Info DirectStorage 1.1 benchmark

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

psolord

Golden Member
Sep 16, 2009
1,875
1,184
136
Guys, if I am not mistaken, copying the folder elsewhere and running the .exe, runs the bench from THAT drive, right?

If so, here are my two runs, on my 8600k+GTX 1070, on the Gammix S11 1TB (placed on a pcie card) and my old SSD EVO 850 512MB.





I was running msi afterburner osd while running the bench and I noticed the gpu usage really did jump to 100% for a second or so. I wonder if it will run on my GTX 970 and Radeon 7950.

Quoting myself, to add some more results (including the quote's).











12400f+3060ti+Gigabyte Aorus 7000s 1TB (nvme pcie 4.0) = 13.30GB/sec
12400f+3060ti+Crucial MX500 1TB (sata)= 2.05GB/sec
8600k+1070+Gammix S11 pro 1TB(nvme pci 3.0)=1.96GB/sec
8600k+1070+Samsung EVO 850 512GB(sata)=1.91GB/sec
2500k+Radeon 7950+Patriot P210 256GB(sata)=1.34GB/sec

Regarding the 8600k+GTX 1070, son, I'm disappoint. Dunno why, I was expecting better. Maybe because in general these two systems provide me with equal gaming experience, but at different resolutions (different houses). It seems the 1070 is the culprit, but I am having some tiny reservations, because that nvme is on a pcie 3.0 card, which is controlled by the chipset and not the cpu. It benchmarks OK with convensional storage benchmarks though, so I guess it's the gpu.

The MX500 and EVO results, both being sata and so close, I think means that a sata drive is good enough up to a GTX 1070. Also, the MX500 is limited by itself on the 3060ti, while the Gammix is limited by the 1070 I think. I'll see on the next primary gpu upgrade, which will upgrade all systems by one tier up (3060ti goes in the 8600k system etc).

Very surprised with the 2500k+7950 system. I didn't expect it to even run. But it did and it was not too shabby for a 12yo cpu and a 11yo gpu. Also for whatever reason, the avocado scrolling was the smoothest of all systems. Have no idea why.

I tried to run it on the work's i5-8400 system (only igp), but it doesn't even run. GPUz says Direct X 12_1 compatibility while the 7950 is DX12_0, so it can't be it.

Next and final step, will be a GTX 970 testing.
 
Last edited:
  • Like
Reactions: DAPUNISHER

psolord

Golden Member
Sep 16, 2009
1,875
1,184
136
Yeah, well so much for GTX 970 testing. It doesn't run at all. Beats me why. Thought maybe it was MSI Afterburner because it is setup differently on that system. Closed it, still nothing.

I know it's not the Sandy, or the old GTX 970 or windows version, since all these apply for the 7950 system and it run there. So let's see if any new version comes out later on.
 

shotgunl

Junior Member
May 10, 2006
5
6
81
Yeah, well so much for GTX 970 testing. It doesn't run at all. Beats me why.

Interesting...Well, at least it's not just me where the bulkload demo doesn't work with a GTX 970 (paired with an i3-10100f, MSI B460M-A PRO, 16GB DDR4-2400, and a 512GB WD SN730 NVMe drive...also tried from two SATA SSDs: 500GB Samsung 850 EVO and 2TB Leven JS600. It fails with a buffer overflow exception, so maybe it's the the segmented 3.5GB/.5GB VRAM?

I don't know. It does work on Maxwell parts though because it works on a laptop I have with a GM108 Geforce 940M + i5-5200U + 16GB DDR3 + 480GB Micron M500 SATA SSD:
Screenshot_20230124_072123.png
 

psolord

Golden Member
Sep 16, 2009
1,875
1,184
136
Well, it seems a Forspoken PC demo will be coming today on all PC stores. So I guess we can check direct storage loading there.

According to DF, It drops even to 720p in performance mode....on the PS5 and also 1080p and/or 20fps in quality mode. So the 970 should be able to do 720p/10fps low, lol.
 

SteveGrabowski

Diamond Member
Oct 20, 2014
6,806
5,769
136
Well, it seems a Forspoken PC demo will be coming today on all PC stores. So I guess we can check direct storage loading there.

According to DF, It drops even to 720p in performance mode....on the PS5 and also 1080p and/or 20fps in quality mode. So the 970 should be able to do 720p/10fps low, lol.

They're claiming almost a 50% drop in load times on an M.2 SSD vs a SATA SSD thanks to Direct Storage.

 

Ranulf

Platinum Member
Jul 18, 2001
2,331
1,139
136
1.7s vs 3.7s. Nice, I guess.

From a 4790k, 2060S system on a newer TLC 1TB sata ssd (not OS) with 200Gb or so empty space. This was first load of the demo. Subsequent runs had lower cpu usage. 1.27 GB/s, 4.28 GB loaded in 3.36s.

dirstor1b.png
 

shotgunl

Junior Member
May 10, 2006
5
6
81
Well, it seems a Forspoken PC demo will be coming today on all PC stores. So I guess we can check direct storage loading there.

According to DF, It drops even to 720p in performance mode....on the PS5 and also 1080p and/or 20fps in quality mode. So the 970 should be able to do 720p/10fps low, lol.

Here's the loading times with the Forspoken demo that I got with the GTX 970 + i3-10100f, MSI B460M-A PRO, 16GB DDR4-2400 CL16, 512GB WD SN730 NVMe SSD, on Windows 11.
Because...why not?

With the preset on Ultra-High, texture memory maxed at Ultra-High, and Model memory maxed on High, the loading time on the NVMe drive is about 8 and 1/12 of a second.

With the same ultra-high settings, but instead with the game on a Samsung 850 Evo SSD, the loading time is about 12 and 2/3 seconds.

And, again with the same ultra-high settings, except using a 4TB Seagate SSHD HDD (5900RPM + 8GB MLC cache), the loading time is about 19 and 5/6 seconds.
https://www.youtube.com/watch?v=fPsDD_sU2zI

And as a comparison, with the preset on low and model and texture memory also set on low, the loading time on the NVMe drive is about 4 and 1/3 seconds.
https://www.youtube.com/watch?v=SO3o63KYQm8


The title screen background scene also takes noticeably longer to load on the SATA SSD and much, much longer to load on the traditional HDD.

Not DirectStorage related, but on the GTX 970 with render scale at 100% and DRS and FSR 2.0 + TAA shutoff (they are enabled by default with low preset) at 720p, it'll bounce between just above 30fps down to about 24fps just in the very beginning area. At 320x240 and 33% render scale with everything at the lowest, you can sometimes hit just over 50fps at near 100% gpu utilization. ha.
 

psolord

Golden Member
Sep 16, 2009
1,875
1,184
136
I'm really glad that you guys posted some real life examples.

I did manage to do a quick test on the forspoken demo myself. I will post two videos and a reminder this is a hobbyist channel, non monetized, not clickbaiting.

I did a test on my 8600k+1070, although I used custom settings, with something the system could actually run.


At the very beginning, after I show the settings (some of them were on high mind you), I load a save and the game starts in like 1 second. And IT IS actually loading. I have the MSI Afterburner OSD measuring the HDD1 throughput, which is the nvme on that system. It is the first time I have seen the nvme of that system going so far up to 1GB/sec reading, during gaming. Moreover it loads like crazy during gameplay, you can clearly see it showing hundreds of MB/sec at times.

Therefore I have to take back my "son,I'm disappoint" meme comment. I am not disappoint at all. I am actually impressed! And also the game did not run that bad, for how heavy it is, in conjunction with these specific settings. Considering it's a 2016 card, it's doing great.


===========

For the kicks of it, I did a test on my 2700k+GTX 970, lol.


Same format, first showing the settings, then load the same save. It loads really fast on this old geezer as well. Not 1 sec fast, but fast. You can again see the MSI Afterburner OSD hitting 500MB/sec speed, which is good. This is the first time I have used the SSD of this system for gaming though. It could be the same on other games. However you see it though, it is admirable that a system with what is essentially a 12yo architecture, manages to have all this world setup and ready in 3-4 seconds, even at low. I will try to do more testing in the future, with a better video card and higher settings.

As for the game itself, yeah even at low, the GTX 970 is no good for it, for PC standards. It hits vram limits hard and the pcie 2.0 bus, ain't helping.

==============

For now I can't wait to test this on the 12400f+3060ti. And I think I will do three tests there, nvme pcie 4.0, sata ssd and a hard drive just for laughs. I will try to run it on the 7950 system as well. I think it is higher level than DX12_0 though, but I'll try anyway.

Thank you Square Enix for providing this demo. I wish more publishers provided demos for everything. I will actually buy it when it gets discounted, just for providing the demo.
 

Borealis7

Platinum Member
Oct 19, 2006
2,914
205
106
shocking! SSDs and M.2s are faster than HDDs! who would've thunk it???
can DS be turned off for testing? maybe disabled in the OS level?
maybe test with a graphics card that doesn support DS? if the game can even run on those cards)
 

psolord

Golden Member
Sep 16, 2009
1,875
1,184
136
It seems there is a performance hit with nvmes.

form pcgh and announced by videocardz


 

Leeea

Diamond Member
Apr 3, 2020
3,599
5,340
106
It seems there is a performance hit with nvmes.

form pcgh and announced by videocardz
Interesting.

Would like to see further reporting before assuming that is true. The results do not feel right.


That said, it kind of makes sense. No free lunch and all that, if the GPU is doing the work rather then the CPU the resources have to come from somewhere.


Definitely a wait for further reporting from my viewpoint. Why would SATA require less GPU processing / decompression? Yea, it is going through the chipset, but work is work.
 
Last edited:
  • Like
Reactions: psolord

utahraptor

Golden Member
Apr 26, 2004
1,052
199
106
I am hearing the 10% hit was a false positive. Some are speculating frame rates were recorded during load screens which are higher during those periods. Since the sata SSD has longer load times they have a lot of bonus high frame rate times raising the total average. 10% in this case. More testing is needed.
 

DeathReborn

Platinum Member
Oct 11, 2005
2,743
734
136
I am hearing the 10% hit was a false positive. Some are speculating frame rates were recorded during load screens which are higher during those periods. Since the sata SSD has longer load times they have a lot of bonus high frame rate times raising the total average. 10% in this case. More testing is needed.

This might be to what you refer:

 

Stuka87

Diamond Member
Dec 10, 2010
6,240
2,559
136
Got around to trying this.

5800X3D
Asrock X570 Steel Legend
32GB DDR4 3600 (8x4)
RX 5700 XT


I ran it on two drives. The NVME drive was of course way faster, but it also used double the CPU.

Samsung EVO 970 Plus 1TB (NVME):
1674951267970.png

Intel 730 480GB (6gb SATA):
1674951304797.png
 
  • Like
Reactions: psolord

solidsnake1298

Senior member
Aug 7, 2009
302
168
116
Mostly a ~1 second difference in load times according to ComputerBase. What a nothing-burger.

Ironically even mechanical HDDs are getting similar benefit.
I thought the point of direct storage wasn't JUST load times. But also loading assets, for example, while you are turning around. 100% no way a mechanical HDD can keep up with the low latency that kind of usage would demand.
 

psolord

Golden Member
Sep 16, 2009
1,875
1,184
136
Following my previous post, I did three more tests with forspoken, on the 12400f+3060ti system. Unfortunately I did not do a gaming test with the sata ssd, but I did a hard disk one, which is very interesting and is more than enough I think.

I am posting the hard disk one fist, since it is the most interesting. (all vids non monetized, just a hobbyist channel since 2006).




I used FSR on this one too, to see how it runs compared to dlss. I used ultra preset, then enabled fsr and all texture and memory settings are maxed. So the HDD has a lot of work to do. The hard disk reading is shown in HDD3 string of MSI Afterburner's OSD. You can see it hovering up to 100MB/sec at times and other times lower than that. But the fact remains that it is mostly OK.

===============

The other one is a run on the same system, but this time from the nvme. Of course the stage loads in like half a sec.


The primary difference with the HDD run, is that it goes up to 1.3GB/sec for a half a sec or so, the stage loads very fast as I said, but instantly after that I see lower chunks being loaded, while the HDD was loading more. They are not exactly the same runs, they are just on the same broader area, but you will get an idea.

==================

The third run has to do with resolution and DLSS setting, so it's irrelevant for this thread.
 
  • Like
Reactions: shotgunl

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
Mostly a ~1 second difference in load times according to ComputerBase. What a nothing-burger.

Ironically even mechanical HDDs are getting similar benefit.
I think you're ignoring the important aspect here, the main benefit of DirectStorage is moving the compute part of texture decompression from CPU to GPU. Remember all the talk about Spider Man Remastered hammering CPUs? AFAIK part of that was due to texture decompression running on the CPU.

The fact that HDDs can keep up with this implementation of DS shows how scalable the tech is, and it's good news for people who are still running games from slower drives. In fact I hope the PC implementation will make good use of the RAM available on a PC, in the sense that games might be able to leverage excess RAM as more (compressed) texture cache. This could potentially relieve pressure from the other parts of the chain (disk speed, VRAM load).

1675072104344.png
1675072118514.png
 
  • Like
Reactions: Leeea and Makaveli

BFG10K

Lifer
Aug 14, 2000
22,674
2,824
126
I thought the point of direct storage wasn't JUST load times. But also loading assets, for example, while you are turning around. 100% no way a mechanical HDD can keep up with the low latency that kind of usage would demand.
This has already been done and failed, ALA megatextures in Rage. Even a console DVD could dynamically stream those assets, but even today's fastest NVMe drives still have visible texture pop-in, and it's certainly not because of a high CPU load.

Streaming is garbage because it absolute destroys how CPU caching works. Not to mention that people seem to have this strange idea that streaming from NVMe drives is somehow magically faster than storing it locally in much faster (V)RAM.

It's like putting a pagefile on NVMe instead of buying more RAM, then calling it progress because "it's much faster than old HDDs doing the same thing!"
 
Last edited: