Boss, how can you get 702MB/s loading speed @SATAIII (Game load time performance@page 6) - did I miss something? Thanx in advance, DG21

DG21:

(Game load time performance@page 6) - did I miss something?

Correct, I have noticed that myself as well and have been noticing it ever since a recent Windows 10 update. I am still investigating but it looks like some sort of file-cache is kicking in and hindering that result set. I might need to remove that test completely in the future, but felt I should add it as I always present numbers as they are measured.

great job, as usual HH. hopefully this will drop in price as 2tb (or more) has been prohibitively expensive for home use (i.e. not a business expense).

Hilbert Hagedoorn:

Correct, I have noticed that myself as well and have been noticing it ever since a recent Windows 10 update. I am still investigating but it looks like some sort of file-cache is kicking in and hindering that result set. I might need to remove that test completely in the future, but felt I should add it as I always present numbers as they are measured.

Part of Redstone 4 and the cache to RAM I guess, for a larger game you can see superfetch kicking in via the task manager after you exit and it begins filling up RAM to it should be around 80% or so depending on how much data it can fit and then leaving a little bit of entirely free space left. Successive runs will then be much faster using the cache data but it's possible to empty out the "stand-by" memory now cached via for example this. https://wj32.org/wp/software/empty-standby-list/ And I think there's a few apps on this forum for this very feature too since before the recent cumulative updates for both Redstone 4 and Redstone 5 there was a issue with slowdowns as the cached memory took all the RAM and it had to swap data around instead of leaving a part of it free plus it didn't want to unload unless you ran out of memory but I believe this if not entirely fixed has at least improved now. 🙂 https://forums.guru3d.com/threads/fix-game-stutter-on-win-10-1703-1809.420251/ There should be a Windows blog announcement on the feature too but essentially superfetch and into RAM to speed things up, not too complex but quite effective considering the speed of modern RAM kits. 🙂 (Even compared to a high-end SSD.)

Hilbert Hagedoorn:

Correct, I have noticed that myself as well and have been noticing it ever since a recent Windows 10 update. I am still investigating but it looks like some sort of file-cache is kicking in and hindering that result set. I might need to remove that test completely in the future, but felt I should add it as I always present numbers as they are measured.

I noticed that after recently updating to the latest version of Windows 10 v1809 (previously I was on v1803), that virus scan run times are nearly half of what they were prior to the update. I've got a full SSD system (no HDD), Sandisk Ultra II and Crucial MX300, I was surprised to see those decreased virus scan run times in both Avira Free Antivirus and Malwarebytes. So, don't know if that's related to what you saw there, but figured I'd mention it.

JonasBeckman:

Part of Redstone 4 and the cache to RAM I guess, for a larger game you can see superfetch kicking in via the task manager after you exit and it begins filling up RAM to it should be around 80% or so depending on how much data it can fit and then leaving a little bit of entirely free space left. Successive runs will then be much faster using the cache data but it's possible to empty out the "stand-by" memory now cached via for example this. https://wj32.org/wp/software/empty-standby-list/

Superfetch is deprecated--can't remember when the superfetch service was removed, but it was a while ago. The newer caching scheme (I'm on 1809, build 17763.134) seems to handle things a bit differently. I think I even posted to these forums about it when I noticed it Superfetch was no more.

Redstone 5 was when it got changed but it's still used although instead of calling it Superfetch it's back to being named the same as the service itself so it's called "SysMain" now both under Services and the Task Manager. 🙂 You can see it spin up every so often so it's pretty easy to spot and with the cache function from Redstone 4 (Creators Update.) it'll also run every time you exit a program as it works it into the cache which under the task manager and memory you should see it divided between "In use" "Modified" "Standby" and "Free" with standby being the cached data. There's tools for more thorough viewing of exactly what is cached in the standby memory but when it's required the OS can dump this data or you can empty the data yourself such as with the earlier listed utility or other programs though super fetch should start up immediately to re-optimize the cache memory for any current running software. Earlier the standby memory tended to consume almost all available RAM but after the recent updates it seems to be capped to always leave a part of free memory and it also unloads earlier allowing other running programs to use RAM as needed although I don't know if that's specific to Redstone 5 or if it's been included in the newer cumulative updates for Redstone 4 too by this point so the various software for emptying out this data might still give a performance increase if there's still any issues with how this is cached or how the cache is handled. Well bit of a longer post but even using a SSD Windows 10 should still retain super fetch and with the stand-by memory and using RAM as a cache it should still allow for a speed-up for any subsequent starts of a software as long as the data is still retained in RAM and hasn't been unloaded though I can see this also skewing testing and benchmarking unless the data is emptied out first. 🙂 As for QLC and SSD's I'm glad to see alternatives for storage space and affordability even if QLC brings additional restrictions over TLC though for regular usage it's probably not going to wear out or slow down notably and should still have a edge over a traditional HDD in most cases though I have to do more reading about this so I'm not particularly well versed in how this all works myself so it's something to look up for the next major system build I suppose. Hmm wonder if PCIE 4.0 and something like Optane would be able to compete with DDR but then we also have high-speed DDR4 now and triple and quad channel memory so probably not, only problem here is between AMD and Intel and getting the kits stable on respective hardware but it sounds like AMD has made several improvements already. Well that's not hitting consumer grade hardware just yet I suppose and then devices that could really benefit from PCIE 4.0 and current devices are already capable of hitting Gigabyte speeds. (And improvements on a OS level and software in general taking advantage of high-speed storage devices like SSD's more thoroughly, guess that's also going to be a thing but with time it should improve further.)

Well it looks like my WD "Green" 2TB drives have their spiritual successor...

And how long will data last without power?

Hilbert Hagedoorn:

Meet the NAND storage future, Samsung today outs its first QLC based SATA3 SSD. Join us as we review the all new Samsung 860 QVO 2TB SSD. Don't worry, we'll explain what QLC is, but the bottom line ... Review: Samsung 860 QVO 2TB SSD (QLC NAND)

Hey HH, do you think reliability of big SSDs is enough to assume we can ditch HDDs for gaming purposes (instantaneous textures streaming and blazing fast loading) and keep them for pure storage only?

Really don't trust this QLC memory enough for my data. Wouldn't wanna fill this up either (as you never should with SSD's) Better off buying a 2 to 4 TB HDD for your storage needs where you can actually use the entire drive. Stick to pro's and maybe EVO's for the important stuff.

Hey HH, do you think reliability of big SSDs is enough to assume we can ditch HDDs for gaming purposes (instantaneous textures streaming and blazing fast loading) and keep them for pure storage only?

Not with QLC drives, i wouldn't.

As I write in the article, I do not foresee any issues with reliability based on normal PC usage workloads. Samsungs problem is trust, they should have gone for 5-years warranty (and lower prices). We've seen exactly the same happen with TLC when it was released and at this stage, TLC has become a commonly accepted type of NAND. Bigger is better though for TBW values, with 100 GB per day writes (!) the SSD would last almost 20 years before the last cells are depleted. 2TB SSD is rated at 720 TB 720TB writes / (100GB/day x365) = 19.73 Years. Also if I may throw in an argument, there are plenty of tests on the web where say SSDs rated at 200TB lasted over a petabyte ... then again, 4 bits per cell is new technology, we do not know how it'll behave long term.

Interesting, my old rusty samsung 830 ssd, is almost 7 years alive and kicking. And it was also a 3-year warranty given disk. 🙂 I believe playing some games during the day aren't so tiring for the disk after all. Only when installing and patching you could slightly reduce faster that lifespan in the end. Thanks. I'll keep an eye on the news if something happens after some usage worldwide.

coth:

And how long will data last without power?

it's non-volatile memory. it will last longer than a hdd

tunejunky:

it's non-volatile memory. it will last longer than a hdd

Non-volatile memory easily leaks without sustained power through the time. A month would be enough to get few bad blocks (leaked cells) with non-fresh MLC.

for active storage (like game-storage) this looks good value but for passive storage (archival purpose) i guess i need to wait abit, to see how reliability goes i personally still prever HDD simply because it been tested/proved until now

Hilbert Hagedoorn:

Correct, I have noticed that myself as well and have been noticing it ever since a recent Windows 10 update. I am still investigating but it looks like some sort of file-cache is kicking in and hindering that result set. I might need to remove that test completely in the future, but felt I should add it as I always present numbers as they are measured.

You could use Rammap and cacheset to purge the caches between tests

coth:

Non-volatile memory easily leaks without sustained power through the time. A month would be enough to get few bad blocks (leaked cells) with non-fresh MLC.

i'm sure you have a reason to say that. and i know it's anecdotal, but i've never had that problem. i keep several (full) ssd's and use a hot swap drive often. many times it's over 6 months between drive swaps.

I got a 1TB 840 EVO a little over 4 years ago, $476 back than lol. It has something like 12+TB of writes and still kicking. I replaced it with a 2TB 860 EVO last month, $311. My game saves and important docs are in the cloud so I'm not too worried. At one point my main PC with the 1TB 840 EVO was powered off and unplugged for a full year in storage.

ST19AG_WGreymon:

I got a 1TB 840 EVO a little over 4 years ago, $476 back than lol. It has something like 12+TB of writes and still kicking. I replaced it with a 2TB 860 EVO last month, $311. My game saves and important docs are in the cloud so I'm not too worried. At one point my main PC with the 1TB 840 EVO was powered off and unplugged for a full year in storage.

12TB isn't much, most of these drives are rated for 500TB+ depending on the make, Think my 850 2TB EVO is rated for 2000TB, most normal users will never hit this