Has enough time passed to trust 850 EVO?

[GreenMeters]

Really torn between two 1 TB SSDs: Mushkin Reactor and Samsung 850 EVO. EVO's obviously been the more populate, wins on power consumption, theoretical speed, and HW encryption. Mushkin wins on price (though not by much now with 850 EVO on sale) and on being trustier MLC instead of maybe questionable TLC.

Has the 850 EVO been out long enough now to prove it doesn't have the same TLC-related performance degradation as the 840 EVO? Any known issues with the EVO?

 

[Coup27]

AFAIK there are no issues with the 850 EVO. I just built a system with a Haswell Refresh 3.6 Ghz i3, H97 board, 8GB of RAM and a 250GB 850 EVO and the speed increase over the system in my sig was amazing. Booting is half the time of my rig which I thought was already quick.

I think I'll be finding a way to upgrade my rig with one soon!

 

[GreenMeters]

Cool, getting a really itchy trigger finger for the EVO. How long did it take for people to start talking about performance issues with the 840 EVO, and are we past that point with the 850?

 

[Byte]

It took them almost a year to finally admit to the 840 EVO issues, the 850 EVO only came out on DEC, so i would not say it is past any point yet. Regardless they are having some crazy firesales on them and i got a bunch anyways at it is almost cheaper than any other SSD!

 

[QcGuy]

I just finished building a new gaming rig for a friend of mine. We put the Mushkin Reactor in it and it seems to be a very solid drive. The Mushkin does not come with any software like the Samsung Magician software so the tweaks like turning off Superfetch,Prefetch etc you will have to do yourself.
Depending on the benchmark program used the drive is pulling around 530-545 read and 435-465 writes. I guess the best way to describe the Reactor is its not a flashy with a lot of marketing hype but one that just performs and is very close to the fastest drives out there.

I don't think you would be unhappy with it if you picked the Reactor over the EVO. I think Newegg was selling the Reactor last week for like $360-365, we got ours off Amazon for $380.

Hope this helps

 

[ggadrian]

I think that the 840evo had been on the market longer than the 850evo when the problems surfaced.

I might be a little conservative, but I don't really trust TLC nand, I might buy it if it were significantly cheaper, but it isn't compared to crucial's MLC.

I have installed many crucial drives (M500 and MX100) and they have been rock solid and a grea value for the money. They might not be the best performers, but reliability (or at least peace of mind) and price are more important than a small performance difference.

 

[redzo]

Really torn between two 1 TB SSDs: Mushkin Reactor and Samsung 850 EVO. EVO's obviously been the more populate, wins on power consumption, theoretical speed, and HW encryption. Mushkin wins on price (though not by much now with 850 EVO on sale) and on being trustier MLC instead of maybe questionable TLC.

Has the 850 EVO been out long enough now to prove it doesn't have the same TLC-related performance degradation as the 840 EVO? Any known issues with the EVO?

http://www.anandtech.com/show/8747/samsung-ssd-850-evo-review/3
As per Kristians's review, VNAND TLC is quite different than planar TLC. The new 850 evo is very different than the old 840 one. Reliability and consistent performance / time is questionable as it is with any other brand new drive out there.

 

[Charlie98]

The new 850 evo is very different than the old 840 one.

Yes, but I'll bet the terrible outsourced warranty service is the same.

Since my experience with my 840PRO and the Samsung SSD RMA, I've gone to either Crucial or Intel.

 

[Puffnstuff]

The rma experience is enough to convince most people that it just isn't worth it especially with so many quality products on the market right now. It certainly convinced me hence the intel 520 ssd in my signature. The refurb 840 pro went into my laptop and when it croaks another intel will take its place.

 

[hojnikb]

I think that the 840evo had been on the market longer than the 850evo when the problems surfaced.

I might be a little conservative, but I don't really trust TLC nand, I might buy it if it were significantly cheaper, but it isn't compared to crucial's MLC.

I have installed many crucial drives (M500 and MX100) and they have been rock solid and a grea value for the money. They might not be the best performers, but reliability (or at least peace of mind) and price are more important than a small performance difference.

You trust 16nm MLC (which is the absolute minimum cell size you can reliably make) but TLC at ~4x nm is all of a sudden crap, just because it holds 3 bits instead of 2 ?

Some people have really poor understading of NAND sadly.

 

[ggadrian]

I believe that I understand NAND well enough.

It's just that it seems that TLC has some flaws that have to be worked out (and there's people that believes that it's not possible) and until it's proven that they are I don't trust it.

16nm NAND might be the smaller/cheaper node, but I don't really care. The drive performs well and reliably, and if it has do die sooner because it wears faster I'm not going to be sad to replace a 6 or 7 years old drive, it's not like I'm gonna keep a computer longer than that anyway (in fact an MX100 is the 3rd drive that my 5 years old laptop sees).

I am not Crucial fanboy or anything, they just happen to be dirty cheap and in my experience (which I know that it's not statistically relevant) reliable. If some company offers the same at a cheaper price I'm sold.

 

[hojnikb]

I believe that I understand NAND well enough.

It's just that it seems that TLC has some flaws that have to be worked out (and there's people that believes that it's not possible) and until it's proven that they are I don't trust it.

16nm NAND might be the smaller/cheaper node, but I don't really care. The drive performs well and reliably, and if it has do die sooner because it wears faster I'm not going to be sad to replace a 6 or 7 years old drive, it's not like I'm gonna keep a computer longer than that anyway (in fact an MX100 is the 3rd drive that my 5 years old laptop sees).

I am not Crucial fanboy or anything, they just happen to be dirty cheap and in my experience (which I know that it's not statistically relevant) reliable. If some company offers the same at a cheaper price I'm sold.

Not, its not like that. TLC is not some magical techology within NAND, that makes SSDs crap or perform worse. Or even cause issues like we've seen with 840evo

TLC, in simple terms, is just a way to differentiate voltage states within the cell. Nothing else. Having 3 instead of 2 bits per cell does infact mean lower performance and endurance (since having more bits per cell causes more sensitivity for voltage drifts). But its not the sole reason why 840evo is causing such issues with reads.

Smaller geometry NAND combined with TLC means, that you will run into degradation issues like we see in the 840evo. TLC on its own doesnt mean drive will perform like that, as long as cells are big enough to compensate.

So, simply put one can treat 3D tlc at 40nm class fab quite equal to 2D mlc at sub 20nm in terms of performance and endurance.

In other words, what TLC loses with 1 more bit per cell quickly compensates with much much larger cells.
Same could be said for sub 20nm MLC; what it gains by using only 2 bits per cell, quickly loses for having very very very small cells.

 

[Hulk]

Not, its not like that. TLC is not some magical techology within NAND, that makes SSDs crap or perform worse. Or even cause issues like we've seen with 840evo

TLC, in simple terms, is just a way to differentiate voltage states within the cell. Nothing else. Having 3 instead of 2 bits per cell does infact mean lower performance and endurance (since having more bits per cell causes more sensitivity for voltage drifts). But its not the sole reason why 840evo is causing such issues with reads.

Smaller geometry NAND combined with TLC means, that you will run into degradation issues like we see in the 840evo. TLC on its own doesnt mean drive will perform like that, as long as cells are big enough to compensate.

So, simply put one can treat 3D tlc at 40nm class fab quite equal to 2D mlc at sub 20nm in terms of performance and endurance.

In other words, what TLC loses with 1 more bit per cell quickly compensates with much much larger cells.
Same could be said for sub 20nm MLC; what it gains by using only 2 bits per cell, quickly loses for having very very very small cells.

That all makes sense but we won't really know for a few years. Also, Samsung claims the 3D NAND is 40nm but we don't really know the that for sure since it's a totally new process and even Anandtech had a hard time figuring out the node size. Samsung has been known to push the truth a little bit.

And saving/reading 4 voltage states is much less than 8. In this case 8 is a really big number. Is 20nm MLC = 40nm TLC? Do we know this is a linear relationship as far as endurance and data retention are concerned?

I'm not saying that what you wrote isn't the truth, it's just that enough time with this technology hasn't passed for people (myself included) not to have doubts about TLC NAND. The 840 EVO issue seems quite serious.

 

[hojnikb]

Also, Samsung claims the 3D NAND is 40nm but we don't really know the that for sure since it's a totally new process and even Anandtech had a hard time figuring out the node size. Samsung has been known to push the truth a little bit.

Its not 40nm, but a 40nm class node. So it can either be 49 or 40nm. In any case, it would make little sense to go to a smaller process fab, given the advantages of the 3D stacking.

And saving/reading 4 voltage states is much less than 8. In this case 8 is a really big number.

Its really not that big of an issue. The bigger issue is dealing with errors, which are inherent for smaller fab. It gets increasingly difficult to reliably program and read smaller and smaller cells. Even if its only 4 states (mlc).

Is 20nm MLC = 40nm TLC? Do we know this is a linear relationship as far as endurance and data retention are concerned?

This is just a very rought guess. But given the much much bigger cells, its bound to have much better endurance and data retention.
Remember, back in the ~50nm days, MLC was good for 10k+ p/e. So its not that big of a stretch to say TLC is good for 2-3k p/e at that fab (thats not even including benefits of 3d stacking). And that figure is right on par with 2D mlc at sub 20nm.

it's just that enough time with this technology hasn't passed for people (myself included) not to have doubts about TLC NAND.

Again, TLC is not the issue here. Very small cells are.

Think of this techologies as a leaking container. MLC container only has 1 small holes for water to go out, while TLC has 3 of them. If we have two containers with the same amount of liquid, TLC version will obviously run flat sooner.

Now we go and get much much bigger container (so 40nm class cell) and again punch the same 3 holes in it... Yes, its gonna leak the same, but because it has much much more liquid inside, it will take a lot more time to get empty.
Now compare a very small container with only one hole and that bigger container with 3 holes. If 3 hole container is big enough, it will match or even outmatch the 1 hole container in the time, it takes to run out of liquid.


Roughly thats how sub 20nm flash and 40nm class flash relate to each other.

 

[Hellhammer]

Also, Samsung claims the 3D NAND is 40nm but we don't really know the that for sure since it's a totally new process and even Anandtech had a hard time figuring out the node size. Samsung has been known to push the truth a little bit.

Fortunately TEMs exist and allow us to see inside the chips:

http://www.3dincites.com/2014/12/samsungs-3d-v-nand-flash-product-ceaselessly-marching/

 

[ashetos]

It seems like 850 EVO has similar endurance to MX100 when the 850 EVO NAND flash cell size is much bigger.

It seems like TLC is much more difficult to handle, it doesn't look like a TLC cell twice the size of an MLC cell has similar endurance.

Maybe it has to do with the difficulty in handling the different voltage states. Concerning that, it's not fair to compare 2 to 3 bits, but 4 to 8 voltage states.

 

[hojnikb]

It seems like TLC is much more difficult to handle, it doesn't look like a TLC cell twice the size of an MLC cell has similar endurance.

40nm class cell is more than twice as large as their 20nm counterpart.

http://images.anandtech.com/doci/8216/cell size actual_575px.png

 

[ashetos]

40nm class cell is more than twice as large as their 20nm counterpart.

http://images.anandtech.com/doci/8216/cell size actual_575px.png

I know, in Hellhammer's link it says 24 times larger. I was trying to make a point.

 

[ashetos]

To further clarify, the point is that having twice the cell size does not make endurance similar, even though you have twice as many voltage states (MLC 4, TLC 8). We can conclude it doesn't work like that.

It seems that a huge cell (850 EVO, 24 times larger than 16nm MX100) is needed to achieve endurance parity. This convinces me that TLC is a very problematic technology indeed.

 

[Hulk]

Its not 40nm, but a 40nm class node. So it can either be 49 or 40nm. In any case, it would make little sense to go to a smaller process fab, given the advantages of the 3D stacking.

Its really not that big of an issue. The bigger issue is dealing with errors, which are inherent for smaller fab. It gets increasingly difficult to reliably program and read smaller and smaller cells. Even if its only 4 states (mlc).


This is just a very rought guess. But given the much much bigger cells, its bound to have much better endurance and data retention.
Remember, back in the ~50nm days, MLC was good for 10k+ p/e. So its not that big of a stretch to say TLC is good for 2-3k p/e at that fab (thats not even including benefits of 3d stacking). And that figure is right on par with 2D mlc at sub 20nm.

Again, TLC is not the issue here. Very small cells are.

Think of this techologies as a leaking container. MLC container only has 1 small holes for water to go out, while TLC has 3 of them. If we have two containers with the same amount of liquid, TLC version will obviously run flat sooner.

Now we go and get much much bigger container (so 40nm class cell) and again punch the same 3 holes in it... Yes, its gonna leak the same, but because it has much much more liquid inside, it will take a lot more time to get empty.
Now compare a very small container with only one hole and that bigger container with 3 holes. If 3 hole container is big enough, it will match or even outmatch the 1 hole container in the time, it takes to run out of liquid.


Roughly thats how sub 20nm flash and 40nm class flash relate to each other.

Your counterpoints are very convincing. I'm going to reconsider an 850 EVO for my next SSD purchase. I don't think that endurance is really going to be an issue for just about any SSD 240GB or larger in 99.9% of consumer usage scenarios. It's the data retention that's the possible cause for concern.

Your leakage analogy is a good one. I understand as the cells get larger they can hold more electrons. But do the larger cells leak the same or less rate-wise (electrons lost per unit time) as small cells?

Or put another way how does the rate of NAND cell leakage relate to cell size for a given voltage applied to the cell?

 

[ImDonly1]

Well I hope it is fine, because I just bought a 256gb 850 evo for $100 from newegg.
Was thinking about the pro, but missed the last sale.

If the 850 is problematic, next ssd will definitely be crucial (as I already have one of those and it works fine).

 

[VirtualLarry]

If the 850 is problematic, next ssd will definitely be crucial (as I already have one of those and it works fine).

Expect that the next major price drop will be around Intel / Micron's 3D NAND efforts. Samsung was basically the first, but that doesn't mean that they are the only.

 

[PhIlLy ChEeSe]

So,
(not to highjack this thread)But who is the reining top dog now as far as SSD'S go?

 

[Riceninja]

850 evo uses a diff technology and also a larger node in order to combat these issues. i'm pretty sure it wont have the same problems and i put my money where my mouth is and bought one last month.

 

[hojnikb]

Your counterpoints are very convincing. I'm going to reconsider an 850 EVO for my next SSD purchase. I don't think that endurance is really going to be an issue for just about any SSD 240GB or larger in 99.9% of consumer usage scenarios. It's the data retention that's the possible cause for concern.

Your leakage analogy is a good one. I understand as the cells get larger they can hold more electrons. But do the larger cells leak the same or less rate-wise (electrons lost per unit time) as small cells?

Or put another way how does the rate of NAND cell leakage relate to cell size for a given voltage applied to the cell?

As far as i understand it, bigger cells should leak less, since they have thicker "walls" + much more electrons inside due to being bigger.

As far as data retention is concerned; i suspect it should be much better than 840evo given the large cells.

I mean, even the very very small 16nm MLC doesnt show any degradation issues so far, so i doubt 3d tlc is any worse than that.