Looks like Intel is launching a new generation SSD in the next few weeks.

It's SATA III 6Gb/s but based on the old 34nm chips.

http://www.fudzilla.com/memory/item/216 ... 00mb/s-r/w

Cool. With speed steadily increasing in SSDs, my question is : for daily use (internet, MS office, little bit of gaming) what is the main advantage of having an SSD that does 450MB versus one that does 150MB ? My applications already open instantly and games are blazing fast to load.

1. Yes, when you are talking about a human interface there isn't that much difference between half a second, a tenth of a second, or a millisecond.

I could still use a faster boot and a faster install.... but realistically it isn't worth spending a lot for it.

Maybe in a few months it will be a good time to get a good buy on a current generation SSD.

2. I actually do have something I would like to speed up on my computer. I think it is caused by my idrive backup. When I used windows explorer to load a file into MS Words, it takes about 4 or 5 seconds. If I launch the file directly with MS Word... no problem. I think that has something todo with what idrive is doing. (if anyone understands this please let me know)

But a faster SSD or fast CPU is unlikely to speed that up.

3. It is hard not to get caught up in the numbers game. They do seem to have hit a bump in the road with the SSDs. The 32nm memory has about 10,000 writes in it. Apparently the new generation can only do 3,000 writes. I think the reason they are sticking with 32nm is a reliability / longevity issue.

It isn't much of an advance being able to have 3x more memory in the same chip is that memory wears out 3x faster. I think over the next 10 years we are going to see reliability issues become more important in all electronics. You can make those circuits only so narrow before you start running into fundamental barriers of physics.

Even now, it wasn't so long ago when circuits were solid nonmoving parts that didn't wear out. They are still solid nonmoving parts... but modern chips and their connectors actually wear out... too many volts for too long at too high a temp... actually alter narrow electrical pathways. It still isn't too bad today... but that is likely the problem with the new generation of memory's limited life span.

20nm, 18nm, 12nm, 8nm... at some point the electrons get too overwhelming for the size of the circuits.

For "regular" use 150MB is fine. But which drive can *reliably* write at 150? It's not the spec which matters.
Up the throughput on the largest drives and you're going to up the throughput on the smaller ones as well. The write performance of many small (and therefore cheap) SSDs is really bad. Like slower than a 2.5'' 5400 rpm drive...
And which drive can you issue a large write to without causing latency to skyrocket?
That's the kind of thing that's going to impact real-world performance.

Here is some more info on the Intel SSDs

The new 25-nm Intel SSDs are slower than the new 34-nm Intel SSDs, but can do more IOPS. 50,000 IOPs to 20,000 IOPS. But it delivers those IOPS at almost half the MB/s speed.

http://www.tomshardware.com/news/Emcres ... 12089.html

Does anyone have a sense of how these two different performance envelops will differ in terms of real life user experience? Why would you favor which one for what use?

These are "up to" specs. Wait for tests if you want to know about actual performance.
I assume there are going to be price differences and that the cheaper line will be preferable once that's taken into account because a larger drive of the cheaper line will end up being faster (depending on what metric you're looking at) than a smaller drive of the more expensive. But that's only a guess.

There is way too much we don't know about these drives but from past models here is my best guess.

The 25nm Gen 3 disks have more of an issue with flash lifetime (less writes per cell before failure) so they are overprovisioning at a higher percentage to offset this limitation of smaller geometry flash cells. You'll never notice the difference in lifetime under normal use but as a side effect overprovisioning increases IOPS capability. Intel will do what they need warranty wise to make sure people are confident in the new drives.

For a single user on a desktop or laptop PC you'll prefer the 510 series with faster transfer speeds.

For a file server that is anything less than overloaded you won't care, either would be fine. You won't be bottlenecked on transfer rates or IOPS so you'll buy on $/GB and reliability not performance.

For a heavily overloaded server it will depend on your usage:

overloaded by number of transactions but not bandwidth and you'll want higher IOPS and will go for the Intel G3
overloaded by bandwidth but not number of transactions and you'll want higher R/W speed and will go for the 510 SSD series or one of the 6Gb/s competitors.

Overloaded by both will give you multiple options to fix by using combinations of drives, different quantities of drives in differing RAID levels, and tons of non drive related fixes (server clustering, RAM cache, programming changes, etcetera).

The biggest unknown question is how will they be priced. I personally don't care about IOPS for my home PC use. I care more about price, reliability, boot time, random read, and random write performance. Any drive that does well on random r/w MB/s will exceed my desire for sequential R/W speeds. And since I'll underformat/overprovision the IOPS capacity will be higher than quoted not that I'll ever push the IOPS limit.

And these two Intel drive series won't live in a vacuum. It'll be

Intel 510
Intel G3
Intel G2
C400 and friends
C300
Samsung 470
Sandforce 2xxx
Sandforce 1200
Indilinx Martini

Although the Sandforce 2xxx drives and some of the Intel drives won't be out as soon as the others apparently. Oh and the order of that list isn't by price or performance or reliability or anything you'd want to sort by. It's just a list.

It'll be a long time before all the Intel G2 drives sell out even if the G3 and 510 drives are all the rage someone will have stock of the old drives.

If I'm reading correct, that second article seems to suggest that the 510 will be launched in place of the G3 (non-SATA 6Gb/s). It would make more sense if the 510 was an X25-E replacement. Either way, at least Intel won't be left in the spec-dust with the G3's 1st-generation-Sandforce levels of performance in the midst a rapidly-increasing numbers game.

The assumption is that 450 MB/s would result in much faster virus scans and other such activities, but as we always predicted there will never be another real-world performance jump like the original HDD to SSD comparison.

In 12 -18 months, these SSDs are going to be saturating SATA 3 connections. They are outpacing the rate at which the connectors are speeding up.

But aren't IOPS and random R/W performance intimately related? I mean, a drive that can deliver 10,000 4K random reads is going to be slower than one than can do 50,000. Or is IOPS some bogus number like MIPS that is just the upper limit of what the controller can handle when doing some trivial operation?

I don't know what the standard for measuring IOPS is but I know it's more important for server/database loads. For PC users, I think the most important number is 4K random read, then write.

If you are looking for synthetic tests, then PCMark Vantage has a quite good HDD suite. Even Anandtech uses it, who hates all the synthetic test suites.

Hmm I'll try a car analogy first as everyone loves those

MB/s is a rate like miles per hour

IOPS is a rate like cars per hour

I want my SSD to boot the OS faster so I want more MB/s

I don't want my PC to handle booting the OS for thousands of users in North America so I don't care about IOPS. Say booting a PC will (totally random numbers don't fixate on the quantity)

read 1TB
write 1MB
all in 30,000 operations over a 15 to 60 second period.

A SSD with 450MB/sec read and 20,000 IOPS would do that faster than a SSD with 250MB/s read and 10,000 IOPS but a SSD with 250MB/sec read and 50,000 IOPS capability wouldn't be faster than either.


now say you want to launch 8 virtual PCs to test a network for a college course or just for the fun of it and now you have a load of

read 8TB
write 8MB
all in 240,000 operations over 15 to 480 second period (depending on if the hardware can keep up)

In this case a infinitely fast MB/s drive would be bottlenecked by transaction processing in those operations if it were rated for anything less than 16,000 IOPS (240,000 divided by 15 secs).

Again these are totally made up numbers for the sake of a theoretical discussion of terminology.

Another comparison would be graphics cards. They have a processor and ram on each card. Some programs are limited by the processor being too slow, some are limited by the ram being too slow. Some are limited by both being too slow.

None of the numbers are meaningless. It just matters what you are doing as to how much they mean to you at that moment.

I'm a little sleepy as I've been up for 18+ hours now but I hope that was coherent enough to help you understand.

I multitask, it's common for me to have dozens of Firefox tabs/windows going + an email client + a game + windows calculator + a text editor and/or HTML editor. So yeah I'm sure if I try hard enough (restoring a prior firefox session, even a normal boot) there might be some split seconds where IOPS would be a bottleneck for me. I'm just saying it isn't likely to be a concern if I understand correctly.

I won't swear that I'm right and someone may link to an article telling us how many IOPS occur in a typical windows boot. But at least we can discuss the concept until then.

Don't know how wise or foolish this is, but I always look for the Vantage Benchmark numbers for the time it takes to boot Windows and the time it takes to launch an application. It seems to me that Intel always comes out on top in these two attributes even when it doesn't do so well on other measures.

dhanson865, does that make any sense to you? In your opinion, am I missing anything by taking such an approach?

Maybe you do, because in real world usage, all the newer drives (Vertex 2, C300) are faster than Intel G2.

I agree more with zsero on this one. I like the faster boot time with a SSD but any SSD beats the snot out of traditional hard drives for boot performance.

I'll pay attention to Vantage scores but I still deal with a lot of XP systems and most reviews have vantage boot times for Vista or Win 7.

If you are buying under 120GB SSDs I think you have to seriously watch the sequential write speeds as well but for larger SSDs I don't look at sequential scores. I'm all about the random R/W when it comes to the 120GB+ drives.

Seems to me there are limited areas of performance that are material to a desktop user for which the HD or SSD is a significant factor:
1. Windows boot and closing times
2. How long it takes to launch an application
3. How long it takes for an application to access a data file
4. And I guess fast random read/write capability speeds up web browsers (is that correct?)

Am I missing any?

Most data files aren't that large. I would guess under 1M, or even under 200K.

If an SSD is good at booting windows and launching applications, isn't it going to be good at the other areas?

I think the most important point, what many people is missing that the least important thing is sequential read speed. For normal usage, it doesn't matter at all.

Look at it, my Samsung F4EG has a sequential read speed at about 130 MB/s. The same as a WD Raptor and the same as most of the SSDs during real world usage. Is there a difference between a 5400 RPM drive, a Raptor and an SSD? Yes there is, by at least 10x. But their sequential read speed is more or less the same.

Is there any situation when you are using the sequential read speed? Maybe the only one is when a single user is copying huge ISO files from one drive to another and nothing else is running in the background. Other than that: not likely.

For Windows XP/2K3/Mac OS X: buy something which doesn't need TRIM. Which means Sandforce or the new Toshiba controller.
For Windows 7: buy something from the last generation, which has a good deal, or what brand you trust: anything Sandforce based or the Crucial C300

0. Linux boot and closing times.

* Windows boot is dominated by random reads

* launching an application will be dominated by random reads if there is any contention for disk availability (multiple programs accessing the drive at the same time). If somehow you could get rid of all the background tasks on a PC and the program is huge it might be that sequential reads would be dominant.

* how long it takes to start accessing the file and how long it takes to finish loading the file are two different things but if the file(s) is/are small the the random aspect will dominate. If the file is huge (Photoshop files, movie editing, etcetera) then sequential speed might dominate.

fast random accesses are common in the vast majority of the tasks you do on a PC. Short of editing large files and 3D game texture loads (level loads, scene changes from indoor to outdoor, crossing a zone boundary in a 3D game, etcetera) it's hard to find common tasks that don't favor random access.

And since sequential reads are fast on even poorly designed SSDs it just isn't much of an issue.

* If an SSD is good at booting windows and launching applications it might be good at other stuff or it might be horrible. These tasks are heavily read oriented. Unless you look at write capabilities you are ignoring a huge portion of the behavior of an SSD by focussing only on boot/load/launch times.

I wrote a long thread on this How important are random writes on SSDs. Probably easier for you to go read that thread than for me to rehash it all again.

I would change that first category to not include Windows XP. If you have a Mac and need the drive to perform well without trim OK that is a special case.

But for windows XP you can use Intel drives, Samsung 470 drives, or even Indilinx based drives and use software utilities to cover the OS lacking native TRIM support. In fact with the Intel utility it can be scheduled and be so automated that you don't even have to think about it after setting it up.

Wow, you really know a lot about SSDs. What I found strange in SSD reviews, is that almost all of them only tests middle-range capacities, usually around 120 GB. But what I found out is that different controllers scale really differently depending on the drive's size, and no one talks about that. For example Intel in the 40 GB version isn't fast at all. On the other hand, I choose the Vertex 2, because I found it to be really well performing even in smaller sizes. I have the 60 GB version which is even handicapped because of the small spare area, but it performs really well, I mean in 10% compared 120 GB version in the reviews. Can you explain a bit on this?

I'd suggest you read from about Nov 2010 in viewtopic.php?f=7&t=56949 to the current post. I've made several posts in that thread about how lower capacity drives have lower write speed.

I'm no fan of the sandforce controller. The Vertex 2 might be faster at low capacities under certain usage patterns but once you test with random data that advantage evaporates. That's not the end of the world if the sandforce controller is cheaper than the other controllers but when they want a price premium for it I look elsewhere.

oh and as to why it happens it's just a matter of physics. Just as 10 carpenters can build a house faster than 5 carpenters a SSD with 10 memory chips can do things faster than a SSD with 5 memory chips.

Value SSDs use cheaper controllers that have less memory channels and/or use less flash chips.

Intel has 10 channels in the X-25M
Intel has 5 channels in the X-25V

Same number of flash chips

Other companies may have a 4 or 6 channel controller but have the ability to address more than one chip per channel. Cutting the number of chips from 8 to 4 or 12 to 6 cuts the speed of a SSD.

Then you get into controller logic (wear leveling, TRIM, bad block reallocation, etcetera). The better controllers handle all the background tasks in a way that allows the stuff you want to happen to be done more quickly.

Anyone heard anything more about the availability of the New Generation Intel 510 "Emcrest" SSDs?

They will be released on March 1st:

http://vr-zone.com/articles/intel-510-s ... 1231.html#

Well, Intel 510 is available now in my country, but the price is prohibitively high for the time being. I don't want to pay a great deal to be an early-adopter/beta-tester given SSD's firmware issues in the past........ I guess I'm better off with my budget Kingston SSD for several months before jumping on the next generation SSD bandwagon.

Intel 510 SSD uses the Marvell 88SS9174-BKK2 controller.

I wonder what this means for the supporting software, one of Intel's strong points?

http://www.storagereview.com/intel_ssd_510_review_250gb

French website performance test of the new Intel 510
http://www.hardware.fr/news/11343/test- ... eries.html

Forget being launched today, the damned things are IN STOCK today.

http://www.overclockers.co.uk/search_re ... arch=crest


Andy

And for the record it's the first Marvell controller (C300/C400 style) SSD that works with the Intel Toolbox. So it's the ultimate SSD for Win XP holdouts.

Who would have thought Intel would use a Marvell controller?