Intel is back in the controller business

Proprietary controller debuts in Intel DC S3700 server SSD - The Tech Report

Nice to see Intel making their own controllers again. Maybe they'll replace the SF-2281 controllers with this one in the next generation of consumer-class SSDs? Hopefully?

 

Maybe using the SF-2281 controller started them on the wrong foot? OCZ and others ruined it for Intel previously and this painted a bad picture for the SF-2281 controller (people didn't trust them or refused to buy them altogether). Going to an "in-house" controller might give better resuls for Intel's SSD production and sales (lower production cost, better end-user results and better sales, etc.).

Samsung has a very competitive SSD line, and has gained a lo of momentum. They've done their SSDs in-house all along as far as I can remember. (correct me if I am wrong) Then there are other companies such as Corsair, Crucial, Etc. I want to see more affordable 500GB and 1TB+ SSDs for consumers around $300-500, less being better of course.

 

I am really interested in this 'more consistent' SSD.

I've always said that HDD's were more consistent for me than SSD's ever were (even with leaving up to 70% of the drive 'unallocated'.

Now, with the charts Anand shows, I have that proof. Actually, with Win8 I have that proof too: seeing average response time hit up to 4 SECONDS is pretty ridiculous - no matter how fast it is 'most of the time'.

This is what has caused me to resist moving to SSD's for so long (since 2009) - their performance would simply sink to depths unheard of in the storage subsystems we were used to (multiple, fast, optimized HDD's).

I can hardly wait for a 'consumer' drive that would fit the notebook's characteristics (heat/power) better from Intel.

Yeah, Intel never left the game - they keep redefining it.

 

Intel had no choice but to go with a 3rd-party controller because they were being left behind in the performance race. They had the SATA II Intel 320 (which costs more per GB than the Intel 520) but if they wanted to remain competitive, they needed a stopgap solution until they develop an in-house controller with performance comparable with other SATA III SSDs. Even then, it still took a year of extensive validation before they were able to release their own SandForce SSD.

Glad to see Intel's foray back into the SSD controller world. Hopefully, we'll see their new controller on high-end desktop models with MLC NAND at a lower price point. Granted, it's good to see that the pricing for the HET MLC model is very reasonable.

 

Intel's performance crown (SSD's) has never been disputed: what the SF 'solution' allowed them to do was compete in the consumer side of the market.

If you want(ed) the fastest most reliable SSD you could buy; it has always been Intel for the most bang for the buck.

 

Most reliable, yes. Fastest, that's true back when there were only SATA II drives out. Unfortunately, development on their new controller was slow going so they had to use controllers from Marvell (Intel 510) and SandForce (Intel 330 & 520).

 

Sorry, I should have said:
"If you want(ed) the fastest most reliable Enterprise SSD you could buy; it has always been Intel for the most bang for the buck."

Instead of:
"If you want(ed) the fastest most reliable SSD you could buy; it has always been Intel for the most bang for the buck."


That is what I meant (and is the post you're responding to).

 

But does enterprise even care about bang for the buck? When an Oracle license is something like $80k per CPU (core? Dunno if they ever changed their licensing model from per-core to per-socket...), the price difference between a $1k Intel SSD and a $10k Fusion-io SSD is peanuts.

 

There are different levels of enterprise - and yeah they all do care for bang for the buck.

 

Perhaps not purely bang for the buck but TCO.

 

Well, after just reading the (google cached) AnandTech review of the new Intel DC S3700 200GB SSD, I now have proof that SSD are inconsistent:

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review

I've been saying this for a while (~2009) - SSD's feel slower than a good HDD setup - just called crazy... whatever.

Yeah, I have to agree 100% with the last quote here: this is an exciting development and Intel is on the right track (as usual) with the direction it's taking (consistency of performance).

Actually, this is why I haven't seen a Samsung SSD I've enjoyed in my user scenarios: they feel just like an HDD to me (at SSD prices) - in other words; I have felt that inconsistency and that is why I have equated them as feeling 'slow' - no matter what the benchmarks say.

I think I can stick with my Crucial M4's (512GB capacity), Intel 520 Series and Sandisk Extreme 240GB SSD's until Intel releases (hopefully soon) the consumer/mobile side of this new controller architecture/direction.


While reading/writing this post - I ran the Win8 'maintenance' in the background and saw almost 15 second 'average' response time in task manager - yeah over 14 SECONDS AVG (what was the maximum???) with a Sandisk Extreme SSD (240GB) with 27GB free space and ~75GB 'unallocated' capacity (~30% reserve capacity). Good thing this is only my 'digital notebook' I'm running it on...

Thanks Anand (and Win8's task manager...) for finally quantifying in a non-disputable way what I've been trying to say/communicate for these last few years:

SSD's still have a looooong way to go.

 

Show me where you can get a 15k RPM HDD for a laptop and then I might be inclined to kind of agree with you. For desktops SSD's aren't necessarily the best option. For laptops it's a massive improvement from the 5400 and 7200RPM drives. It's not just benchmarks that show performance, but real world use that is apparent like night and day. Your usage scenarios tiller are far and above worse case from a consumer perspective and it's clear you need to invest in enterprise type drives to work with whatever it is that you do to write as much as you do to HDD's/SSD's daily. Most users write maybe 2-3TB a year where you seem to do that every day.

I own both Samsung and Crucial SSD's and have to say Samsung are by far the fastest performing drives. I never get lags, hangs, or any kind of latency that you speak of. 15 seconds is something everyone would notice.

That being said, I am glad Intel is back into design of their own controllers, as I know they know how to do it right.

 

Yeah; you're right, no 15K drives for a notebook and if the 14+ seconds average latency I noticed in Win8's task manager was something I noticed in anything other than Task Manager (not really noticeable when just reading and writing the original post above - though there were some big 'pauses' in pages loading...) - then I would be putting in a HDD right now.

But it does point to the fact that SSD's are not consistent (whether we notice them with a specific workload or not) and Intel is at least beginning to address this very real issue.

Even Anand says as much with his Samsung 830... and even extreme over-provisioning won't help in all cases.


I have stated this previously, I am not an enterprise user: I am a workstation user - just one that pushes systems to their edge, even SSD based storage subsystems.


Alright: you have swayed my pre-biased opinion of the new Samsungs again. I will try them at some point (not available easily here...) - just to be fair to Samsung.

But I have to say; the last graph on the following page/link doesn't paint the 840 PRO in the best light (not even against the lowly Intel 330 Series).

See:
http://webcache.googleusercontent.c....199.756.5j2.7.0.les;..0.0...1c.1.KAJXrimwnjQ


(Click the different SSD's to see how much of an improvement Intel has achieved...)



I guess I'll just have to see for myself what Samsung 'real world' is like in 'my world'.

 

The Intel DC S3700 200GB SSD from Anandtech is now officially up:

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review


Reading it again. And trying not to drool too much.

 

Anand Lal Shimpi:

See:
AnandTech - The Intel SSD DC S3700 (200GB) Review


The following two attachments are from the same link above and really show what I've been 'feeling' these last few years when running SSD's in my workflows. These are not meant to say one SSD or the other is 'better' here (the 330's run is shown with compressible data, for example) - it's just to show that the high latency (which I have been characterizing as 'slow' and 'slower than HDD's...') all SSD's (other than Intel's new SSD) exhibit is a huge, real-world concern not only to me, but to many others that can feel this 'pause', 'stutter', or otherwise annoying SSD 'phenomenon' until now. This has finally been shown in a way others here can 'believe' instead of taking my word for it.

Maximum performance (benchmarks...) mean nothing without a certain minimum level of performance that is sustainable (indefinitely) over time. Just like needing a minimum FPS in games and also a lack of 'micro-stutter' and input lag in multi-gpu setups.

The SSD's in 2012 finally became usable in my workflows (but only with highly overprovisioned setups - 50% or more); not because they are 'perfect' - no, simply that the average performance over time (and taking price into consideration) is finally above my vRaptors when Everything is considered. In a notebook setting; sure, no comparison to anything we can fit in that form factor (not even with Hybrid HDD's or any SSD based caching solution).

But that still doesn't mean that that performance is superior in every way to traditional HDD's. Those SSD lags that Anand reports and have driven me crazy these last few years are real and detrimental to productivity (make you focus on the system, O/S, components ("Is something going to crash???") instead of the reason you're using the computer...).

Intel changes that now.

When they perfect their new vision with client optimized endurance and even more important; notebook optimized power usage I will no longer have a choice...

ALL SSD's I currently have will be disposed of to get the true 'next gen' SSD's into ALL my systems and workflows. Yes, this 'small' change is how revolutionary this new direction is. And as always, Intel is at the forefront of addressing the real needs for power users with real work flow performance issues.

(Waves a few bucks in Intel's face to make them HURRY UP!). :laugh::laugh::laugh:

 

Based on those graphs, I don't see the new DC S3700 as a monumental leap over Intel's previous enterprise-oriented drives - it's more like an incremental improvement. If you feel that your current X25-E or SSD 710 is a disappointment, I doubt you'll be much happier with the DC S3700 if those numbers are reflective of real-world performance.

Compared to consumer drives it's definitely a huge improvement, but that's an apples-to-oranges comparison.

 

Peon, sorry I wasn't more clear.

Those two graphs are showing the current Intel 330 and the Samsung 840 PRO (and they're both pretty horrible, I agree).

The following graph is from the same link as in my previous post and shows what the new Intel DC SSD achieves:

(Yeah, now we're talking, eh?):

 

Im dying to see the consumer version of that drive

 

Yeah, me too (if I could have 'liked' your post 1000x, I would have).


Intel's 'consumer' drives are better than some manufacturer's Enterprise drives... Even Samsung who is so highly regarded is trailing way behind in what it is 'guaranteeing' in total TB written.

See:
Samsung Debuts High-Performance Enterprise SSDs Destined for the Data Center - HotHardware



What makes me smile with Intel is the claim:

And the above is what is 'guaranteed' for ~15% variable performance at the indicated spec's!

Lol... so 1.8TB to 3.6TB (10-20x 186GB formatted capacity) data written each and every day for 1,825 days. I don't know...


I'm going to have to wait for the REAL SSD's to show up!!!

 

yep! kcetech1 should be jumping about right now, she uses those SSDs for scratch disks, and yes nand do wear down, specially if you do that.

The samsung 840 pro is still a good consumer drive, just that this new controller from intel is superior.

I wonder what they are going to keep from the consumer version, probably and entire change on the pcb including getting rid of those caps and stick to the usual 25nm nand. I do hope that they keep the channel configuration though

 

I was bored and explored a little how the immediate future would change for SSD's.

See:
http://www.sata-io.org/documents/NVMe and AHCI_ (long).pdf



I have mentioned that PCI-e is where SSD's are heading - and, they'll be faster because of that.

For the details of why it will be faster, the above link really goes into detail how each 'standard' of the time is only as good as the components generally available then.

It also shows that AHCI's successor, NVMe, will not simply enable the internal AHCI commands our SSD's are familiar with today.

Rather, the new 'driver' will greatly simply while also greatly empower each SATA Express device that is connected to it.

The quote below illustrates how much overhead is inherent in today's SATA drives; whether they are HDD's or SSD's.

With the new specification, SSD's are expected to 'easily' hit 1 Million I/O's (individually; no RAID0 needed).

I would love to be a fly on the wall at Intel's testing labs and see these devices and interfaces being hammered out right now.

Just as this Datacenter SSD makes everything else look from the 1920's - this technology implemented as explained in the link will make this drive look like it's from the Stone Age.

Is there anybody reading anything else interesting on the SSD/storage subsystem front? I'm still bored...

 

you may not find this interesting, but here is one

AnandTech - NGFF SSDs: Putting an End to Proprietary Ultrabook SSD Form Factors

and another

AnandTech - Two New SATA Specs: SATA Express & µSSD

and another one just because of your high iops

AnandTech - Breaking the SATA Barrier: SATA Express and SFF-8639 Connectors

 

Thanks for the reads! Even though I have read a couple of them before.

That last article? Yeah: we want SFF-8639 (let's just skip SATA Express, okay? ). Looks like we'll have this available by Christmas 2013 (if we're good boys and girls).

 

yeah I didnt get the idea of a stop gag interface. Its pretty stupid.

This matters more on the notebook dept, where what you get is what you have, on the desktop you have more space to put legacy controllers/interfaces, but basically if you are going to transition, just do it. it wont matter as etrash is going to be generated anyway, since most consumers wont bother removing their old HDD from their old systems to the new one.

btw if you find anything more, do please post