Samsung Outs Z-NAND's Performance; Intel's Optane On Notice

  • Thread starter Thread starter Guest
  • Start date Start date
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
3dxp (optane) drives get to 10 usec read times with 1-2 usec of media read time.

Typical NAND flash read time is 50 usec.

It'll be interesting to see how Samsung gets to 12 - 20µs read times with zFlash -- if this is just a DRAM buffer in front of FLASH number that would be very disappointing and applications with random read patterns would see bad read performance.

If this is truly a consistent 12 - 20µs read time then this is something new for NAND. (aside: think this is what the author Paul Acorn meant by saying "Samsung hasn't released QoS metrics, such as 99th percentile performance".)
 
A DRAM buffer is actually very good for random reads.

In fact Tom's has mentioned in previous articles, ssds without DRAM buffers tend to not have great performance.

The main factors are:
1. The speed of the DRAM buffer, not all emulated slc is the same.
2. The algorithm used, magical black box that turns data into awesome data.
3. Size of DRAM buffer, if you only have a 5 gigabyte buffer it doesn't matter what values you use for the first 2.

A best case scenario, but still valid, the Samsung 850 Evo 4 Terabyte has 96 gigabytes of buffer before performance drops to tlc levels.

http://www.tomshardware.com/reviews/samsung-850-evo-4tb-ssd-review,4623-3.html
 


"..ssds without DRAM buffers tend to not have great performance..." agree completely for NAND flash based SSDs today, this is not true for 3DXP based drives. THIS is exactly why the samsung z-nand flash is interesting. Hope it is *not* using a cache to get it's published numbers.

"...A DRAM buffer is actually very good for random reads...." unless you get some form of locality (non-random) behavior a read cache is useless. For example, you do email for a while then shift to loading a game. The records you need to load the game are not in the cache. Unless you detect sequential (non-random) patterns and pre-stage none of the records you need will be in the cache. For random read workloads a 3xdp drive will have an advantage over a nand flash drive with a dram cache. For sequential and predictable workloads the NAND flash drive might do OK.

...has 96 gigabytes of buffer before performance drops to tlc levels..." This is for WRITES, not reads. It allows the drive to quickly consume a lot of writes (much faster response then if it tried to write to TLC flash). But that is not used for reads (unless the data happens to have just been written to the buffer and has not yet been destaged).
 

You're mixing up DRAM with pseudo-SLC NAND buffer/cache.

Either way these numbers they're claiming can't be the result of DRAM alone. They had to make other improvements, probably at the cost of density. I don't think it will fully challenge Optane in terms of performance, but it will serve as a more-affordable alternative until they have something other than NAND to deploy as a challenger.
 

If the benchmark isn't complete junk, then a DRAM buffer will be virtually useless with random reads (which should be distributed throughout the whole drive).
 
Where this thing probably still can't compete is in NVDIMMs. Those require good random access performance down to the size of a cacheline - or even byte! The performance of NAND flash is too reliant on large block operations for it to compete, there.
 

Comment on the forums. There's nothing wrong with the forum software, and they have no ads.
 
As Bit said, you can click "Comment from the forums" to solve the second problem. As far as ads on articles go: Install Adguard or similar. I personally disable Adguard protection for select sites including Tom's, because they need revenue to keep putting out reviews I like to read without a paywall. I find performance is reasonable even with Adguard turned off (FCU, Edge). But YMMV and I'd rather see people run a decent adblocker rather than fret about the ads.

The forum software is definitely superior for commenting. But technically there are still SOME ads when browsing comments from the forum. Banner ads - not nearly as obtrusive though.
 

I'm using no ad blockers - just standard Firefox - and I never see ads on the forums.
 
Has to be something in the browser. They're there. Just simple banner ads, though.
 

Try this:

http://www.tomshardware.com/forum/id-3569107/samsung-outs-nand-performance-intel-optane-notice.html

After it loads, paste it in your address bar and hit Enter. Sometimes, when the forums pages first load, it's like I get a different stylesheet. Then, I load them again and it's back to the same old look as it's been for a long time.
 


Some of the ads being displayed, are from domains blacklisted by McAfee's WebAdvisor.....specifically, the Revcontent ads....
 


No its not.

Bandwidth=!Latency

System memory is at less than 100ns. At 20us, this device would be 200x+ slower. Not to mention the gap in IOPS between read and writes, that says the fundamentals are still NAND, nothing revolutionary.
 
What about endurance? Those specs are suspiciously absent from Samsung's claims.

I thought one of optane's biggest selling points was it's ridiculously good endurance.
Samsung's NAND product might end up costing half as much, but need be replaced 10x as often. If that's the case, then they end up costing more in the long run.
 

Take another look. It's listed in the table as "Drive Writes Per Day (DWPD)".


...and that's another area where 3D XPoint underperformed Intel's initial claims. It's still not close to DRAM-level endurance, meaning you can't just swap out your DDR4 DIMMs for NVDIMMs. Much better than most current NAND flash, though.

I think it's more plausible to use NVDIMMs with about 4 GB of HBM2. This should be a big win for mobile, since both are power-saving technologies and the performance improvement from HBM2 should generally offset the extra paging overhead, while simultaneously reducing the frequency of writes to the NVDIMMs. Perhaps this is where Intel is going in their joint effort with AMD.
 


PRICE.

An NVDIMM is DRAM plus FLASH plus the mechanism to dump the data from the DRAM to the FLASH when the dimm loses power.

z-nand and 3dxp are persistent media that are slower and cheaper than dram but are fast enough that they can substitute for dram where slower but cheaper makes sense. They can also substitute for nand flash SSDs when use cases pay a price premium for low latency.

Yes the NVDIMM is faster than z-nand or 3dxp. But it is not a price performer,it costs more per byte than DRAM and tons more per byte than z-nand or 3dxp. NVDIMMs do not compete with them as either a lower cost dram replacement or as a fast solid state drive because price breaks those use cases.

HP 8GB (1x8GB) SDRAM NVDIMM Compatible Product by NETCNA
Be the first to review this item
Price: $2,913.73 & FREE Shipping (amazon)

 

Relatively speaking.


Unless your DBs are read-biased, as many are.
 


They actually are read-biased most of the time. But the read stuff used 99.99% of the time fits into RAM. Gota love modern Servers and their max RAM capacity measured in TBs... ;-)

It's the super write biased update/clean/import runs that still take many hours. At least RAID Arrays of current gen SSDs got it down to that. Before on RAIDs of 10k platters some runs took days... Xpoint will take it down another order of magnitude. Maybe not current V1 gen of it. But I have little doubt one of the next gens will...

Decided to delay server upgrade that was planned for next year. One for Xpoint (or at least nvme/U.2 RAID cards). Two for insane RAM prices....


 

With such fast read performance, you mightn't need so much RAM.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom