Why do SandForce SSD controllers have a bad reputation?

[hojnikb]

that's, and I'm sorry, but you're wrong.

 

I'm gonna need a detailed document describing such behaviour.

Otherwise i'll mark this as something you pulled outta ur buttocks (like a lot of things today).

[patrickjp93]

I'm gonna need a detailed document describing such behaviour.

Otherwise i'll mark this as something you pulled outta ur buttocks (like a lot of things today).

It's just Samsung's own engineers' answers to a 2-hour long Q&A with the ECE professors and students at Miami University.

[hojnikb]

You have a source for that? And if you do, maybe you could answer the seemingly obvious question: why the F@CK would anyone limit themselves to 32-bit architecture now by choice? If Samsung could build 4TB drives they would and they should. It would force every other SSD maker into deeper competitive waters and just plain make them look good.

 

 

Just look up Samsung MEX or MDX. Youll see its based on cortex R series (which is 32bit).

But bittnes of controller really doesnt play a role here for adressing flash blocks.

They could build 4TB drives (just raid 4 controllers and you're done) but they are not gonna do that. They'll either develop a new controller to adress more space with current block/page size or simply up the page/block size. You can guess wich one is more likely

[hojnikb]

It's just Samsung's own engineers' answers to a 2-hour long Q&A with the ECE professors and students at Miami University.

So it pretty much could be just some PR bullshit for all i care.

[patrickjp93]

Just look up Samsung MEX or MDX. Youll see its based on cortex R series (which is 32bit).

But bittnes of controller really doesnt play a role here for adressing flash blocks.

They could build 4TB drives (just raid 4 controllers and you're done) but they are not gonna do that. They'll either develop a new controller to adress more space with current block/page size or simply up the page/block size. You can guess wich one is more likely

Well yes, you'd have to be a lazy idiot to raid 4 controllers due to worse performance scaling and greater electrical usage and heat.

[patrickjp93]

So it pretty much could be just some PR bullshit for all i care.

Doubtful. They wouldn't answer proprietary questions at all and only explained how components work together in general. It's not PR bullshit. Bad blocks are only relevant to wear-leveling and triggering leakage checks. I suspect they may even become SLC blocks for turbowrite before being deigned totally useless.

[hojnikb]

Doubtful. They wouldn't answer proprietary questions at all and only explained how components work together in general. It's not PR bullshit. Bad blocks are only relevant to wear-leveling and triggering leakage checks. I suspect they may even become SLC blocks for turbowrite before being deigned totally useless.

 

Thats actually very possible, but then again, one can't be certain, since these things have pretty much zero details on how they works.

 

I would very well be interested in a NDA like documentation on controller tehnology

[hojnikb]

Well yes, you'd have to be a lazy idiot to raid 4 controllers due to worse performance scaling and greater electrical usage and heat.

 

Well, scaling can be quite good (just look at all the pci-e solutions) if done right, but energy use is obviously a drawback.

 

But to get more than 1TB at this point, thats the only way. Hopefully this changes in the future, as i would like to see large SSDs.

[patrickjp93]

Well, scaling can be quite good (just look at all the pci-e solutions) if done right, but energy use is obviously a drawback.

 

But to get more than 1TB at this point, thats the only way. Hopefully this changes in the future, as i would like to see large SSDs.

Parallel scaling on memory access processing never goes well. Merely see SLI and XFire.

[Askew]

I've learned a lot from viewing this debate, thanks for arguing in this thread.

[hojnikb]

Parallel scaling on memory access processing never goes well. Merely see SLI and XFire.

 

Actually it works pretty well (again look at all the pci-e solutions using raid0 architecture -- like revodrive). Obviously only sequential speeds scale very well (to a certain point atleast) while random doesn't scale very well (especially on lower queue depth).

That being said, this has little to do with graphics tech you're mentioning

[hojnikb]

I've learned a lot from viewing this debate, thanks for arguing in this thread.

Sorry buddy, for hijacking your thread like this

[Askew]

Sorry buddy, for hijacking your thread like this

 

I think it's not a hijack now, I asked a question early on and it was answered, now I've spent today migrating my system onto the SSD and as of about 10 minutes ago everything is perfect and I couldn't be happier with the responsiveness I have as a result of my upgrade, SSD master race indeed!

 

Reading the comments of this debate has been a nice break from time to time today I'm happy it happened.

[hojnikb]

I think it's not a hijack now, I asked a question early on and it was answered, now I've spent today migrating my system onto the SSD and as of about 10 minutes ago everything is perfect and I couldn't be happier with the responsiveness I have as a result of my upgrade, SSD master race indeed!

 

Reading the comments of this debate has been a nice break from time to time today I'm happy it happened.

I'm glad you like your ssd

 

I remember back when i bought my first one; I never looked back. Now all of PCs (including PCs from family members) use atleast one SSD.

Dat smooth feel is almost adicting

[Askew]

I'm glad you like your ssd

 

I remember back when i bought my first one; I never looked back. Now all of PCs (including PCs from family members) use atleast one SSD.

Dat smooth feel is almost adicting

 

I am amazed by how much faster I can use the system now, I have a good CPU (3570K) and now even things like the delay when I add/remove displays in Nvidia control panel before the Yes/No to keep tab appears is faster, I expected faster boots and fast launches but little things like that make it even sweeter.

[hojnikb]

I am amazed by how much faster I can use the system now, I have a good CPU (3570K) and now even things like the delay when I add/remove displays in Nvidia control panel before the Yes/No to keep tab appears is faster, I expected faster boots and fast launches but little things like that make it even sweeter.

 

And it gets even better. If you manage your data correctly, you can have near silent system, as you can pretty much shutdown the hdd all the time (you set a low spindown time for HDD -- like few minutes tops).

[Askew]

And it gets even better. If you manage your data correctly, you can have near silent system, as you can pretty much shutdown the hdd all the time (you set a low spindown time for HDD -- like few minutes tops).

 

I am currently trying to keep all the programs that will require data on the SSD and stuff like games etc onto the mechanical drive.

 

My case is a Silencio 650 that is fully lined with sound absorbant material and it's situated in a corner of my room that I have surrounded in convoluted sound dampening foam so the machine is barely audible, all the fans are NF-F12s that spin very slowly when idling despite my system being air cooled it's quite quiet.

 

I'm not going to lie and say it's inaudible because that isn't true but it is very quiet, I have used the studio foam to compensate for not using watercooling I suppose. Certainly much cheaper

[patrickjp93]

Actually it works pretty well (again look at all the pci-e solutions using raid0 architecture -- like revodrive). Obviously only sequential speeds scale very well (to a certain point atleast) while random doesn't scale very well (especially on lower queue depth).

That being said, this has little to do with graphics tech you're mentioning

that splits multiple streams to multiple drives. With the SSD internal scaling it would be redirecting 1 stream to multiple possible locations. It wouldn't work so well.

 

In fact this is the exact same paradigm as SLI and XFire. Everything goes to one card which then starts moving things around to the other cards who then interexchange frames. The scaling would be exactly the same idea, and it won't go well.

[hojnikb]

that splits multiple streams to multiple drives. With the SSD internal scaling it would be redirecting 1 stream to multiple possible locations. It wouldn't work so well.

 

In fact this is the exact same paradigm as SLI and XFire. Everything goes to one card which then starts moving things around to the other cards who then interexchange frames. The scaling would be exactly the same idea, and it won't go well.

 

 

It works pretty well, as its evident from multiple drives using such architecture. But its not meant for every possible application.

But its still better to have a single controller solution. Now that sata express is getting some traction and native pci-e controllers coming out, things will get pretty interesting and drives like revodrive will soon be redundent. Which is a good thing, as i can't image using 4 crappy controllers using RAID0 :o

To be honest, i can't image using any drive in raid0. Guess i value my data a little bit too much

[patrickjp93]

It works pretty well, as its evident from multiple drives using such architecture. But its not meant for every possible application.

But its still better to have a single controller solution. Now that sata express is getting some traction and native pci-e controllers coming out, things will get pretty interesting and drives like revodrive will soon be redundent. Which is a good thing, as i can't image using 4 crappy controllers using RAID0 :o

To be honest, i can't image using any drive in raid0. Guess i value my data a little bit too much

Eh, until Intel and AMD both shore up their mainstream processor lineup with more PCIe lanes, those solutions will only be available on enthusiast-grade chips, and at that point having only 40 PCIe lanes on Intel's best (8 more than AMD's best) leaves you only 8 lanes to work with outside the graphics cards, leaving 2 or 4 for wifi, 2 for sound, and 2 or 4 for a single PCIe/Ultra M.2 SSD solution.

 

With just 16 lanes available, or 20 in the case of Skylake (such B.S. upgrade) it's not even reasonable...

[hojnikb]

Eh, until Intel and AMD both shore up their mainstream processor lineup with more PCIe lanes, those solutions will only be available on enthusiast-grade chips, and at that point having only 40 PCIe lanes on Intel's best (8 more than AMD's best) leaves you only 8 lanes to work with outside the graphics cards, leaving 2 or 4 for wifi, 2 for sound, and 2 or 4 for a single PCIe/Ultra M.2 SSD solution.

 

With just 16 lanes available, or 20 in the case of Skylake (such B.S. upgrade) it's not even reasonable...

Thats why we have things like PLX chips. 16 lanes is plenty, considering the bandwidth of pci-e 3.0 (16GB/s is damn fast if you think about it). typical gpus don't even need more than 8 lanes effectivly, 2 for the ssd and you're left with 6 for everything else (can't think of anything, that could use 6GB/s bandwidth outside of those things i've mentioned in a client PC).

 

So yeah, pci-e is the future. Only thing left, that needs to be fixed is freaking pci-e 3.0 support for Intels PCH. It saddens me, to see pci-e 2.0 on their latest chipsets. Its pathetic.

[Askew]

Thats why we have things like PLX chips. 16 lanes is plenty, considering the bandwidth of pci-e 3.0 (16GB/s is damn fast if you think about it). typical gpus don't even need more than 8 lanes effectivly, 2 for the ssd and you're left with 6 for everything else (can't think of anything, that could use 6GB/s bandwidth outside of those things i've mentioned in a client PC).

 

So yeah, pci-e is the future. Only thing left, that needs to be fixed is freaking pci-e 3.0 support for Intels PCH. It saddens me, to see pci-e 2.0 on their latest chipsets. Its pathetic.

 

What about latency when using a PLX chip? I hear complaints about that all the time but I've never seen it substantiated or used a board that used PLX to juggle lanes myself.

[hojnikb]

What about latency when using a PLX chip? I hear complaints about that all the time but I've never seen it substantiated or used a board that used PLX to juggle lanes myself.

 

That could be an issue. But then again, unless you're really a power user, you won't need that and neither you will need more than 16 lanes.

16 lanes is plenty.

But chances are, that intel and amd could incorporate more pcie lanes in their future processors, once sataexpress gains mainstream popularity. But its gonna take a while for that to really happen.

[patrickjp93]

That could be an issue. But then again, unless you're really a power user, you won't need that and neither you will need more than 16 lanes.

16 lanes is plenty.

But chances are, that intel and amd could incorporate more pcie lanes in their future processors, once sataexpress gains mainstream popularity. But its gonna take a while for that to really happen.

SATA needs to die anyway. Thunderbolt-based SSDs should have become a standard long ago. Leave behind 2 SATA 6 Gbps ports for rotating rust and just reallocate the extra bandwidth from the old SATA ports to PCIe lanes and thunderbolt.