Separate names with a comma.
Discussion in 'SSDs & Data Storage' started by Brahmzy, Apr 5, 2018.
Was the 950 EVO an NVMe drive?
I don’t do placebo. Been doing this stuff way too long. Anybody that actually USES a 900p all say the same thing.
The reviewers say the same thing.
You must be dense. Can you not read the link to the anandtech chart showing the qd1 random read performance?
^^ If you cannot understand what’s happening there, you need to go back to school.
Sounds like a perfect plan...
One of the problems is that we don't have a benchmark that directly quantifies all those little UI latencies and stutters that something like an Optane drive is supposed to improve.
It's kind of like how people used to argue about how some video cards seemed subjectively smoother than others, pointlessly arguing over average frame rates -- there was no hard evidence until reviewers started to quantify micro-stutters, because existing benchmarks were showing the wrong metric to answer the question.
I got one. Also had a 960 Pro as a boot drive before.
Can't honestly tell the difference, but i know i would if the 900P had RGB LEDs on it sync'd to IO activity.
Would be ludicrous fast. ... and i mean F A S T . >>>
Lay off the personal attacks.
I can read charts, and I know what that means, but I personally would like to see some real world benchmarks which show the results of those charts. Whether it be starting programs, running some custom hard drive write software or database work. It’s like memory performance, looks good on a chart, but show me the actual performance difference in applications.
Like when having 2 processors went mainstream, and smpy smoothness was a thing discussed. Hard to measure based on subjective experience
I do suspect that is the sort of thing here. The actual performance difference is not going to register highly for real workloads, but it there is a difference folks can perceive and enjoy. More power to them, but this really does seem like "my mclaren will blow away your viper" sorts of stuff. Okay, but both get to the grocery store at the same time.
I wonder if it's like frame-pacing in multi-GPU; the issue (AMD's shit-tastic implementation) was there and reported on, but took near a decade to prove empirically.
Maybe we don't have the right tools to measure the (I'll assume) very real performance difference?
I can only say, as I’ve said above, EVERYTHING is snappier. Browsing, opening apps, running Windows Update inventory (almost all random read), opening Kodi, running Kodi, opening movies and closing movies in Kodi, general TV/DVR usage in Kodi, start up, shut down, reboot, MyMovies SQL DB optimization etc.
Intel is marketing Optane as storage that makes the OS and everyday functions feel faster. This is solely because of low-qd random read performance. They specifically designed it to do what they say it does and it works.
Why on earth is it so hard for some to grasp this concept that has been known as fact since consumer SSDs were first available?
You really need to stop with the tone of "why is it so hard to grasp" stuff, and you'll do much better.
As you know, you can increase random read performance quite a bit, and still not make significant differences in the actual end-user benchmarks. This is what is being discussed. You can make one thing faster, but it's the whole chain that matters. For many or most tasks, there are clearly things besides the raw I/O which are obviously in the critical path. We know this because end user benchmarks do not show anywhere near the same delta as the super-focused synthetic ones. They're faster, but not to a degree where folks such as yourself would stand up and take notice.
Thus, since we believe your subjective evaluation, we are trying to determine a good way of measuring this "feels faster" in a repeatable, deterministic, and objective way.
I don’t need to stop any tone - don’t get so butthurt about other people.
How about you go buy one for yourself?
That’s what I do.. First hand experience.
Low-QD random read has been the holy grail since 2006 when SSDs went mainstream. It is well known FACT (not my subjective opinion) that this is what gives a user a better “experience”. Intel knows this, the industry knows this.
So it does baffle me when you demand proof of something, after seeing the IOmeter graphs, after knowing why low-qd random reads are so important, hear numerous first hand accounts, yet still don’t believe it could possibly be true.
So go freakin buy one!
Or don’t and live in bliss with the other more-than-adequate offerings from Samsung and others.
This thread is painful to read, why do people not understand that low latency will significantly increase performance. Optane is like 5000x faster than Samsung 960 Pro when running SQLite benchmarks. You know what applications that use SQLite a lot? Your web browser!
And there are like billions of other applications and even games that uses SQLite.
The revolution with Optane can be compared with transportation. 10 Busses(Samsung 960 Pro) can move a lot of people at once, but it takes a while to transport all the people on the bus to their respective location.
100 taxis (Intel P3700) can move people to their respective location a lot quicker.
But 1000 teleportation devices(Optane) can move people to their respective location almost instantly.
so post this: https://openbenchmarking.org/result/1802112-TY-900PBYTEE44 ffs
Reviewers should benchmark this!
Could you share the web browser and game benchmarks showing the significantly improved performance from Optane? I haven't run across those.
As for the transportation analogy, I'd suggest some corrections. It would be more accurate to simply compare a Ferrari and a SmartCar for a couple different use cases. Without contention, the Ferrari is at least 10x as fast.
How much faster could the Ferrari complete a lap at Nürburgring?
How much faster could the Ferrari get across downtown LA?
The first scenario is like the storage heavy benchmarks. It allows full utilization of the components, so their relative differences will translate directly into more performance.
The second is like the desktop benchmarks. They are really not storage-bound very often. Thus the amount of time where the faster component could pull away is dramatically limited. Most of the time there's just an irate cardiologist in the Ferrari screaming about traffic.
This is all we're saying. Nobody at all is saying the Optane isn't the faster drive. We're just trying to ascertain how much of it's blistering speed we'd actually get for things we actually do. End user benchmarks produced so far are throughput focused, and do not show a significant difference.
However, since some owners like Brahmzy DO note a considerable difference in experience, it would be interesting to find a way to quantify that difference in experience. I'm pretty sure it's the 10s of ms deltas here and there which aren't going to tilt a benchmark result, but are quite perceptible to the user. But it would be nice to really see those numbers, to allow much simpler communication of potential value for the investment in a top-end part - without expecting everyone to just "go freakin buy one".
I picked up 3 of these when they were ~$160 net for the 280GB.
Should have grabbed 3 dozen.
If you are building a ZFS box, do yourself a favor and grab the tiny m.2 one for your zil/slog. The 32GB is as cheap as $55 at microcenter, if they are oos it might have been me, sorry! You will not wear it out in this kind use case before that computer is totally irrelevant or fails for other reasons.
Its literally as good as having a battery-backed dedicated ramdisk just for sync writes. Obviously turn on sync writes. Intel was totally on the money about the caching drive utility of these, just demoing the wrong platforms.
These are also the obvious goto "SLC" drives for DIY/embedded stuff if write endurance is on your radar at all.
I suggest reading the PCPer review, lots of good detail in there about how the IRQ driver model is a hinderance since 3DXP is so fast (there’s a Linux polling driver that offers a nice performance increase). They also talk about some of the drawbacks of NAND SSDs that 3DXP doesn’t have, like how NAND drives will lock up for a few seconds if you delete a large amount of data.
Then there’s this very limited test that shows the Optane drive outperforming the 960Pro by about 9% - not a lot for the price difference, but not insignificant http://www.legitreviews.com/game-load-time-benchmarking-shootout-six-ssds-one-hdd_204468
I can understand the opinion that it’s not worth the cost for some people, but it is noticibly faster. Installing Windows was comically quick.
This thing is freakin amazing. Bouncing back and forth between 3 W10 boxes, all @ 4.8-5GHz, the 900p box is ridiculous fast.
I see no other choice but to get 900ps for the other 2. Everything else with 960 Pros feel noticeably slower, lol.
You got me thinking now and I am contemplating an 900p since I found out about them
Is there a downside if I get the 280GB U.2 version?
I kinda want to pick one up for my project drive. I have a couple huge compiles that hammer the drive pretty hard for extended periods of time, and I have a feeling the random read perf of the 900P would tear it to shreds.
IMO no downside - it’s the format I’m using.
I can’t afford to burn PCIE slots for the other format, either from room, cooling or whatever.
The U2 size is thicker than a std SSD, but it’s mounting holes line up the same. So if, for instance, you own a Corsair case with the plastics clips for SSDs, the U2 is too thick for those. So just mount it another way. The M2 > U2 cable is plenty long and flexible so no concern there.
Just wish I’da grabbed 2 more when NewEgg was having its $329 sale. Microcenter has em for $339 I think, but then tax.
Wake me when we have a few reviews testing real world scenarios, especially game load times. I can see why one would want the smaller ones for zfs or certain work loads, maybe the 58gb for windows install* but for most this is a waste of money. When the u.2 drive 480gb is the same price as a 1-2tb ssd I'll pay attention.
* I really would hate to have windows these days on anything less than 120gb, really 256gb to be safe.
I’m trying to grab a 16gig to benchmark as a cache drive for my 950 pro, if I can pick it up cheap enough I will benchmark and time it
I don't like the idea of the PCI-E card version and I can place the U.2 drive at a good position behind an 120mm fan in my modded Coolermaster Cosmos II case. I was wondering if there is a performance difference (shouldn't) besides thermal throttling that I can take care of.
No performance difference.
My U2 benches have matched what I’ve seen for the AIC.
I have one of my two 480GB 900P’s connected via a U.2 to PCI-E adapter to maximize PCI-E lane usage and it works well
Great I am planning to get the U.2 280GB version to use it as an OS drive and also take advantage of the U.2 slot in my M/B.
$160 ? That sounds crazy.. What site are we talking about?
People were buying them and selling the Star citizen code when it came out for $150-200.
On a side note, took the plunge recently and if you are really sensitive to the os drive, you can feel it compare to the 960 pro
LOL at the Samsung 970 Pro/Evo release today. 900p still owns the 4k RR.
I've been benchmarking couple 900p Optanes recently. The performance is incredible. ~500k 4k random writes with 32 depth. It is really great and I will use one of those in my system
The new 905ps are about to release...
So I managed to acquire a 32gig 800p yesterday and have been testing it. I will report back soon.
So using the 32 Gig optane as a cache drive with version 3.0.2 of primocache. I am consistently getting about 390-450 mb/sec read and write @ 4k QD1 from the 800p as a cache drive. While it is no 800/900mb/s per the 900p, This setup is a lot cheaper. I looked into potentially raiding two 32gig optanes but my motherboard doesn't support bifurcation so it's not going to work.
Some real life test would be more useful. Seconds saved and such
Ok - give me a piece of software to test?
Don't know a specific benchmark. Test some program/game loading times, but you would need before/after for those.
Before/after is easy - you can just disable cache for the drive at will .
Loading times in general are so quick that one second here or there makes little difference.
Do checksum calculation on a set of large and a set of small files.