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

[Admin]

German news outlet PC Games Hardware measures the performance penalty from Direct Storage with a GeForce RTX 4090 graphics card.

DirectStorage Causes 10% Performance Hit On RTX 4090 In Forspoken : Read more

 

[TJ Hooker]

All the results in the table in this article are listed as using DirectStorage. How can you can say the difference in results is from DS, when all the results use DS?

Maybe it's explained in the video, but I don't speak German.

 

[nimbulan]

All the results in the table in this article are listed as using DirectStorage. How can you can say the difference in results is from DS, when all the results use DS?

Maybe it's explained in the video, but I don't speak German.

I believe the assumption is that DirectStorage requires an NVMe drive, so wouldn't be used with the SATA setup. But when they first showed off DirectStorage with this game, they showed DirectStorage numbers for NVMe, SATA SSD, and even an HDD. Really makes me wonder if there's something else going on.

 

[Markis82]

Sorry to say but this not actually due to DirectStorage directly, but something else. The reason is because Forspoken does not actually use DirectStorage 1.1. Dev's confirmed it did not have the GPU decompression aspect. Whether performance took a hit due to indirect reasons either coincidence or indirectly as a byproduct of assets swapping or some other behavior would require more analysis.

 

[Kamen Rider Blade]

Maybe the Texture Decompression used the GPU Compute aspects of the 4090 since there was no dedicated ASIC for that specific Decompression step that it needed to do.
Ergo the loss in performance.

 

[Yakumo_unr]

Forspoken uses DirectStorage 1.0, it does not have GPU Decompression, that came later in DirectStorage 1.1.

The DirectStorage SDK has a benchmark sample you can build, it does run on a SATA drive, it gives no indication that doing so is doing anything other than using DirectStorage but just slower than an NVME.

 

[KyaraM]

All the results in the table in this article are listed as using DirectStorage. How can you can say the difference in results is from DS, when all the results use DS?

Maybe it's explained in the video, but I don't speak German.

Yeah, he touches on it in the first couple minutes in the video. What they did was testing different types of SSDs, under the assumption that, for example, a SATA-SSD doesn't support it. He also says they cannot actually make out how much is from DirectStorage and how much from the type of SSD, since it cannot be turned off. Have to watch further to get to the part where he talks about FPS.

Edit:
The SATA-SSD got over 80 FPS, the fast, DirectStorage-drives a bit less.

 

[PlaneInTheSky]

We shouldn't draw conclusions from a game that was horribly unoptimized.

Couldn't get 60fps in 1080p on my 3060. Refunded.

 

[Amdlova]

10% hit on 4090 the 4070ti will hit 45%?

 

[KyaraM]

10% hit on 4090 the 4070ti will hit 45%?

Could you maybe not throw random numbers around? Thanks.

 

[Amdlova]

Could you maybe not throw random numbers around? Thanks.

Damage control?

 

[serafin152]

Doing tests and judging performance on such unoptimized game is just wrong.

 

[KyaraM]

Damage control?

I'm not working for NVIDIA, SquareEnix or Microsoft, and I have absolutely no clue what your problem is. All I know is that you made a needlessly caustic comment and threw around imaginated numbers for absolutely no reason that people who don't know much about the subject might take seriously. All I want is for the discussion to stay factual. Not even PCGH can say what the reason is and admits that they couldn't test it fully. It's also completely needless and senseless, so please just not do that.

 

[jp7189]

10% is a lot at the top end. People spend big money chasing that last couple of FPS. That it hit the 4090 so hard has me worried about lesser cards. Although this may be more due to game or driver optimization than hardware limits. Who knows?

It would be interesting to see a full review of various GPUs, and of course, another game or two.

 

[FunSurfer]

Direct Stuttetorage! Get it now! HDD Stuttering Performance on an NVMe SSD! Feel the Nostalgia! Only Today! At 150% Normal Game Price! Get it Now!!

 

[evdjj3j]

All the results in the table in this article are listed as using DirectStorage. How can you can say the difference in results is from DS, when all the results use DS?

Maybe it's explained in the video, but I don't speak German.

I wonder if we won't see direct storage accelerators built into future SSDs.

 

[evdjj3j]

I believe the assumption is that DirectStorage requires an NVMe drive, so wouldn't be used with the SATA setup. But when they first showed off DirectStorage with this game, they showed DirectStorage numbers for NVMe, SATA SSD, and even an HDD. Really makes me wonder if there's something else going on.

Nope. Tom's has already done a DS article in which the performance of a SATA drive is measured.

https://www.tomshardware.com/news/directstorage-performance-amd-intel-nvidia

 

[TJ Hooker]

I can't find any reliable source for whether the game uses DS 1.1 (with GPU decompression) or 1.0, but as others have pointed out, DS very much works with SATA SSDs (and even HDDs, as per previous Forspoken benchmarks no less). If PC Games Hardware is assuming that it only works on NVMe then I guess they just don't know what they're talking about, and none of the sites reporting their results bothered to question it.

 

[bit_user]

Leaving aside the various problems noted on this specific benchmark...

Someone should run a GPU profiling tool to examine what happens to the compute load on a GPU that's using GPU-based decompression, in DirectStorage 1.1. I've said it before and I'll probably say it again: GPUs aren't good at decompression. It's not very SIMD-friendly, so I expect it'll use a disproportionate amount of the GPU's resources. However, I'll wait until we have a clean benchmark and then we'll know for sure.

 

[bit_user]

I wonder if we won't see direct storage accelerators built into future SSDs.

Wouldn't that defeat the point? The idea seems to be sending compressed assets across PCIe, into host memory, and into GPU memory. In this case, you don't want to decompress them too early.

 

[Markis82]

10% hit on 4090 the 4070ti will hit 45%?

No, lets not exaggerate.

Users have been testing it on Steam forums reporting their results. Almost all of them are reporting a loss of FPS and stuttering issues when using -noDirectStorage. Interestingly, some are also reporting blurry textures are resolved by disabling DirectStorage.

Now, none of them I see mentioning their tests are checking vs a SATA like in the linked article. The linked article tests different drives and both SSD result in reduced performance while the SATA gets improved. This means it is unlikely to be simply DirectStorage because these results make no sense unless the SATA is loading data assets in so slowly its causing low quality temporary place holder assets during the transition to result in a boosted FPS. This does not appear to be DirectStorage oriented. They also don't test with DirectStorage with the SATA for reference.

 

[KyaraM]

I can't find any reliable source for whether the game uses DS 1.1 (with GPU decompression) or 1.0, but as others have pointed out, DS very much works with SATA SSDs (and even HDDs, as per previous Forspoken benchmarks no less). If PC Games Hardware is assuming that it only works on NVMe then I guess they just don't know what they're talking about, and none of the sites reporting their results bothered to question it.

What I see mentioned pretty much everywhere is that it's DS 1.0, not 1.1.

 

[InvalidError]

The DirectStorage SDK has a benchmark sample you can build, it does run on a SATA drive, it gives no indication that doing so is doing anything other than using DirectStorage but just slower than an NVME.

Well, what is DirectStorage in the first place? Nothing more than a streamlined pipeline to load data from storage to system memory with reduced file system and OS overhead, then from system memory to wherever else it is needed. The only real difference between SATA and NVMe besides bandwidth and latency is that with SATA, it is the SATA controller handling DMA instead of NVMe. One extra abstraction layer but otherwise fundamentally the same.

It would be quite inconvenient for game developers to have to write completely different loading and asset streaming code depending on storage device type and performance.

 

[Chung Leong]

Read the update, people. Story turns out to be completely bogus.

 

[KyaraM]

Read the update, people. Story turns out to be completely bogus.

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.