Intel Launches its Next-Gen Intel Optane 900p SSD

[Sauron]

Well, now we're talking. I'm still surprised they didn't debut the technology with something like this.

[porina]

11 hours ago, ARikozuM said:

Buying the ~500GB 900p on launch? Please keep the price around the 960 EVO, Intel. 

Keep dreaming. It destroys the 960 Pro for random which is where anyone that really cares about it is looking. If you want sequentials, get the Samsungs.

[Drak3]

5 hours ago, hey_yo_ said:

Which makes me wonder when will AMD make their own M.2 NVME SSD? ?

I can't wait for the AMD 1080P.

[DocSwag]

On 10/27/2017 at 6:24 AM, porina said:

Is the take away point meant to be the difference between those SSDs, particularly in game load times, is basically nothing? Without numbers on hand, I think we're in the ball park of 30MB/s for a budget branded SATA SSD, to maybe 50MB/s for a high end NVMe SSD, when it comes to QD1 random reads. The 32GB optane modules were in the hundreds. It may be questioned how much is due to sequential vs. random, but I'm willing to try this one out myself.

 

I don't play the game much, but when I've done testing with GTAV benchmark, there is a visible difference in dynamic loading stutter between cheap SATA SSDs and a 960 Evo. Right now I mostly game off SATA SSDs due to price and am wondering if that was suboptimal, should have got something faster. I don't have an unlimited budget, but the one in this thread is within reach for testing.

 

I'm thinking here... I also have the FFXIV:Stormblood benchmark, and I think that reports a load time value too. Some of my systems have multiple drives in them, so I could try running it off each and see what if any difference there is. That way system variables would be controlled also. For example, my main system has 1 AHCI, 2 different SATA SSDs, and one HD. My VR system has one NVMe, one SATA SSD.

I really don't know what to tell ya. Everything I've seen indicates very small differences between even the best and worst SSDs out there.

 

It is worth noting though that generally the reason might be these games are generally more tailored towards HDDs, I believe (more compressed stuff which requires better CPU, not storage).

[porina]

1 minute ago, DocSwag said:

I really don't know what to tell ya. Everything I've seen indicates very small differences between even the best and worst SSDs out there.

 

It is worth noting though that generally the reason might be these games are generally more tailored towards HDDs, I believe (more compressed stuff which requires better CPU, not storage).

Since your previous post I have been benching away using FFXIV:Stormblood benchmark. This might seem obscure, but this is what I spend over 90% of my PC gaming time on and is what I want to optimise. This has been an interesting learning time... in the following results, I list the tested drive, capacity, total load time reported by benchmark, and CrystalDiskMark Q1 random read speed result.

 

SATA SSD

Sandisk SSD+ 480GB 17.7s 40.4MB/s

Sandisk Ultra II 960GB 18.2s 31.2MB/s

Crucial MX200 1TB 16.7s 29.3MB/s

Kingston SSDNow 240GB 20.2s 20.1MB/s

 

PCIe SSD

Samsung PM951 256GB 15.5s 45.8MB/s

Samsung SM951 512GB 14.3s 37.9MB/s (AHCI version, not NVMe)

Samsung Evo 960 500GB 30.6s 38.5MB/s (with unknown system problem)

Samsung Evo 960 500GB 16.9s 55.0MB/s (after installing FCU)

Samsung Evo 960 500GB 14.6s 55.0MB/s (overclock CPU to 4.3 GHz from stock 7800X)

 

HD

WD Blue 1TB 56.8s 0.55MB/s

 

 

The benchmark is about 2GB in size. For all of these except the HD, sustained transfers could shift that in seconds. Also note the benchmark specifically resets the system disk cache, so when re-running the bench it has to be reloaded. Early on I repeated the measurements several times in each configuration and saw only a small difference, of the magnitude of a couple % or so, so I didn't do repeats later on unless the results seemed suspect.

 

Roughly speaking, SATA SSDs are on average a bit slower than PCIe ones, which shouldn't be a big surprise. The question then is how much gain for how much speed. The correlation with measured random doesn't seem to be too strong, but I don't know how much balance there is between randoms, sequentials, and non-disk activities. Also each system had different CPUs which might also have some impact.

 

That would be an interesting exercise, the disk cache can be selected to NOT be wiped on re-runs. Reading out of cache (2GB will fit no problem in my ram) should then mostly indicate loading time due to non-disk related activities, like possible CPU decompressing stuff. I think I'll have to do that next. An alternative would be to create a ramdisk, which should be even better than Optane. I'll see how it goes.

 

BTW in case anyone is wondering why I listed the Evo 960 multiple times, the benchmark actually revealed a problem with the system I didn't know I have. Disk benchmarks showed sequential transfer rates were normal, but random was much lower than expected. I tried everything I could think of, and started another thread on it. The root cause was not found, but I was offered FCU on that system this morning, applied it, and since then speeds have been back up to expectations. Overclocking it also seemed to have helped, so that is showing some CPU contribution. I guess ideally I should run all the drives on the same system...

[ARikozuM]

I think we're all forgetting that disks are accessed by games quite often and open-world games would see great benefits if they're sending a good number of files across their protocol.

[DocSwag]

1 hour ago, porina said:

Since your previous post I have been benching away using FFXIV:Stormblood benchmark. This might seem obscure, but this is what I spend over 90% of my PC gaming time on and is what I want to optimise. This has been an interesting learning time... in the following results, I list the tested drive, capacity, total load time reported by benchmark, and CrystalDiskMark Q1 random read speed result.

 

SATA SSD

Sandisk SSD+ 480GB 17.7s 40.4MB/s

Sandisk Ultra II 960GB 18.2s 31.2MB/s

Crucial MX200 1TB 16.7s 29.3MB/s

Kingston SSDNow 240GB 20.2s 20.1MB/s

 

PCIe SSD

Samsung PM951 256GB 15.5s 45.8MB/s

Samsung SM951 512GB 14.3s 37.9MB/s (AHCI version, not NVMe)

Samsung Evo 960 500GB 30.6s 38.5MB/s (with unknown system problem)

Samsung Evo 960 500GB 16.9s 55.0MB/s (after installing FCU)

Samsung Evo 960 500GB 14.6s 55.0MB/s (overclock CPU to 4.3 GHz from stock 7800X)

 

HD

WD Blue 1TB 56.8s 0.55MB/s

 

 

The benchmark is about 2GB in size. For all of these except the HD, sustained transfers could shift that in seconds. Also note the benchmark specifically resets the system disk cache, so when re-running the bench it has to be reloaded. Early on I repeated the measurements several times in each configuration and saw only a small difference, of the magnitude of a couple % or so, so I didn't do repeats later on unless the results seemed suspect.

 

Roughly speaking, SATA SSDs are on average a bit slower than PCIe ones, which shouldn't be a big surprise. The question then is how much gain for how much speed. The correlation with measured random doesn't seem to be too strong, but I don't know how much balance there is between randoms, sequentials, and non-disk activities. Also each system had different CPUs which might also have some impact.

 

That would be an interesting exercise, the disk cache can be selected to NOT be wiped on re-runs. Reading out of cache (2GB will fit no problem in my ram) should then mostly indicate loading time due to non-disk related activities, like possible CPU decompressing stuff. I think I'll have to do that next. An alternative would be to create a ramdisk, which should be even better than Optane. I'll see how it goes.

 

BTW in case anyone is wondering why I listed the Evo 960 multiple times, the benchmark actually revealed a problem with the system I didn't know I have. Disk benchmarks showed sequential transfer rates were normal, but random was much lower than expected. I tried everything I could think of, and started another thread on it. The root cause was not found, but I was offered FCU on that system this morning, applied it, and since then speeds have been back up to expectations. Overclocking it also seemed to have helped, so that is showing some CPU contribution. I guess ideally I should run all the drives on the same system...

So basically...

 

pcie ssds are faster than sata ones, but not by a huge margin. And the cpu is somewhat bottlenecking.

 

That's an interesting point. I guess you're right, optane may end up delivering better performance in real world scenarios... assuming the cpu doesn't bottleneck.

[porina]

5 minutes ago, DocSwag said:

That's an interesting point. I guess you're right, optane may end up delivering better performance in real world scenarios... assuming the cpu doesn't bottleneck.

Still benching. If I enable persisting disk cache (system with 960 Evo) the reported load times goes down to 9.2s average of two runs. Running out of a ramdisk with cache wipe on start gave 9.9s. The ramdisk was pushing out numbers in CDM more like what I'd expect from the 900p, so that might be indicative. Is it worth it for a (potential) 30% reduction in load time?

 

It also shows a fast CPU helps too. I'll try it on my Coffee Lake system later as that's an easy 5 GHz.

 

In case anyone wants to point out that reads will be cached in normal use, the game client is on the ball park of 32GB last time I checked. I don't know how much of that I actually use in a typical session, but it would likely need a ram upgrade to make some impact, and that would also be costly.

[captain_to_fire]

So basically the Intel Optane 900p lasts longer than the Samsung 960 Pro?

[ARikozuM]

10 minutes ago, hey_yo_ said:

So basically the Intel Optane 900p lasts longer than the Samsung 960 Pro?

Are you saying Intel is the better man? 

[captain_to_fire]

17 minutes ago, ARikozuM said:

Are you saying Intel is the better man? 

Yup but the 960 Pro is cheaper and it's not like the 960 Pro is bad. It's actually one of the best.

[DoctorWho1975]

Will the new SSD require a key? Only work on Intel systems?

[leadeater]

12 hours ago, porina said:

-snip-

I had a thought then just as I was about to write the reply I realized it's unrealistic for most people to even test. I was going to say create a RAM disk and test with that since that is about best case as you're going to get, then you can compare between that and SATA SSD etc, but yea total RAM required to do it lol.

 

If I had a spare GPU I could test it out but I don't 

[NumLock21]

So a Intel exclusive game item. There was another game that did something similar where the game enabled more features when it detects a Intel SSD. Forgot what that game was. Came out like 3-4 years ago.

[ARikozuM]

7 minutes ago, NumLock21 said:

So a Intel exclusive game item. There was another game that did something similar where the game enabled more features when it detects a Intel SSD. Forgot what that game was. Came out like 3-4 years ago.

Arizona Sunshine with i7

[NumLock21]

1 minute ago, ARikozuM said:

Arizona Sunshine with i7

That's another game with exclusive items or features? Well the game I saw wasn't that one. It was more cyberpunk/apocalyptic.

[ARikozuM]

10 minutes ago, NumLock21 said:

That's another game with exclusive items or features? Well the game I saw wasn't that one. It was more cyberpunk/apocalyptic.

I don't know then. The only exclusives that I know of is AS and some football VR exclusivity they had this year or last.

[NumLock21]

Just now, ARikozuM said:

I don't know then. The only exclusives that I know of is AS and some football VR exclusivity they had this year or last.

The only thing I remember with that game was, when it detects a HDD, then those jumbotron hanging on the building is turned off. When it detected a SSD, those jumbotrons are turned on and they start to animate. Intel promoted that feature is exclusive to their higher end 500 series SSD, but I have their 300 series SSD. I even went on to download that game demo to try it out to see if my Intel 300 series SSD will work and it does too. Can't find that game demo, must have delete it..

[vanished]

I wonder what thermals are like.  I know some existing high performance SSDs already get hot and this is on another level.  Then again, it's also a totally different tech so maybe it doesn't get hot the same way...

[porina]

4 hours ago, NumLock21 said:

So a Intel exclusive game item. There was another game that did something similar where the game enabled more features when it detects a Intel SSD. Forgot what that game was. Came out like 3-4 years ago.

AMD gave out Star Citizen ships too. I got the one in the link below with some past promotion. Note the logo on the nose.

 

https://robertsspaceindustries.com/pledge/ships/mustang/Mustang-Omega

 

[Coaxialgamer]

Holy f these drives are fast. Finally they are showing the potential of Xpoint. 

[Coaxialgamer]

On 10/28/2017 at 10:11 AM, hey_yo_ said:

Which makes me wonder when will AMD make their own M.2 NVME SSD? ?

Why? Amd doesn't make memory. They've sold drives in the past. All it would be is a rebranded SSD from some OEM. 

[leadeater]

39 minutes ago, Coaxialgamer said:

Why? Amd doesn't make memory. They've sold drives in the past. All it would be is a rebranded SSD from some OEM. 

AMD Intel OEM SSDs, I'd love to see that 

[Zodiark1593]

Considering I haven't jumped on NVME myself, those drives look very appealing.

[Drak3]

2 minutes ago, leadeater said:

AMD Intel OEM SSDs, I'd love to see that 

Inb4 AMD becomes a second source for Intel SSDs, comes up with their own next gen tech that enables far higher bandwidth at similar latency, then pushes out a NVMe 9590 Pro.