How realistic are SSD benchmarks for gamers?

[StrangerGuy]

exception: does not help with stupid games like BF4 which refuse to use physical memory (when there is ample) and want to revert back to page file on ssd.



back to the original ssd question. once you get to a certain level of ssd performance (~300MB/s). any higher performance become diminishing.

weakest performance link is the 4k read. even a 950 pro is still marginal. that is where performance needs to be focus.

Where's the proof that 4K read is a real-world bottleneck other than that's the lowest number you can find in a synthetic benchmark?

 

[UaVaj]

there u go... look at the 4k access.

http://www.gamersnexus.net/guides/1577-what-file-sizes-do-games-load-ssd-4k-random-relevant



also - ever wonder why it takes ~8sec to boot and not instant boot. cause most windows files are 4k read. just imagine if 4k were in the 300mb/s.

 

[BFG10K]

there u go... look at the 4k access.

http://www.gamersnexus.net/guides/1577-what-file-sizes-do-games-load-ssd-4k-random-relevant

This shows nothing relevant in the real-world. In reality there's no difference between SATA and NVMe with virtually any game's load time.

Again, I showed RAM disk tests earlier, which is the upper limit of performance. No storage system will beat RAM.

Also everyone harps on how Shadow of Mordor takes so long to load because of I/O, when this is absolutely untrue:

Ironically the SATA EVO is faster than all of the NVMe drives. Again, once you hit any SSD (including old SATA 2 variants), the vast majority of games are no longer I/O bound.

also - ever wonder why it takes ~8sec to boot and not instant boot. cause most windows files are 4k read.

Nope, it's because most of it is POSTing which has nothing to do with I/O. Most of the rest is kernel/driver/user code being loaded into the RAM, executing, and acquiring resources such as networking and peripherals.

Why don't you show us these massive improvements in boot times you expect with NVMe in the chart below?

Half a second boot improvement from NVMe, so clearly I/O isn't the problem. Ironically the 750 takes the longest of the bunch because its firmware adding to POST times.

Here's the entire page: http://techreport.com/review/29221/samsung-950-pro-512gb-ssd-reviewed/4

 

[Deders]

Just pointing out that the thread title is about SSD's in general, not specifically NVMe.

 

[UaVaj]

anyone have system with 96gb+ ram? how about a ram test loading current generation game (bf4 and/or shadow or mordor).

then compare the results to say a 950, then a 850 pro or a 850 evo, then a slower sata2 ssd.

 

[Deders]

anyone have system with 96gb+ ram? how about a ram test loading current generation game (bf4 and/or shadow or mordor).

then compare the results to say a 950, then a 850 pro or a 850 evo, then a slower sata2 ssd.

Making sure they restart each time or it will just be reading from memory.

 

[StrangerGuy]

This shows nothing relevant in the real-world.

I find it amazing and increasingly disturbing at the same time that a concept so painfully obvious can not be understood by so-called hardware enthusiasts.

 

[UaVaj]

I find it amazing and increasingly disturbing at the same time that a concept so painfully obvious can not be understood by so-called hardware enthusiasts.

there is a measureable difference should your specific task requires it.

reality is that such difference is irrelevant to the typical everyday task.



car terms. since we everyone knows cars.

on the track. a 10second car is faster than a 11second car. races are won. undisputed.

on the street. both 10 second and 11second car are faster than you ever want to go. even a 12second car is fast.

 

[beginner99]

Most important point of SSD is random read and writes. That's what makes it feel fast and snappy compared to a HDD. It's what makes the PC start up faster. Therefore for normal use you will not see much difference between SSDs in a "gamer workload".

I had an Intel G2 80 GB for years. Note that it has pretty bad sequential write, slower than many HDDs. Anyway after upgrading to an MX100 512 gb, no noticeable performance improvement.

Therefore I would only use the benchmark to avoid especially crappy drives. Personally I would also avoid TLC drives. Get the cheapest MLC version. That will usually do the trick.

 

[StrangerGuy]

nevermind.

 

[StrangerGuy]

The point is anything can be made to look mighty impressive when conditions are sufficiently contrived. For example, a network adapter that has a latency of 0.1ms vs one with 0.001 ms to the router. The latter technically has a 1000x speed advantage in a tiny part of chain but is absolutely worthless in the face of the ISP latency.

 

[therealnickdanger]

More often than not, I find the impact of an SSD vs HDD to be entirely game dependent. My system has 32GB RAM and my GTX 970 has 3.54GB VRAM. Plenty of fast RAM to load assets to.

I play a few MMOs like Rift, Marvel Heroes, and WoW. Not only do those games consistently launch faster from my SSD game directory, but warping to different regions is faster, bringing up large inventory is faster, and little hitches, frame rate dips, and stutter completely disappear no matter how many other players are in the area. When I play these games off the HDD game directory, it is a terrible experience (in comparison). There are so many assets that just simply aren't loaded into memory until they are needed - no matter how much memory you have.

I think games that are designed around the lowest common denominator (2-4GB RAM, slow HDD, weak CPU) are the games that will truly shine using SSDs because the game is built around bottlenecks rather than expecting gamers to build around them with superior hardware. Also, some game engines are built entirely around streaming textures from a console DVD drive or hard drive, where RAM is limited, so when those games are ported to PC, the behavior of the engine consumes all the IO it can get.

Meanwhile, other games like Battlefield only see improved loading times, but not much else. From my own experience, SSDs always end up being better for gaming, the degree to which depends on the game.