Personal computing discussed

Moderators: renee, morphine, Steel

 
slushpuppy007
Gerbil In Training
Topic Author
Posts: 9
Joined: Fri Jun 10, 2016 2:40 am

SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 12:55 pm

Hi,

When I read the reviews on the new NVMe SSDs, like the Samsung 960 Pro, I see that the windows boot times and load times for games are pretty much the same.

Like 0.5 seconds difference between a SATA SSD and a NVMe SSD, sometimes the SATA SSD even beats the NVMe Drive with game / application load times, and its got only 550MB/s Seq Read vs 3200MB/s Seq Read.

Can anyone explain the technical detail on why this is the case?

Reading some large game media files seems like a Sequential Workload, and the NVMe Drive should shine here.

Thanks
Slushpuppy007
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 1:05 pm

When the drive gets fast enough that the time required to access the data is comparable to the time required to process the data, speeding the drive up even more doesn't help load times much because you're waiting on the CPU, not the drive.
Nostalgia isn't what it used to be.
 
chuckula
Minister of Gerbil Affairs
Posts: 2109
Joined: Wed Jan 23, 2008 9:18 pm
Location: Probably where I don't belong.

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 1:11 pm

In addition to what JBI said, during a boot sequence there might be a delay due to the drive initialization process that occurs when the drive is first turned on that nullifies the I/O speed advantage that occurs later in the bootup sequence.
4770K @ 4.7 GHz; 32GB DDR3-2133; Officially RX-560... that's right AMD you shills!; 512GB 840 Pro (2x); Fractal Define XL-R2; NZXT Kraken-X60
--Many thanks to the TR Forum for advice in getting it built.
 
slushpuppy007
Gerbil In Training
Topic Author
Posts: 9
Joined: Fri Jun 10, 2016 2:40 am

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 1:58 pm

okay, thanks, that makes sense.

Assuming the game loading process is single threaded, having a "faster" single threaded performance cpu will give you an edge in game loading speed, rather than a faster SSD?
 
Waco
Maximum Gerbil
Posts: 4850
Joined: Tue Jan 20, 2009 4:14 pm
Location: Los Alamos, NM

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 2:15 pm

slushpuppy007 wrote:
okay, thanks, that makes sense.

Assuming the game loading process is single threaded, having a "faster" single threaded performance cpu will give you an edge in game loading speed, rather than a faster SSD?

Game load times can be significantly improved by multiple fast cores (depends on the game, not all of them are terrible single-threaded beasts).
Victory requires no explanation. Defeat allows none.
 
K-L-Waster
Gerbil Elite
Posts: 576
Joined: Thu Feb 12, 2015 8:10 pm
Location: Hmmm, I was *here* a second ago...

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 2:40 pm

It all comes down to "where are your bottlenecks?" For operations like OS loading and game loading, when using a spinning HDD the storage subsystem was a major bottleneck and the CPU spent most of its time waiting to receive data. Moving to a SATA SSD alleviates much of that wait time. 

The results we see here are consistent with SATA SSDs being "fast enough" to provide the CPU with data sufficiently quickly that the CPU is no longer waiting around. Once you get to that speed, further data transfer increases no longer provide a significant benefit because data transfer isn't what is holding you back anymore. Going to NVME for that use case doesn't provide a significant speed increase because if anything the storage is now waiting for the CPU to request more data rather than the other way around. 
Main System: i7-8700K, ASUS ROG STRIX Z370-E, 16 GB DDR4 3200 RAM, ASUS 6800XT, 1 TB WD_Black SN750, Corsair 550D

HTPC: I5-4460, ASUS H97M-E, 8 GB RAM, GTX 970, CRUCIAL 256GB MX100, SILVERSTONE GD09B
 
Krogoth
Emperor Gerbilius I
Posts: 6049
Joined: Tue Apr 15, 2003 3:20 pm
Location: somewhere on Core Prime
Contact:

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 2:49 pm

Waco wrote:
slushpuppy007 wrote:
okay, thanks, that makes sense.

Assuming the game loading process is single threaded, having a "faster" single threaded performance cpu will give you an edge in game loading speed, rather than a faster SSD?

Game load times can be significantly improved by multiple fast cores (depends on the game, not all of them are terrible single-threaded beasts).


Not likely going to happen. Loading data onto memory tends to be a heavily serial workload. You are not going to see that much of a benefit from parallelization.
Gigabyte X670 AORUS-ELITE AX, Raphael 7950X, 2x16GiB of G.Skill TRIDENT DDR5-5600, Sapphire RX 6900XT, Seasonic GX-850 and Fractal Define 7 (W)
Ivy Bridge 3570K, 2x4GiB of G.Skill RIPSAW DDR3-1600, Gigabyte Z77X-UD3H, Corsair CX-750M V2, and PC-7B
 
jihadjoe
Gerbil Elite
Posts: 835
Joined: Mon Dec 06, 2010 11:34 am

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 3:13 pm

A fast CPU with lots of threads helps a lot. Game resources are often stored compressed, and decompression is a task that scales well with both core count and frequency. Tom's did a test on this a couple of years back by disabling cores and scaling clocks on a CPU. The results are pretty interesting. Most games scaled well both with increasing core count and frequency, but one game (Supreme Commander 2) peaked out pretty early, showing only an improvement going from 1 to 2 cores, but not at all beyond that.

Storage, OTOH has a pretty minimal effect once your disk is able of reading fast enough to keep the CPU saturated as it decompresses assets and executes code. Techspot did some extreme testing on this subject, comparing SSDs to RAID-0 SSDs, and finally to a RAMDrive which is literally 100s of times faster than the fastest SSDs, and according to their report even the RAMDrive didn't make much difference to the load times.
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 3:32 pm

jihadjoe wrote:
... RAMDrive which is literally 100s of times faster than the fastest SSDs ...

Not true.

Even under ideal conditions, a RAMDrive on a system with dual-channel DDR4-2133 RAM has a hard ceiling at 17 GB/sec (total RAM bandwidth is 34 GB/sec but you only get half of that because accessing the data causes it to be copied from one location in RAM to another).

The Samsung 960 Pro that was just written up on the front page has a sequential read speed of 3.5 GB/sec.

So the difference is less than a factor of 5.

Another way to look at it: Today's high-end SSDs are faster than the system RAM of 15 years ago.
Nostalgia isn't what it used to be.
 
thecoldanddarkone
Minister of Gerbil Affairs
Posts: 2449
Joined: Wed Mar 26, 2003 4:35 pm

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 5:06 pm

Most people have pointed this out, most loading is cpu limited.  In my simple testing some operations are so CPU limited that the difference between quad channel memory @2133 and my intel aic 1.2TB is insignificant in day to day operations.  I felt the same way with the 950 pro, so I assume the 960 pro would be the same. 
I7 4930k, 32 GB Ballistix DDRL3@2133 , 1.2 TB Intel 750 AIC, 500 GB mx200, Sapphire R9 Fury, asus x79 ws, HP ZR24w, edifier s730
HP Pro x2 612- i5-4302Y, 8 gigs of memory, 256 ssd
 
ClickClick5
Gerbil Elite
Posts: 581
Joined: Fri Aug 15, 2008 8:28 pm
Location: Alaska
Contact:

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 5:17 pm

I once ran Skyrim on a WD Green drive, WD Black drive, an OCZ ssd, and from RAM. I then tried different loading points to get a metric of how fast the game loads on these devices. The HDDs were the slowest obviously, while the SSD was a massive jump over the HDDs. Yet the RAM only shaved off about a second or two vs 10+ seconds.

The CPU is going to be the bottleneck with these fast drives. Just that simple.

EDIT: Found it: http://i.imgur.com/u2OO3z2.png
The computer was rebooted before each drive change to clear any preloaded content. 
i7-5930k (4.1GHz), 32GB DDR4-2400, 2080Ti, GA-X99-UD3, Samsung 860 Pro 256GB (os/programs), Toshiba 5TB (games), WD VelociRaptor 500GB (games/scratch disk), Win 10 Pro x64
Check out my site for trance sets and other goodies! clickclick5.com!
 
Waco
Maximum Gerbil
Posts: 4850
Joined: Tue Jan 20, 2009 4:14 pm
Location: Los Alamos, NM

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Wed Oct 19, 2016 8:17 pm

Krogoth wrote:
Waco wrote:
slushpuppy007 wrote:
okay, thanks, that makes sense.

Assuming the game loading process is single threaded, having a "faster" single threaded performance cpu will give you an edge in game loading speed, rather than a faster SSD?

Game load times can be significantly improved by multiple fast cores (depends on the game, not all of them are terrible single-threaded beasts).


Not likely going to happen. Loading data onto memory tends to be a heavily serial workload. You are not going to see that much of a benefit from parallelization.

Wrong, but okay. The only operation being done is not just loading data into memory...if it was, it would scale pretty nicely with better IO subsystems. :)

just brew it! wrote:
jihadjoe wrote:
... RAMDrive which is literally 100s of times faster than the fastest SSDs ...

Not true.

Even under ideal conditions, a RAMDrive on a system with dual-channel DDR4-2133 RAM has a hard ceiling at 17 GB/sec (total RAM bandwidth is 34 GB/sec but you only get half of that because accessing the data causes it to be copied from one location in RAM to another).

The Samsung 960 Pro that was just written up on the front page has a sequential read speed of 3.5 GB/sec.

So the difference is less than a factor of 5.

Another way to look at it: Today's high-end SSDs are faster than the system RAM of 15 years ago.

If there's any amount of latency affecting things, RAM drives are hundreds of times faster than even the best NVMe drives, as you can essentially saturate the bus even with small low-queue depth workloads.
Victory requires no explanation. Defeat allows none.
 
Krogoth
Emperor Gerbilius I
Posts: 6049
Joined: Tue Apr 15, 2003 3:20 pm
Location: somewhere on Core Prime
Contact:

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Thu Oct 20, 2016 1:38 am

Waco wrote:
Wrong, but okay. The only operation being done is not just loading data into memory...if it was, it would scale pretty nicely with better IO subsystems. :)


Not every process and operation can be easily parallelized and there's an hefty opportunity cost on writing proper multi-thread code (more manhours spend on testing and QC). There's a reason why most mainstream application are still single and dual-threaded at most. There is simply isn't enough benefits to justify the extra cost.
Gigabyte X670 AORUS-ELITE AX, Raphael 7950X, 2x16GiB of G.Skill TRIDENT DDR5-5600, Sapphire RX 6900XT, Seasonic GX-850 and Fractal Define 7 (W)
Ivy Bridge 3570K, 2x4GiB of G.Skill RIPSAW DDR3-1600, Gigabyte Z77X-UD3H, Corsair CX-750M V2, and PC-7B
 
slushpuppy007
Gerbil In Training
Topic Author
Posts: 9
Joined: Fri Jun 10, 2016 2:40 am

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Thu Oct 20, 2016 4:31 am

Is it possible that an Intel Kaby Lake CPU (i7 K-Series) + Optane SSD could show significantly better load times than what we are seeing with the current Samsung 960 Pro?
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Thu Oct 20, 2016 6:37 am

Waco wrote:
If there's any amount of latency affecting things, RAM drives are hundreds of times faster than even the best NVMe drives, as you can essentially saturate the bus even with small low-queue depth workloads.

I'd be a little surprised if the difference is hundreds of times even for latency-sensitive workloads. You're still treating it as if it was a disk, and operating through an API and device driver stack that is emulating a block I/O device. So although you don't have the overhead of sending commands to the drive and waiting for it to respond, you do still have quite a bit of software overhead for each request. It's also likely to be more CPU intensive, since the CPU is moving the data around itself instead of offloading the data movement to a hardware DMA engine in the disk interface.

A few months ago at the day job we were evaluating a PCIe-based SSD card. Much to our surprise, the performance of the SSD was actually better than a RAMDisk on certain workloads. (The testbed we were using was a ~4 year old server using DDR3 RAM; I expect that the RAMDisk would've won on newer hardware, but the result was still a bit of an eye-opener, and we initially thought we had screwed up the test somehow.)
Nostalgia isn't what it used to be.
 
Chrispy_
Maximum Gerbil
Posts: 4670
Joined: Fri Apr 09, 2004 3:49 pm
Location: Europe, most frequently London.

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Thu Oct 20, 2016 7:25 am

ClickClick5 wrote:
I once ran Skyrim on a WD Green drive, WD Black drive, an OCZ ssd, and from RAM. I then tried different loading points to get a metric of how fast the game loads on these devices. The HDDs were the slowest obviously, while the SSD was a massive jump over the HDDs. Yet the RAM only shaved off about a second or two vs 10+ seconds.

The CPU is going to be the bottleneck with these fast drives. Just that simple.

EDIT: Found it: http://i.imgur.com/u2OO3z2.png
The computer was rebooted before each drive change to clear any preloaded content. 

That's the sort of real-world sanity-check that I like to see. I do the same sort of thing at home and at work and I'm quite happy to pass on the latest NVMe storage stuff I see.

I used to refer people back to a screenshot I made testing a drive on SATA2 vs SATA3 when everyone was getting all upset about not having SATA3. Real world usage was almost indistinguishable even though the REALLY BIG NUMBERS were doubled for the SATA3 tests. QD1 random reads are by far the largest part of an SSD's performance and even the incredible new 2TB Samsung 960 cannot saturate SATA1's 150MB/s bandwidth yet for small random reads in a consumer workload.
Congratulations, you've noticed that this year's signature is based on outdated internet memes; CLICK HERE NOW to experience this unforgettable phenomenon. This sentence is just filler and as irrelevant as my signature.
 
Waco
Maximum Gerbil
Posts: 4850
Joined: Tue Jan 20, 2009 4:14 pm
Location: Los Alamos, NM

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Thu Oct 20, 2016 9:50 am

Krogoth wrote:
Waco wrote:
Wrong, but okay. The only operation being done is not just loading data into memory...if it was, it would scale pretty nicely with better IO subsystems. :)


Not every process and operation can be easily parallelized and there's an hefty opportunity cost on writing proper multi-thread code (more manhours spend on testing and QC). There's a reason why most mainstream application are still single and dual-threaded at most. There is simply isn't enough benefits to justify the extra cost.

I'm well aware there are costs, but you're ignoring a massively complicated set of parameters and sweeping them under the rug by saying game loading is single threaded and "only loading data into memory". That's not reality. Game assets have to be decompressed, data structures built, etc, and these all affect loading times even if the thread doing I/O is "just loading things".

just brew it! wrote:
Waco wrote:
If there's any amount of latency affecting things, RAM drives are hundreds of times faster than even the best NVMe drives, as you can essentially saturate the bus even with small low-queue depth workloads.

I'd be a little surprised if the difference is hundreds of times even for latency-sensitive workloads. You're still treating it as if it was a disk, and operating through an API and device driver stack that is emulating a block I/O device. So although you don't have the overhead of sending commands to the drive and waiting for it to respond, you do still have quite a bit of software overhead for each request. It's also likely to be more CPU intensive, since the CPU is moving the data around itself instead of offloading the data movement to a hardware DMA engine in the disk interface.

A few months ago at the day job we were evaluating a PCIe-based SSD card. Much to our surprise, the performance of the SSD was actually better than a RAMDisk on certain workloads. (The testbed we were using was a ~4 year old server using DDR3 RAM; I expect that the RAMDisk would've won on newer hardware, but the result was still a bit of an eye-opener, and we initially thought we had screwed up the test somehow.)

Sure, there's overhead, but in my experience tmpfs/ramfs are much much faster than even the best PCIe or NVMe storage options when it comes to access latency. Throughput? That's a much closer call, but with a mixed workload, tmpfs tends to be faster in pretty much every measure. Streaming GB/s with 4K random IO is trivial in RAM and reasonably difficult to do with flash. It's not hundreds of times faster in practice, but the underlying storage is never the bottleneck, so it's much easier to hit peak throughput and be entirely CPU limited.

It's hard to compare 1:1 unless you're running the same filesystem, but I don't think I've ever seen a good reason to use anything other than tmpfs for RAM drives.
Victory requires no explanation. Defeat allows none.
 
Krogoth
Emperor Gerbilius I
Posts: 6049
Joined: Tue Apr 15, 2003 3:20 pm
Location: somewhere on Core Prime
Contact:

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Thu Oct 20, 2016 5:29 pm

Waco wrote:
I'm well aware there are costs, but you're ignoring a massively complicated set of parameters and sweeping them under the rug by saying game loading is single threaded and "only loading data into memory". That's not reality. Game assets have to be decompressed, data structures built, etc, and these all affect loading times even if the thread doing I/O is "just loading things".


Most of that stuff in mainstream applications and games are serial in nature and it doesn't yield much in returns if try to parallelized it. If it was so simple to parallelize and it yield massive benefits then why almost every piece of software that parallelized focuses on massive number crunching with workloads that easily be broken-up? Reality itself is you claim is much more complicated.

Parallel computing isn't some magical wand that makes everything 2x as you keep throwing more cores onto it. People are so desperate on trying to keep the "Moore's law" meme alive when it has been dead for almost a decade now.
Gigabyte X670 AORUS-ELITE AX, Raphael 7950X, 2x16GiB of G.Skill TRIDENT DDR5-5600, Sapphire RX 6900XT, Seasonic GX-850 and Fractal Define 7 (W)
Ivy Bridge 3570K, 2x4GiB of G.Skill RIPSAW DDR3-1600, Gigabyte Z77X-UD3H, Corsair CX-750M V2, and PC-7B
 
Waco
Maximum Gerbil
Posts: 4850
Joined: Tue Jan 20, 2009 4:14 pm
Location: Los Alamos, NM

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Thu Oct 20, 2016 6:54 pm

Krogoth wrote:
Waco wrote:
I'm well aware there are costs, but you're ignoring a massively complicated set of parameters and sweeping them under the rug by saying game loading is single threaded and "only loading data into memory". That's not reality. Game assets have to be decompressed, data structures built, etc, and these all affect loading times even if the thread doing I/O is "just loading things".


Most of that stuff in mainstream applications and games are serial in nature and it doesn't yield much in returns if try to parallelized it. If it was so simple to parallelize and it yield massive benefits then why almost every piece of software that parallelized focuses on massive number crunching with workloads that easily be broken-up? Reality itself is you claim is much more complicated.

Parallel computing isn't some magical wand that makes everything 2x as you keep throwing more cores onto it. People are so desperate on trying to keep the "Moore's law" meme alive when it has been dead for almost a decade now.

Well this is my day job, but sure, whatever you say. :)

If you want to think a single core is all that matters for loading times, by all means, continue to do so. Here, a quick little sample of why you're wrong from 6 years ago: http://www.tomshardware.com/reviews/gam ... 38-14.html
Victory requires no explanation. Defeat allows none.
 
Krogoth
Emperor Gerbilius I
Posts: 6049
Joined: Tue Apr 15, 2003 3:20 pm
Location: somewhere on Core Prime
Contact:

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Thu Oct 20, 2016 8:53 pm

Waco wrote:

If you want to think a single core is all that matters for loading times, by all means, continue to do so. Here, a quick little sample of why you're wrong from 6 years ago: http://www.tomshardware.com/reviews/gam ... 38-14.html


LMAO

#1 - using THG as a source

#2 - the test system in article was using a bloody HDD not a SSD. .

#3 - They didn't even bother to overclock on "pseudo-" 2-core and 1-core setup and the most important item you have completely overlooked. When they did overclock the full i5-750 chip. It made a significant gain on some of those "games" despite being a HDD more so then just trying more cores at the damn thing (if it made a difference). I'm willing to beat you also see similar gains on gimping that i5 to 2-core and 1-core.

Hopefully you can see why THG is not really held in high regards within the enthusiast circles.
Gigabyte X670 AORUS-ELITE AX, Raphael 7950X, 2x16GiB of G.Skill TRIDENT DDR5-5600, Sapphire RX 6900XT, Seasonic GX-850 and Fractal Define 7 (W)
Ivy Bridge 3570K, 2x4GiB of G.Skill RIPSAW DDR3-1600, Gigabyte Z77X-UD3H, Corsair CX-750M V2, and PC-7B
 
Waco
Maximum Gerbil
Posts: 4850
Joined: Tue Jan 20, 2009 4:14 pm
Location: Los Alamos, NM

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Thu Oct 20, 2016 9:33 pm

Krogoth wrote:
Hopefully you can see why THG is not really held in high regards within the enthusiast circles.

I'm well aware. My point stands, and you've provided no evidence to the contrary thus far.
Victory requires no explanation. Defeat allows none.
 
anotherengineer
Gerbil Jedi
Posts: 1688
Joined: Fri Sep 25, 2009 1:53 pm
Location: Northern, ON Canada, Yes I know, Up in the sticks

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Sat Feb 11, 2017 5:06 am

About this.

Does or has anyone upgraded from a good 2.5 sata SSD to an NMVe on a newer CPU??  Just wondering if there is a human perceivable noticeable real world improvement??

I am planning on building a new machine probably this summer, and I don't know if I should pay the premium to get NVMe or a larger capacity 2.5 SATA SSD??

I basically use the PC as a web-surfer and occasional gaming, with some occasional AutoCAD 2015.  Just wondering if it's worth it??
Life doesn't change after marriage, it changes after children!
 
JustAnEngineer
Gerbil God
Posts: 19673
Joined: Sat Jan 26, 2002 7:00 pm
Location: The Heart of Dixie

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Sat Feb 11, 2017 8:04 am

There are at least six different combinations to consider:
SATA+AHCI - 2½"
SATA+AHCI - 1.8"
SATA+AHCI - M.2
PCIe+AHCI - M.2
PCIe+NVMe - M.2
PCIe+NVMe  - PCIe card

SATA+AHCI drives in the M.2-2280 form factor have reached price parity with their 2½" versions (led by the Crucial MX300 drives).  When you start trying to keep track of SLC vs. MLC vs. TLC and 3D vs. 2D, it gets even messier.  Here are some current prices for ½ TB SSDs:

2½" SATA:
Sandisk X400 - $260/TB
Crucial MX300 - $273/TB
Plextor M6S - $275/TB
WD Blue - $280/TB
Mushkin Eco2 - $293/TB
Samsung 850 Evo - $320/TB
AData XPG SX930 - $375/TB
Intel 540s - $375/TB
AData SU900 - $396/TB
Samsung 850 Pro - $465/TB

M.2 SATA:
Crucial MX300 - $278/TB
WD Blue - $280/TB
Sandisk X400 - $289/TB
Mushkin Atlas Vital - $328/TB
Kingston SSDNow G2 - $333/TB
Samsung 850 Evo - $336/TB
Intel 540s - $383/TB

M.2 PCIe:
Intel 600p - $332/TB
AData XPG SX8000 - $449/TB
Plextor MBPe - $469/TB
Samsung 960 Evo - $500/TB
Kingston Predator - $625/TB
Samsung 960 Pro - $643/TB
· R7-5800X, Liquid Freezer II 280, RoG Strix X570-E, 64GiB PC4-28800, Suprim Liquid RTX4090, 2TB SX8200Pro +4TB S860 +NAS, Define 7 Compact, Super Flower SF-1000F14TP, S3220DGF +32UD99, FC900R OE, DeathAdder2
 
strangerguy
Gerbil Team Leader
Posts: 262
Joined: Fri May 06, 2011 8:46 am

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Sun Feb 12, 2017 1:16 pm

slushpuppy007 wrote:
Hi,

When I read the reviews on the new NVMe SSDs, like the Samsung 960 Pro, I see that the windows boot times and load times for games are pretty much the same.

Like 0.5 seconds difference between a SATA SSD and a NVMe SSD, sometimes the SATA SSD even beats the NVMe Drive with game / application load times, and its got only 550MB/s Seq Read vs 3200MB/s Seq Read.

Can anyone explain the technical detail on why this is the case?

Reading some large game media files seems like a Sequential Workload, and the NVMe Drive should shine here.

Thanks
Slushpuppy007

With a typical SATA3 SSD the bottleneck on load times is now the underlying programming itself of the app and OS.  It's pretty obvious with early 2000s games; on modern CPUs and SSDs they still do not load instantly.
8700K 4.3GHz @ 1.05V | Cryorig H7 | MSI Z370M AC | 32GB Corsair LPX DDR4-3200 | GTX 1070 @ 0.8V | 500GB Evo 850 | 1TB M550 | 3TB Toshiba | Seasonic G650 | Acer XB271HU
 
credible
Gerbil XP
Posts: 499
Joined: Fri Oct 07, 2011 7:47 pm

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Sun Feb 12, 2017 3:19 pm

anotherengineer wrote:
About this.

Does or has anyone upgraded from a good 2.5 sata SSD to an NMVe on a newer CPU??  Just wondering if there is a human perceivable noticeable real world improvement??

I am planning on building a new machine probably this summer, and I don't know if I should pay the premium to get NVMe or a larger capacity 2.5 SATA SSD??

I basically use the PC as a web-surfer and occasional gaming, with some occasional AutoCAD 2015.  Just wondering if it's worth it??


If I had to do it over I would go with the larger drive as you mentioned, I put this in mine and while fast I don't believe it was worth the money.

On the other hand I put a 960GB each in my boys new computers and they seem plenty fast to me.

Btw I am using an i7-6700k


http://www.canadacomputers.com/product_info.php?cPath=179_1229_1296&item_id=090002
 
credible
Gerbil XP
Posts: 499
Joined: Fri Oct 07, 2011 7:47 pm

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Sun Feb 12, 2017 3:20 pm

Btw, we both use our pc's for essentially the same thing.
 
whm1974
Emperor Gerbilius I
Posts: 6361
Joined: Fri Dec 05, 2014 5:29 am

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Sun Feb 12, 2017 3:33 pm

So from reading all of this, a good SATA SSD is good enough for most folks, including us most of the time? I have long thought that was the case.
 
Dposcorp
Minister of Gerbil Affairs
Posts: 2771
Joined: Thu Dec 27, 2001 7:00 pm
Location: Detroit, Michigan

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Sun Feb 12, 2017 6:28 pm

We talked about PCI-E drives (NVMe versus AHCI) here:

viewtopic.php?f=5&t=117642&p=1300971&hilit=nvme#p1300971
 
MOSFET
Gerbil XP
Posts: 373
Joined: Fri Aug 08, 2014 12:42 am

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Sun Feb 12, 2017 7:00 pm

anotherengineer wrote:
About this.

Does or has anyone upgraded from a good 2.5 sata SSD to an NMVe on a newer CPU??  Just wondering if there is a human perceivable noticeable real world improvement??

I am planning on building a new machine probably this summer, and I don't know if I should pay the premium to get NVMe or a larger capacity 2.5 SATA SSD??

I basically use the PC as a web-surfer and occasional gaming, with some occasional AutoCAD 2015.  Just wondering if it's worth it??


I have done this recently. TL;DR would be that for Windows, an SSD with caching software like Crucial or Samsung is the best option for desktop use. Best speeds, best boot times, best app load times, best value, least wear and tear of flash. If that drive costs less in a 2.5" formfactor, get it. It's still more portable overall and is less likely to throttle, overheat, and get otherwise damaged. (spills, screwdrivers, static) Long-winded anecdotes ahead.

Main boot+system drives for about the last 5 years have been a series of Crucial m4, MX100, and a few MX200s (one is m.2 SATA), spread through desktop (MX100 250GB), laptop (MX200 500GB after a Samsung 830 128GB), and NUC (m.2 SATA 250GB). Basically, I have a lot of practical experience with SATA3 SSDs. Desktop has been FX-8320 until recently, when I upgraded to Kaby Lake i5-7600K, and a Samsung 960 EVO 500GB. As awesome as the new NVMe SSD is, the bigger upgrade was the CPU, and probably secondarily was DDR4-2400 from DDR-2133 (Intel vs old AMD IMC is just no freaking contest). The NVMe SSD is an upgrade, but desktop workloads (yes, even us power users) just don't saturate it. It's like octo-channel memory for a Skylake or Kaby Lake i7-x700K. You can benchmark the streaming throughput, but it's hard to notice. Yes, 4K I/O is faster than SATA3 with the Samsung drives, but not always on some of the cheaper NVMe drives now. One game I like is Cities Skylines, and loading a map fills up a good chunk of my 32GB RAM, so it is safe to that faster drives do this operation faster - but I need a stopwatch to tell. CPU and DDR4 were still the bigger upgrade.

IF I was putting together a new NUC (or upgrading a laptop) and budget was a concern, I'd get the taller NUC and a 2.5" drive - and put the savings toward a gadget like a USB3 NIC or DisplayPort hub or whatever. If you want to build a clean-looking desktop beast that's set for the next 3-5 years, then the value proposition of the recent Sammy PRO/EVOs goes up.

One other quick scenario while we're on the topic. For the past few months I've had an all-flash vSAN across 3 hosts in a homelab. Two retired FX-8320s and an Opteron 6328 (also Piledriver and octocore, similar) all with 32GB RAM. All with 10GB Intel NICs and a Quanta LB6M 10GB SFP+ switch. Each FX-8320 system had a 480GB Plextor/Lite-On PCIe AIC NVMe 480GB SSD caching a 1TB Samsung 850 EVO 2.5" SATA3 SSD. The Opteron is in a 1U rackmount with PCIe riser limiting me to one PCIe slot, which is taken by the 10GB NIC, so no NVMe via PCIe. Also, the old-ass AMD server chipset doesn't have SATA3 like the 990FX boards of the FX systems, so it's storage subsystem should be considerably slower since I'm using the old laptop Samsung 830 128GB SSD on SATA2 to cache an SK Hynix 500GB TLC SSD on SATA2.

Reality was that the Opteron's vSAN storage subsystem, which was half the size, with a quarter of the cache, and all limited to SATA2 with no PCIe boost, often outperformed the FX systems' larger NVMe+SATA3 combination. Also, on the FX's, the NVMe cache tier just never outperformed the SATA3 850 EVO storage tier - so having seen my particular vSAN setup in action, I've decided to rethink my homelab strategy. 1) I think the native ESXi 6.5 NVMe driver is not very good. 2) I think the Plextor m8pe series drives, even with the giant heatsink, are not very good. 3) Having setup, tested, played around with, and benchmarked vSAN, I have removed it and migrated the VMs I'd like to keep over to the NUC with the m.2 SATA MX200. The 3 AMD hosts were great for testing but I cannot justify 600W vs 15W any longer than necessary!

One more? My office PC is kind of the management hub of a small business - it loads a lot on startup, and pegs while Dropbox and Dashlane (in particular) start up. And then I load up a bunch more stuff. The Richland A8-6600K APU at 4.4GHz is rarely idle but settles down and copes. I did some minor experimentation to attempt to get faster boot times and possibly a less-loaded CPU. Starting point is an Intel 530 180GB SSD plus Toshiba 2TB 7200rpm HDD (mainly Dropbox and some backups). Attempt #1 - clone system to one of the Plextor PCIe AICs. Boot times unaffected - responsiveness during boot unaffected. Hmm...tried migrating to an FX-8320 for 8 cores instead of 4, plus L3 cache. Unaffected, except for a much hotter office - instead of reaching 100% CPU during boot, the system reached 50%, but wasn't much more responsive. In the end, the best combo seemed to be Richland with a dGPU and a Crucial drive (the MX100 from my home desktop that was replaced by the 960 EVO) with their caching software. I tried an 850 EVO, but Samsung Magician would not enable caching with the SSD attached to an AMD SATA/AHCI controller. I managed to eventually fool it by moving the EVO to the ASMedia controller, booting, installing Magician with caching, then moving the drive back to the AMD controller. It worked, but ultimately I did not need to boot from a 1TB SSD at work - so I settled on the MX100. I also swapped the 2TB 7200rpm Toshiba for a 1TB VelociRaptor, which made Dropbox's boot-up indexing a lot snappier.
Be careful on inserting this (or any G34 chip) into the socket. Once you pull that restraining lever, it is either a good install or a piece of silicon jewelry.
 
slushpuppy007
Gerbil In Training
Topic Author
Posts: 9
Joined: Fri Jun 10, 2016 2:40 am

Re: SATA SSD vs NVMe SSD (similar boot and load times)

Mon Mar 20, 2017 2:50 am

Looking at this slide from the Optane Marketing Material: https://www.servethehome.com/intel-opta ... -nand-ssd/

At Queue Depth 1, the drive is sitting just under 100 000 IOPS.

Loading a game surely falls in this Low QD spectrum.

Any thoughts?

Who is online

Users browsing this forum: No registered users and 1 guest
GZIP: On