SAMSUNG 850 PRO IN RAPID MODE
During our attendance last year at the Samsung SSD Summit in Seoul, South Korea, David Lin of former NVELO fame, now Samsung, introduced us to RAPID (Real Time Accelerated Processing of I/O Data). When enabled through the Samsung SSD Magician, RAPID uses spare DRAM as a cache in order to provide read acceleration and write optimization for a system level performance boost. At that time there was a cap of 1GB on the maximum memory size that could be used and we asked David why they didn’t increase this limit for those of us with 32 and even 64GB of memory in our systems. His response… “Wait for it.”
This years Samsung Magician Ver. 4.4 release contains a new and improved RAPID. Samsung has included a new DRAM usage algorithm that now allows for a maximum memory usage of 4GB when system memory in use is 25% or less. Last year, RAPID increased SSD performance to just over 1GB/s. Take a look at this years results and get ready for some massive cached speed…
An important note with this ATTO result is how unpredictable performance can be when using DRAM as a cache. Ideally, we like to see ATTO speeds gradually increase with file size.
With Crystal DiskMark, RAPID claims to reach speeds of over 7GB/s read and 4.5GB/s write, with a low 4K write score of 692MB/s. The true question will be how these speeds translate into real life transfer speeds.
As unbelievable as these are, we were definitely surprised with the AS SSD Copy Benchmark scores as this benchmark is a simple real life benchmark where three different files are moved from one part of the disk to another.
Last but not least, Anvil provides a complete picture of throughput and IOPS speeds.
When we look at RAPID performance as shown here, an understanding that these results are the product of our DRAM being used as a cache is necessary. This performance doesn’t create a new standard and shouldn’t be viewed as increasing ones PC performance more than tenfold, but rather, RAPID will provide benefits in specific scenarios. As for reliability, we have had one of our systems running RAPID for just over a year now and RAPID is still running strong. To sit at that system, and than any other in our office however, one could not visibly see any difference in ‘typical user activities’. Still, amazing number though eh?
waahhhh a price point would be nice as well
Pricing is listed in the article….thanks ahead!
Not worth the price premium IMO. I don’t care much about SSDs faster than the Crucial M550 or Samsung 830 Evo until the interface becomes faster (ie. SATA Express next year)
> Each chip has a RAW value of 128GB
That is wrong. you will find different types of flash chips on both sides of the ssd:
16 x 86 GBit = 172 GByte x 4 = 688 GByte
8 x 86 GBit = 86 GByte x 4 = 344 GByte
Sum: 1032 GByte
So basicly inside the packages there is no need for stacked dies, since there are already so many tranzistors packed on each other.. Is this correct or can 3D nand have multiple dies aswell ?
It looks like this drive is the new king – of SATA 3 drives. That and well, it seems to be bottlenecked by SATA 3. I imagine an XP941-like drive with V-NAND would be awesome.
I’m not sure what to make of it though.
– PCI-E, M.2, and likely SATA Express drives will probably be faster
– Among the PCI-E drives, well, the MX100 is offering literally double as much storage for the same price, assuming these MSRPs are to be believed
– Then there’s the risk of V-NAND issues if things don’t work out
– On the upside, if it does, well V-NAND because it’s using a bigger process than the others, ought to have pretty awesome endurance. We are looking at 40 nm V-NAND here.
Hmm … at 20 nm, the amount of storage available in a few years ought to be like 16Tb? Especially if they get more layers of V-NAND on the chips.
Hopefully samsung brings this to mainstream drives aswell. It would be awsome to see a MX100 killer 🙂
No supercaps on this model ???
Hmmm, now I’m wondering if this V-NAND has power issues and can’t be driven by supercap level current if the need arises..
You gave a 5 star for performance of the Sandisk Extreme Pro.
And Samsung 850 Pro is MUCH! FASTER and MUC MORE POWERFUL! than the Sandisk Extreme Pro.
But you give only 4.5 star for the performance of Samsung 850 Pro.
I do not trust the Honesty of your reviews any more.
After updating to Magician 4.4 last night and enabling RAPID, my 840 EVO reads at nearly 8GB/s. Again, not sure if it’s real-world noticeable, but maps seem to load PFQ.
My 1TB EVO 840 was only $400, I don’t see paying the $320 or so difference for this drive. It’s not worth the extra money.
Where did you get the figures of 3W for active and .4W idle power consumption?
Official site is mum on the specifications so far.
That would have come from documentation in our possession and received from Samsung.
What would be of great interest to many is the result of enterprise benchmarks (those used for 845DC Pro) run on 850 Pro that was formatted (overprovisioned) to match 845DC Pro capacity – i.e. to 400GB or 800GB.
If benchmark results are comparable, it seems possible to save tons of money by using 850 Pro in a more read-oriented server environments (lack of supercaps can be dealt with at the system – not SSD – level).
Any chance you can run those tests?
A 850 Pro 512 GB (with Rapid mode) or a M2 XP941 512 GB would be the most performant option for a new X99 build?
Absolutely the better of the two is the XP941 as it gives you that performance all the time and full throttle. RAPID is a caching program and acts only as such in certain scenarios.
Why is “The performance of this Crystal DiskMark result is the highest we have ever had, bar none”? Most of the results are lower than the 840 Pro, by almost half in the sequential results:
It would be nice if the results were all clear-cut (e.g. drive A is superior to drive B in every rating in every test) instead of the rock-paper-scissors game, but that’s the way it was for my 840 Pro pick to begin with.
Ummm… are you aware that you are comparing a single SSDs results to that of an SSD in RAPID mode which uses DRAM to increase performance? Totally different concept…
That would explain the difference. The original 840 Pro review displays scores that are still higher than the 850 Pro’s by some measurements, but no where near the RAPID scores. The 850 Pro has some of the RAPID scores posted near the end of the interview, but does not display the results of every test following enabling RAPID.