Intel G3 SSD's postponed till Febuary 2011?

[=Wendy=]

the OCZ has speed but no reliability.

According to an OCZ rep, from a sample of over 50,000 Sandforce based drives sold. Less than 1% have been returned under RMA, and from those returned, most of them have suffered a bad firmware flash or the drive has locked in panic mode. In either case the drive is easily recovered ( I assume OCZ have some special software tools for this).
Very few drives have actually died, according to OCZ.


Regarding the G3. There could be many reasons why it has been delayed. Maybe their reason for delaying is correct.

 

[exar333]

Anyone have proposed specs on the new G3? Me want.

 

[AbRASiON]

Any specs out? would be nice - new controller? more cache?
Knowing intel it'll probably have a lightpeak port on it.

 

[IntelUser2000]

My guess:

340MB/s read, 250MB/s write, SATA III-600, better controller that can handle much more loads before degradation starts to happen. Peak IOPS might actually decrease but sustained performance will dramatically improve.

 

[Emulex]

100,200,400,800gb mlc for enterprise man. 6gb sas (full duplex), with controllers to the market that can handle (trim,8 of them).
x25-e g2 - for sure 100% hands down new SLC with capacitor backup (novembers is the end of battery back write cache anyways, its all capacitor/flash for traditional dasd now)
i think the main problem is the enterprise is ready for all of the flash it can buy - and is chomping at the bits. micron probably can't make enough.

keep in mind it's micron 25nm chips - intel has a stake but not all - so crucial and intel get the divvy the goods up - any scraps might go to the others

so pricing may not be at surprising (above msrp) as you expect. I suspect it will be cheaper for many to buy enterprise mlc from their work (if its a big ass company) discounted way more than any etailer.

just sayin'

 

[taltamir]

same as the intel G2, only with SATA3, and firmware improvements (including relaxing the built in cap on speed). kinda like the G2 was compared to G1.

 

[FishAk]

same as the intel G2, only with SATA3, and firmware improvements (including relaxing the built in cap on speed). kinda like the G2 was compared to G1.

If the sequential file transfers aren't improved substantially above the G2, there will be no need for SATA III. Intel is the weakest here,compared to the other good controllers, and is only saved by their great small file and I/O rate.

 

[taltamir]

If the sequential file transfers aren't improved substantially above the G2, there will be no need for SATA III. Intel is the weakest here,compared to the other good controllers, and is only saved by their great small file and I/O rate.

1. intel is weakest because they cap it, artificially.
2. intel managed to greatly improvement performance from G1 to G2 on firmware alone.
3. 25nm MLC is inherently faster than its predecessor.

but yes, intel needs to provide some solid improvements.

 

[FishAk]

1. intel is weakest because they cap it, artificially.

I wasn't aware of that. It seams silly to me, unless they had problems with the higher rates.

Honestly, this is the only reason I wouldn't buy an Intel SSD. Without higher sequential rates, there is no reason to have anything except the OS and programs on the SSD. For everything else, a good HDD is just as fast as the Intel for much less money. This makes the 80Gb too big, and the 40Gb is missing to much I/O performance, so neither can really compete with the likes of SandForce, or the C300 in terms of performance/value. The sequential speeds of their competitors are 2x better than a HDD, so the higher capacity of those SSDs have value.

 

[taltamir]

I wasn't aware of that. It seams silly to me, unless they had problems with the higher rates.

product differentiation. Their firmware artificially limits their speed on the 25-M drives so that they aren't too close to the speeds of the 25-E drives.
Otherwise a lot of companies would buy the 25-M instead of 25-E

 

[Obsoleet]

My G2s and 5870 were some of the best PC purchases I've done, outside of my 3dfx Voodoo1.


While I would love to have a G3, honestly the G2s are so good that I don't care. Buying and using a G2 has been one of those "everything works" and reliably since day 1.. for me, the era of SSDs was here a long time ago. I'm glad I didn't wait.

 

[Obsoleet]

If the sequential file transfers aren't improved substantially above the G2, there will be no need for SATA III. Intel is the weakest here,compared to the other good controllers, and is only saved by their great small file and I/O rate.

In my experience with SSDs, sequential is the least important part of the equation.

 

[taltamir]

My G2s and 5870 were some of the best PC purchases I've done, outside of my 3dfx Voodoo1.

While I would love to have a G3, honestly the G2s are so good that I don't care. Buying and using a G2 has been one of those "everything works" and reliably since day 1.. for me, the era of SSDs was here a long time ago. I'm glad I didn't wait.

I completely agree with everything you said.
I :heart: my G2.

 

[Emulex]

remember that server grade mlc is coming out - current intel gear has no write-back cache (no battery, no capacitor) - all the ram is for read cache and write strategy.

the other guys are using ram (unprotected i guess) to accomplish massive increases in speed - which seems terribly unsafe.

So when they put capacitors on the next gen and rock some write-back caching i suspect we'll see some massive improvement in all directions. I hope they do this for the consumer line as well. But i suspect if they make the x25-m G3 that good - some people might shy away from the server grade mlc drives. The SLC of course is still going to be there but with mad faster speed with capacitor write back cache and of course the fact that slc lasts 10x as long. For everything you can do in firmware to prolong the write cycles of MLC - you can do the same * 10 for SLC.

i wonder if anyone will use the new 3bit for uber cheap ssd

 

[taltamir]

the other guys are using ram (unprotected i guess) to accomplish massive increases in speed - which seems terribly unsafe.

it doesn't SEEM terrible unsafe, it IS terribly unsafe. This makes them unacceptable for server use, period.

So when they put capacitors on the next gen and rock some write-back caching i suspect we'll see some massive improvement in all directions.

they didn't put capacitors on current gen because capacitors are too expensive... have capacitors gone down in price?

 

[Old Hippie]

In my experience with SSDs, sequential is the least important part of the equation.

Gets my vote.

I thought I would like these Vertex 2s better that the Intels because of their superior sequential write speeds but I was wrong again.

The Vertex 2s put the Intels to shame on compression jobs (RAR/unRARing) but they lose some of their luster on day-to-day normal use.

AAR, I've purchased my fair share of SSDs (5) in the last 1.5 yrs and I'm gonna wait for the new Intel SATA6 stuff before another HD upgrade.........at least...that's the plan!

 

[Voo]

it doesn't SEEM terrible unsafe, it IS terribly unsafe. This makes them unacceptable for server use, period.

Actually a server should have a UPS which would make it rather safe to use, wouldn't it?

 

[Emulex]

batteries cost more than capacitors lol and fail. failure is not good. batteries are very heat sensitive and require alot of control to maintain life. they only hold 36-72 hours tops when new and degrade about 50% each year or so (depends on so many factors).

capacitors have been in use in san for years. they are not that expensive relative to cost of an X25-E. lol.

 

[=Wendy=]

remember that server grade mlc is coming out - current intel gear has no write-back cache (no battery, no capacitor) - all the ram is for read cache and write strategy.

the other guys are using ram (unprotected i guess) to accomplish massive increases in speed - which seems terribly unsafe.

So when they put capacitors on the next gen and rock some write-back caching i suspect we'll see some massive improvement in all directions. I hope they do this for the consumer line as well. But i suspect if they make the x25-m G3 that good - some people might shy away from the server grade mlc drives. The SLC of course is still going to be there but with mad faster speed with capacitor write back cache and of course the fact that slc lasts 10x as long. For everything you can do in firmware to prolong the write cycles of MLC - you can do the same * 10 for SLC.

i wonder if anyone will use the new 3bit for uber cheap ssd

Which guys are you talking about, regarding using ram?
I know Indilinx used a 64MB cache, and if i recall correctly, Samsung used a larger one for their controller.

 

[Emulex]

intel uses 256meg for read only iirc. is that megabit or byte i would have to google that.

A UPS does go out/bad. A UPS does fail (fires,poof), a UPS does get shutdown accidentally (oops we're human). There are always N-ways to rain on your parade it seems. remember those doofus that hit the kill-switch the single button power down (required in a datacenter for firemen)? human err wrecks !@#% .

 

[tweakboy]

Still too expensive price per gb their competition are mechanical drives and you wont see GB price drop until 2013 imo. This technology is going slowly. 500GB SSD for 100 dollars by 2015. what you say..

 

[Voo]

A UPS does go out/bad. A UPS does fail (fires,poof), a UPS does get shutdown accidentally (oops we're human). There are always N-ways to rain on your parade it seems. remember those doofus that hit the kill-switch the single button power down (required in a datacenter for firemen)? human err wrecks !@#% .

Well I know several large server farms where stuff like that is highly regulated and controlled (for good reason imho). And while there's the minimal chance that the UPS (and backup) will go bad between a controll and the problem, well you can probably always construct a situation where nothing will help, but I'd think the chances are really slim.

But obviously depends on how serious you take the whole business (just think about all those people with backups who never test how they'd restore those.. read funny stories about stuff like that^^), but I wouldn't have a problem with SSDs without a capacitator in there.. wouldn't want to use one in my desktop PC though and all in all I assume the negligible extra cost isn't worth losing sleep over and safer is safer.

 

[rgallant]

-it's Q4--checked road map again for G3's.

-checking the web every day to see if the delay is correct,
-delay posted from the one site only -wouldn't be taken-ed as true on any other high end tech. -like 5 months.[not including nv 4xx]
-not sure if it's right, should have some leaks-conformation one way or the other by now ?

 

[Emulex]

G3's consist of several lines now - the consumer line (V/M) - i'm not so sure how they are going to satiate the demand enterprise wants - and will pay. i'm going to get 3-4 400gb drives (server grade) for raid-1 (tmp/log) should kick it hard - i can afford to pay 800-$1200 each - i suspect the same flash is used to make the server mlc and consumer mlc - so who do you think they'll sell to - either way higher than msrp.

intel only gets half of the micron production too(nand). it's going to be tight.

can't wait for ES samples need to find a x16 raid controller or two

 

[taltamir]

Actually a server should have a UPS which would make it rather safe to use, wouldn't it?

no, because there could be failed PSU, a prolonged power outage, actually a powerloss due to ANY failure of ANY hardware is a problem.
See, the problem with a server is that the drive LIES to the OS in order to gain its speed... it tells the OS "it has been written" when it is just in the ram and has NOT, in fact, been written.
This means you WILL have dataloss even if your software is smart enough to know better... A bank, or any other important database, cannot afford such lies... as long as it KNOWS there is dataloss at a power loss (because the drive did not lie to it), then it can prevent dataloss via redundancy and distributed server platforms... but if the drive lies to it then it thinks it was written and discards that data forever / doesn't bother fixing it.

This is a problem with spindle drive cache as well, which is why they either disable it (server drives allow it to be disabled, or don't even have it in the first place)

With home user software it doesn't matter, we don't have software that can pick up from where it was and avoid damage in case of a power loss, thus, any file we worked on is lost in a power loss, capacitor or no capacitor. So its not useful for a home user...

Another issue btw, is that if 1 single bit of data is corrupt it will FAIL to rebuild (it will literally abort halfway with an error message, part of why classic RAID sucks so bad and needs to be replaced with smarter implementations like zfs's raidz)... anyways, if you have multiple drives, and they lie about what was written (due to cache), then it WILL cause data mismatch, which will fail your raid rebuild in case of a drive replacement (for any reason). Again, the solution is to disable cache... and that is done for spindle drives as well.