News DirectStorage Causes 10% Performance Hit On RTX 4090 In Forspoken

[InvalidError]

Not really bogus, just not thought through to the end before publication. It does make quite a bit of sense, though, that it would be that way.

Counting "very high fps during loading black screens" in a benchmark is a very silly omission and letting fps surge in load screens is a silly waste of power and processing time on the game engine's part.

 

[KyaraM]

Counting "very high fps during loading black screens" in a benchmark is a very silly omission and letting fps surge in load screens is a silly waste of power and processing time on the game engine's part.

Did I say with a single word it was impeccable testing or something? No, I didn't. I literally said they didn't think it through to the end. As in, something was missing from their test terminology. A better way to do the testing would indeed be to filter out blackscreen FPS. It makes sense that longer blackscreens at higher FPS would skew results towards the slow SATA drive. I thought it was obvious from my post that I think they missed something, but apparently not.

 

[Friesiansam]

Counting "very high fps during loading black screens" in a benchmark is a very silly omission and letting fps surge in load screens is a silly waste of power and processing time on the game engine's part.

Given those factors, noted in the article, the results are, effectively, completely meaningless.

 

[InvalidError]

Did I say with a single word it was impeccable testing or something? No, I didn't. I literally said they didn't think it through to the end.

I wasn't meaning to get into an argument, merely restating the edits for people who jump in here later without reading the update.

 

[PEnns]

I would rather wait a 3-4 more seconds for the game to load than having my (very expensive GPU these days) take a performance hit!!

Having said that, the jury is still out on dozens of other games yet to be tested.

 

[drivinfast247]

It's not just about initial loading times, it is also to help in the seamless transition between different areas of a game.

 

[InvalidError]

I would rather wait a 3-4 more seconds for the game to load than having my (very expensive GPU these days) take a performance hit!!

Enjoy your longer 300fps loading screens. Longer static loading screens rendering at stupid high FPS inflating the SATA's stats is where SATA's higher scores in this benchmark come from.

 

[RichardtST]

Lol! So, in some arbitrary game, that might not even use DS, because it renders black frames very fast, and when loading from a slow drive there are naturally more black frames while it loads, the data came out exactly backwards! Nice! Could we get some people testing this that aren't amateurs? Seems like the games that need this sort of technology are just horribly badly written. Perhaps teaching the programmers how to program and fixing the game would be a better path. There is no excuse for loading delays with today's hardware. Bad programming. Plain and simple.

 

[bit_user]

Not really bogus,

Huh? It's not outright fraudulent, but "bogus" doesn't necessarily imply an intent to deceive.

From The Jargon File (version 4.4.7, 29 Dec 2003) :

  bogus
   adj.

      1. Non-functional. 'Your patches are bogus.'

      2. Useless. 'OPCON is a bogus program.'

      3. False. 'Your arguments are bogus.'

      4. Incorrect. 'That algorithm is bogus.'

      5. Unbelievable. 'You claim to have solved the halting problem for Turing
      Machines? That's totally bogus.'

      6. Silly. 'Stop writing those bogus sagas.'

It's completely bogus, as it seems the eye-catching results were caused by the high-framerate loading screens displaying for a longer time, on the SATA drive. Their test was fundamentally flawed, as it produced badly biased data. Bogus.

Did I say with a single word it was impeccable testing or something? No, I didn't. I literally said they didn't think it through to the end. As in, something was missing from their test terminology.

It's not just that it wasn't impeccable, it's that the error was probably an order of magnitude higher than any signal they might've been measuring for. That's a broken testing methodology. I'm sorry if you don't like to see people be called out on malpractice, but defending this makes you seem like a person not concerned with data, and therefore risks discrediting other arguments and claims you make.

Good data must be sacrosanct.

If you want to say something nice, you could simply credit the publication for coming out and owning their mistake.

 

[bit_user]

letting fps surge in load screens is a silly waste of power and processing time on the game engine's part.

Really? Black screens is a silly waste of power & processing time? It'd be silly if it used an appreciable amount of either, but that's not a given.

What would be silly is to waste development time solving a non-issue, if black screens indeed don't use appreciable processing time or power (as I expect).

 

[InvalidError]

Really? Black screens is a silly waste of power & processing time?

I don't have the game to check for myself but if it is coded like the bulk of games out there, the main rendering thread is likely pegged at 100% CPU usage spamming black frames and the GPU likely has also non- trivial usage from just doing that.

 

[bit_user]

if it is coded like the bulk of games out there, the main rendering thread is likely pegged at 100% CPU usage spamming black frames and the GPU likely has also non- trivial usage from just doing that.

Um, probably not. For the CPU, a blanking the frame and displaying it probably amounts to putting just 2 commands in a queue. That doesn't take any real time. And once it hits the refresh rate of the display, the thread is likely back-pressured.

As for the GPU, it's only 2 GB/s to memset an entire 1440p image @ 144 Hz. That's only 1% or below their total memory bandwidth. And they have so much more computer power than bandwidth that it's probably less than 0.1% of their compute budget.

If you just think about it, it's the same bandwidth equation to fill an image as it is to display (just writes instead of reads). So, if it would actually use a lot of bandwidth, then so would simply displaying a static screen.

 

[InvalidError]

Um, probably not. For the CPU, a blanking the frame and displaying it probably amounts to putting just 2 commands in a queue. That doesn't take any real time. And once it hits the refresh rate of the display, the thread is likely back-pressured.

Unless they have dedicated code spitting out black frame in a closed loop, those black frames come out of the game engine going through all of the baseline work it normally goes through for in-game rendering before ending up with empty display lists. On the GPU side, I bet the "black screen" still uses far more bandwidth and processing power than it should due to shader post-processing and other silly stuff - you want your black screen correctly anti-aliased and raytraced!

 

[bit_user]

Unless they have dedicated code spitting out black frame in a closed loop, those black frames come out of the game engine going through all of the baseline work it normally goes through for in-game rendering before ending up with empty display lists. On the GPU side, I bet the "black screen" still uses far more bandwidth and processing power than it should due to shader post-processing and other silly stuff - you want your black screen correctly anti-aliased and raytraced!

At this point, one of us should really just fire up the game with some profiling tools, except I don't have a PC which is even capable of running it.

 

[InvalidError]

At this point, one of us should really just fire up the game with some profiling tools, except I don't have a PC which is even capable of running it.

With my GTX1050, the black screens are probably the only thing that would run well on 2GB of VRAM

 

[Deleted member 1353997]

I would rather wait a 3-4 more seconds for the game to load than having my (very expensive GPU these days) take a performance hit!!

While that's reasonable, the updated article suggests the test was meaningless, as it doesn't really prove anything.

Since you probably missed it:

Unfortunately, the publication didn't consider that the Forspoken benchmark contains black screens with high FPS that affect the average frame rate.

The average FPS was higher because the loading screen (which renders really fast) is shown for a longer amount of time, which biases the average FPS.

 

[Chung Leong]

letting fps surge in load screens is a silly waste of power and processing time on the game engine's part.

Probably a consequence of v-sync & co having been turned off. These are numbers generated with non-real-world settings on non-real-world hardware set-up (RTX 4090 on a computer with a SATA SSD) we're dealing with here.