagon44]

Maybe not thousands of times faster, but considerably faster.

Bandwidth and access time (latency) are the two metrics measured. Bandwidth on a fast SSD will be limited by the SATA3 bus, and tops out at around 550mb/s. Looking at this page, bandwidth for DDR3 can be as high as 17GB/s. So that is over 30 times faster. Not thousands of times faster, but considerably faster.

Latency for RAM is measured in nanoseconds, while it is measured in milliseconds (I think) for SSDs.

 

[BarkingGhostar]

Why would the fact that SSD's were slower than RAM be surprising? I would think if a) SSD's were as fast as RAM then the RAM market would be demolished or the SSD market way over priced, or b) if RAm was as slow as SSDs then someone would have come up with a way to physically implement an SSD as a RAM module.

Does anyone have a serious problem with RAM being faster than an SSD?

 

[Topweasel]

Modern ram is faster, don't know for sure if its on a magnitude of thousands.

DDR3 1600 is 10,500 MB/s

The average current gen SSD is 530MBs.

So about 1/20 the performance of memory.

 

[Deders]

DDR3 1600 is 10,500 MB/s

The average current gen SSD is 530MBs.

So about 1/20 the performance of memory.

Some ram is faster than that. Mine running at 1800MHz does 21,000 MB/s, and more recent memory controllers will drive it faster.

 

[Essence_of_War]

The data rates are different, but latency is important too.

http://blog.codinghorror.com/the-infinite-space-between-words/

1 CPU cycle ~ 0.3 ns -> 1 s
Main memory access ~ 120 ns -> 6 min
Solid-state disk I/O ~ 50-150 μs -> 2-6 days

Table originally from:
http://www.amazon.com/dp/0133390098/?tag=codihorr-20

 

[Topweasel]

Some ram is faster than that. Mine running at 1800MHz does 21,000 MB/s, and more recent memory controllers will drive it faster.

Nope. That's the fastest a DDR3 1600 memory chip can communicate. Yours is capped around 15GBs.

If your getting only 21GBs. Its probably because you are running with dual channel configured on an inefficient memory controller. That or you have it overclocked quite a bit.

http://en.wikipedia.org/wiki/DDR3_SDRAM#JEDEC_standard_modules

But my numbers is based on actual memory capabilities and SSD capabilities to establish a 1 on 1 comparison and wanted to keep it as general as possible.

 

[Deders]

Yeah but most people have dual channel nowadays

 

[Topweasel]

Yeah but most people have dual channel nowadays

And most people have Raid controllers on board.

I was assessing 1 Memory stick vs. 1 SSD. SSD's have endless amount of raiding that can be done on single computer. The best you can get out of RAM is about 16 channels over 4 CPU with UMA.

But 1 average SSD is 1/20 1 average memory stick.

 

[Deders]

And I was saying that generally ram bandwidth in machines setup in ways that people generally use make the divide even bigger

 

[Topweasel]

And I was saying that generally ram bandwidth in machines setup in ways that people generally use make the divide even bigger

I understand where you were coming from. But I don't think that's a fair assesment. Like I said if there wasn't such a cavernous difference and it actually impacted performance then people would just buy two SSD's and stripe them. An SSD is ~1/20 an average stick of ram. No reason to add complexity of dual channel memory vs. Raid performance into the mix. If we wanted to go overboard we could compare it to my X79 setup or someone with an X58 setup.

 

[Fred B]

With IBM eXFlash and memory channel storage it is putting ssd on to the ddr3 controller with very low latency compared to pci and sata .

So i think with new development around 3D nand it will be a matter of time and flash is faster than ram . But the interface got to be faster first , most important is support for the interface on all hardware to be successful .

 

[Topweasel]

With IBM eXFlash and memory channel storage it is putting ssd on to the ddr3 controller with very low latency compared to pci and sata .

So i think with new development around 3D nand it will be a matter of time and flash is faster than ram . But the interface got to be faster first , most important is support for the interface on all hardware to be successful .

I wouldn't count on it. I don't see how Non-volatile memory can ever pass volatile memory if given the same development funds. The fact that one has to deform its structure when writing data should mean it's always at a disadvantage.

Obviously if your looking for different things like recording a Petabyte a second from a supercomputer and money is no object a company like IBM will come up with a solution.

 

[h9826790]

If you setup a RAM driver, you will realise that the RAM driver handle files about 5-10 times faster than the SSD (for both sequential and random read write)

 

[Makaveli]

Some ram is faster than that. Mine running at 1800MHz does 21,000 MB/s, and more recent memory controllers will drive it faster.

Just looking at these numbers something is off.

going from DDR3 1600 to DDR3 @ 1800 isn't going to double your bandwidth from 10,500 to 21,000!

 

[Topweasel]

Just looking at these numbers something is off.

going from DDR3 1600 to DDR3 @ 1800 isn't going to double your bandwidth from 10,500 to 21,000!

He was getting his from a bandwidth benchmark on the computer, so he was seeing the dual channel memory results and not the transfer speed of a single stick.

 

[Makaveli]

He was getting his from a bandwidth benchmark on the computer, so he was seeing the dual channel memory results and not the transfer speed of a single stick.

lol makes sense cause when I first saw that was like wtf.

 

[Binky]

Hi,
The SSD is still thousands of times slower than RAM.

This cannot be true, mathematically. It's also one of my pet peeves.

Something that is "1x" or "100%" lower/slower/less than anything is zero. More than 1x less would imply nothing really, although I guess it could imply a negative number, but that's nonsensical for most things. Sure, something can easily be 1,000,000,000x more/higher/greater/faster, but the reverse does not work.

I see this crap all the time in advertising, and people probably pick it up from there.

On point, RAM is significantly faster, but not by that (poorly phrased) magnitude.

 

[Lonyo]

He was getting his from a bandwidth benchmark on the computer, so he was seeing the dual channel memory results and not the transfer speed of a single stick.

On that basis you should compare a 64GB or 128GB SSD to a single stick of RAM, because SSDs are already effectively in multi-channel mode, and usually the major difference is 128GB vs 256GB due to the number of NAND chips used, hence why a ~128GB SSD performs substantially slower than a ~256GB one for some operations.

Comparing typical to typical makes most sense. Typically computers have dual channel RAM, and SSDs typically have enough channels to saturate SATA-6GBps, therefore typically RAM is over 30x better in peak throughput (sequential SSD reads), probably 60+x sequential write speed, and has way better latency.

Otherwise you're arbitrarily deciding what numbers to compared based on a single stick of RAM vs some random choice of SSD which may not be particularly representative and is not specified.

http://www.anandtech.com/show/8066/crucial-mx100-256gb-512gb-review
128GB - 150MB/s sequential write
256GB - 330MB/s sequential write
512GB - 500MB/s sequential write

 

[Topweasel]

On that basis you should compare a 64GB or 128GB SSD to a single stick of RAM, because SSDs are already effectively in multi-channel mode, and usually the major difference is 128GB vs 256GB due to the number of NAND chips used, hence why a ~128GB SSD performs substantially slower than a ~256GB one for some operations.

Comparing typical to typical makes most sense. Typically computers have dual channel RAM, and SSDs typically have enough channels to saturate SATA-6GBps, therefore typically RAM is over 30x better in peak throughput (sequential SSD reads), probably 60+x sequential write speed, and has way better latency.

Otherwise you're arbitrarily deciding what numbers to compared based on a single stick of RAM vs some random choice of SSD which may not be particularly representative and is not specified.

http://www.anandtech.com/show/8066/crucial-mx100-256gb-512gb-review
128GB - 150MB/s sequential write
256GB - 330MB/s sequential write
512GB - 500MB/s sequential write

We could go on and on about it. Then why don't we look at EDO Simms instead of DDR DIMMS.

I was analyzing 1 average stick performance against 1 average hard drive to show the discrepancy. I didn't want to get into some controller vs ram sticks war.

Also I think you would be surprised in the amount of single channel users still out there.

 

[Deders]

Just looking at these numbers something is off.

going from DDR3 1600 to DDR3 @ 1800 isn't going to double your bandwidth from 10,500 to 21,000!

He was getting his from a bandwidth benchmark on the computer, so he was seeing the dual channel memory results and not the transfer speed of a single stick.

It can depend on the benchmark too, Sisoft Sandra tells me it's 21,000, wheras Aida tels me it's 25,000, so I guess there is a difference in what kind of data the memory is dealing with.

 

[KillerBee]

Thought this was interesting video someone made:

loading times between a conventional big HD,
a highend SSD and a dedicated 25GB Ramdrive for BF4
https://www.youtube.com/watch?v=KzAtiDGHEjM

On paper it seems ram would be a lot faster for program loading
but in reality SSD is only a lil slower

ps: here's overclock.net thread about the ImDisk ramdisk he used (free)
which can also be used to mount .iso files

http://www.overclock.net/t/1086220/imdisk-open-source-ram-drive-with-no-size-limitations