-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
DirectStorage 1.2 BulkLoadDemo Benchmark
- Thread starter winjer
- Start date
Thebonehead
Gold Member
Nice.Wanted to come back to this fun thread with my newly assembled system. GEN5 nvme and 5090.
I've just bought a Crucial T705 4TB gen5 which is around14,000 Mbps.
Just waiting for the 5090 in that system at the moment. Scan reckon around March on the priority list.
CamHostage
Member
There's lots to be frustrated about in how game development went this gen, but I'm still surprised how little we've gotten out of the switch to SSDs and rapid IO.
Of course, we've greatly enjoyed how little loadtime there's been, that part has been great and alone has been well worth the upgrade.
But otherwise, games have not been revolutionized by this generational SSD upgrade as promised/expected. Games still have relatively similar asset density (and when they do have a lot of "stuff" in them, the procedural gen does a lot of work,) patches are still massive, only a few novelty concepts of "rifting" have been attempted, and just nothing brand new in game design seems to have come yet or been shown coming from this sea change in drive speed and bandwidth.
Of course, we've greatly enjoyed how little loadtime there's been, that part has been great and alone has been well worth the upgrade.
But otherwise, games have not been revolutionized by this generational SSD upgrade as promised/expected. Games still have relatively similar asset density (and when they do have a lot of "stuff" in them, the procedural gen does a lot of work,) patches are still massive, only a few novelty concepts of "rifting" have been attempted, and just nothing brand new in game design seems to have come yet or been shown coming from this sea change in drive speed and bandwidth.
Last edited:
SHAKEZ
Member
Devs just use new tech as an excuse for bad optimization, so in the end we end up gaining almost nothing from the new tech being used.
CamHostage
Member
Naw, I reject simple notions of "lazy developers" being to blame for all the ills of game production.
Myself, I know from research and talk, from relationships of a few industy workers, from even a tiny bit of experience, that there's way, way, way, way, way more to any aspect of these types of discussions. Sure, developers cut a corner and/or push their luck at times yet then get by with a "Phew, it worked..." relief, but on an industry level (and with as much independent, experimental work is being done out there, albeit much of it in established engines or toolsets which can reinforce the bounding box,) there are people out there looking for the next way through.
It's for sure disappointing, whether it be a mystery or an outrage why.
This advancement of IO specifically, there was such a clear case for what it would solve and offer (going back to the Cerny chats or the MS DS demo.) I honestly expected to see some sparks of innovation come from the transition.
There are unfortunately reasons why the results haven't been revolutionary: there wasn't a similar jump in storage space, there's been consolidation of middleware technology, procedural generation has become the hotter and possibly more promising concept, the money has steered a certain way, the bounds are potentially too big for small experimental offices to explore, the publishing industry is not interested in risks, etc. None of those reasons closes the book on breakthrough play coming from breakthrough IO, but as a key aspect of the generational upgrade, it didn't offer the expected heights in its generational jump.
I'm still holding out hope that some developers out there are still pushing towards realizing the promises that this tech came with when it was first announced, and I'm crossing my fingers that when a developer does come up with a path forward, there's a publisher confident enough to back them for pushing into that void.
Last edited:
Negotiator
Banned
I think it's a shame DirectStorage/GDeflate isn't being used more...
Just found out God of War Ragnarok is 190GB on PC vs 100GB on PS5.
With my slow ADSL connection and the fact I got the disc PS5 version a bit cheaper in BF sales vs the cost of the PC/Steam version, I find it hard to justify a purchase.
I've heard GDeflate is worse than PS5's Kraken decompression... I think PC GPUs (and maybe CPUs too) need to standardize a common decompression algorithm.
Add an ASIC block to perform decompression on-the-fly on both GPUs and CPUs. Kinda like H.264/HEVC/VP9/AV1 for video or SHA extensions in CPUs.
It's crazy we have jumped from 5GB (PS360 era) to 200GB games, but the jump in quality is nowhere near 40 times higher.
Stuff like these make me wish the physical medium should stay a bit longer.
Just found out God of War Ragnarok is 190GB on PC vs 100GB on PS5.
With my slow ADSL connection and the fact I got the disc PS5 version a bit cheaper in BF sales vs the cost of the PC/Steam version, I find it hard to justify a purchase.
I've heard GDeflate is worse than PS5's Kraken decompression... I think PC GPUs (and maybe CPUs too) need to standardize a common decompression algorithm.
Add an ASIC block to perform decompression on-the-fly on both GPUs and CPUs. Kinda like H.264/HEVC/VP9/AV1 for video or SHA extensions in CPUs.
It's crazy we have jumped from 5GB (PS360 era) to 200GB games, but the jump in quality is nowhere near 40 times higher.
Stuff like these make me wish the physical medium should stay a bit longer.
rofif
Banned
The drives capacity and internet speeds stagnated.
I've had 1-2tb drives for past 10 years at least. Before that it was growing each year or two yet we don't all run 512tb drives nowadays.
And the interned speed also capped in my city. 600mb is what I had for past…. 8 years and best they offer today if I wanted an upgrade is 900.
That's not changing that much when you are downloading 200gb game.
Negotiator
Banned
HDD test (7200 RPM, must be reading data from the 256MB cache, since it's close to the theoretical SATA3 speed, while the mechanical speed is no faster than 270MB/s):
SSD test (PCIe 4.0 connected to the CPU, not the southbridge):
Any idea why read speed is 0 MB/s? It should have been 6-7GB/s.
It seems RTX 4070's GDeflate performance is faster than PS5's max Kraken decompression speed (22 GB/s), but it's pointless if games don't utilize it (because devs have to cater to the lowest common denominator)... I don't buy the stutter excuse.
To me this whole situation reminds me of Amiga Commodore (dedicated coprocessors doing the heavy lifting) vs IBM PC (generic processors doing all kinds of stuff, not necessarily in the most efficient manner).
SSD test (PCIe 4.0 connected to the CPU, not the southbridge):
Any idea why read speed is 0 MB/s? It should have been 6-7GB/s.
It seems RTX 4070's GDeflate performance is faster than PS5's max Kraken decompression speed (22 GB/s), but it's pointless if games don't utilize it (because devs have to cater to the lowest common denominator)... I don't buy the stutter excuse.
To me this whole situation reminds me of Amiga Commodore (dedicated coprocessors doing the heavy lifting) vs IBM PC (generic processors doing all kinds of stuff, not necessarily in the most efficient manner).
Negotiator
Banned
The GDeflate compression ratio seems to be 3.75x (1.86GB/s vs 507MB/s), so 23.43GB/s means it must be reading at 6.23GB/s from the NVMe, but I don't see it in Task Manager. Weird...
In 3DMark I get over 40GB/s (would need PCIe 5.0 NVMe to saturate that) from the GPU (I wonder how faster 4090/5090 would be, not that it really matters in PC ports though):
In 3DMark I get over 40GB/s (would need PCIe 5.0 NVMe to saturate that) from the GPU (I wonder how faster 4090/5090 would be, not that it really matters in PC ports though):
Last edited:
Hoddi
Member
DirectStorage reads on NVMe drives aren't reported by Windows' disk counters on Win11 due to BypassIO. If you want to test it without then you can disable it in the registry by searching for StorageSupportedFeatures for your drive and setting it to zero.
YeulEmeralda
Linux User
Technology is useless if nobody uses it.
S0ULZB0URNE
Member
Technology is useless if nobody uses it.
Zathalus
Member
The overwhelming majority of games don't need it. So most developers see little point in retooling for the API.Technology is useless if nobody uses it.
Thebonehead
Gold Member
FS2020 could sure use it.
Load times are in the minutes for that.
Not sure if 2024 uses it as I went back to 2020 as the performance in vr took a major hit so I uninstalled it.
May swap the 4090 across into the new rig and retry 2024 as I seem to remember it having direct storage.
Load times are in the minutes for that.
Not sure if 2024 uses it as I went back to 2020 as the performance in vr took a major hit so I uninstalled it.
May swap the 4090 across into the new rig and retry 2024 as I seem to remember it having direct storage.
Puscifer
Gold Member
There's a company called Sonic in California that offers 10gbps for 50 bucks a month. I used to work for them and wanna KMS every time I pay Comcast 45 for 350mbps
America’s Best Internet
Get fiber-optic internet service with speeds up to 10 Gigabits. Fast, fair, and affordable.
Last edited by a moderator:
Corporal.Hicks
Banned
My results. The PS3's peak VRAM throughput was 22GB/s, so it seems that GEN4 NVMe bandwidth is already faster than that.
Last edited:
YeulEmeralda
Linux User
I was under the impression that this feature would be great for open world games (no more pop in, one giant map without loading screens).
That's not what this says at all.
NVMe bandwidth in your example is 8GB/s, the decompressed bandwidth is in memory, not disc I/O.
Corporal.Hicks
Banned
"8.63 GB loaded in 0.36 seconds" already takes into account decompressed bandwidth. My RAW NVMe bandwidth is 7GB per second.
Last edited:
Negotiator
Banned
PC ports need it though, but not every PC gamer has NVMe. Many people still use SATA SSD or even HDD as their primary boot disk.
PC game devs would rather brute force the problem (more RAM for caching, more storage space needed)...
DirectStorage and sampler feedback streaming should reduce VRAM requirements a lot (not to mention NTC):
Coming to DirectX 12— Sampler Feedback: some useful once-hidden data, unlocked - DirectX Developer Blog
Why Feedback: A Streaming Scenario Suppose you are shading a complicated 3D scene. The camera moves swiftly throughout the scene, causing some objects to
devblogs.microsoft.com
Interesting, but we lack 10 Gbps Ethernet + 10 Gbps switches. Even high-end mobos support up to 2.5 Gbps.There's a company called Sonic in California that offers 10gbps for 50 bucks a month. I used to work for them and wanna KMS every time I pay Comcast 45 for 350mbps
America’s Best Internet
Get fiber-optic internet service with speeds up to 10 Gigabits. Fast, fair, and affordable.www.sonic.com
You're not wrong, this gen is full of open-world games.
The dataset in the test is compressed 3.5:1, so yes, you load around 2.4GB of actual data from NVMe.
Physiognomonics
Member
Not sure that game is even compressed by Kraken on PS5. The game is not compressed at all on PC (which is why it's running without I/O stutters as there is no textures to decompress).
The compressed game on PC would have caused a sea of I/O stutters like Spider-man or TLOU ports, all blaimed by "it's a bad port" excuse when the main problem is the lack of dedicated I/O hardware (and the low latency APIs) on PC.
Last edited:
Negotiator
Banned
God of War Ragnarok being only 100GB on PS5 (a single BDXL disc) indicates it uses Kraken compression. No?
100 vs 190GB is a hell of a difference... hell, I even remember cross-gen PS4/5 games where the PS4 version is considerably bigger (due to lack of Kraken compression).
Yes, PCs don't have a dedicated ASIC block for decompression, but at least they have tons of shader ALUs for GDeflate. It's better than nothing.
Couldn't they offer 2 versions/downloads of the same game? (vanilla/uncompressed vs GDeflate-compressed)
They need to do something about it, otherwise expect next-gen PS6 ports to cross the 400GB mark... this is not sustainable.
ps: I'm not sure what you mean by "low-latency APIs". What kind of latency exactly?
Zathalus
Member
There are very few games that load any slower than what happens on console. DirectStorage without GPU compression doesn't offer that much better performance and GPU DirectStorage is highly limiting in what compression algorithm can be used.
Storage space is something else, there you would see some reduction on PC but even God of War Ragnarok is not that much bigger, being 102GB vs 159GB (only installing English language pack to match Ps5).
Sampler Feedback is something else entirely, No game uses it, not even on Xbox, as it requires extensive work to actually implement and the PS5 lacks support for it.
Of course it is compressed on PC, It's only 57GB larger when not including the various language packs. And "it's a bad port" is no excuse, Spider-Man 2 and TLOU are bad ports, trying to gaslight people into thinking they are not is silly. There are plenty of games that have equally impressive loading and texture streaming on PC that are just fine. Ratchet & Clank, no I/O stutters. Cyberpunk 2077, no I/O stutters. Forbidden West, no I/O stutters. FF16, no I/O stutters.
Last edited:
Negotiator
Banned
I never said anything about loading times.
So it's best to avoid GDeflate, just because Kraken is not available on PC?
I'm not sure why you compare the console version with all languages available/installed (100GB) with a PC version that only has English. Doesn't sound like a fair comparison.
If you remove non-English languages from the PS5 version, it's even less than 100GB...
Then why did Microsoft release it?
PS5 also lacks DirectStorage and mesh shaders (required in Alan Wake 2), since it has its own proprietary API.
Ratchet uses DirectStorage and rightly so. The rift portal mechanic wouldn't work otherwise.
And as you said, there's no stuttering, so...
Zathalus
Member
It is limiting according to Nixxes:
The big Horizon Forbidden West PC tech interview with Nixxes and Guerrilla Games
How one of PS5's most advanced games made its way across to PC
www.eurogamer.net
" Patrick Den Bekker - Nixxes: That's actually one of the reasons not to support GPU decompression because it locks in your compression format, and GDeflate is very CPU-unfriendly. We'd love it if we could use decompression that works on both the CPU and GPU, because they we could make it optional for the user. We therefore decided to use LZ4 which, I think, is very, very fast on the CPU.
Michiel Roza - Nixxes : I think GDeflate is very useful if your game that is mostly CPU-bound and not GPU-bound, because it adds a lot of stress to the GPU. Therefore, we decided to do a very quick CPU-based decompression. In the past, we have done some tests with multiple different decompression types, and I think LZ4 always came on top. It's not the compression that makes the assets the smallest, but the trade-off between disk size and decompression speed was amazing for LZ4, so we've been using that quite a lot in our projects!
Patrick Den Bekker - Nixxes: As well as GPU bandwidth, there are also GPU scheduling issues from using GDeflate, so your frame-rate can become unpredictable... Hopefully we can convince them to support multiple compression types, becacuse at that point it makes it easier for us to opt in or for users to opt in... people can just experiment with what they like best."
Kraken is available on PC, but the licensing needs to be payed for. It's free on PS5 as Sony payed for it to be free for all PS5 games.
I'm comparing like for like, the PS5 version is 101.51GB with just the English pack installed. Each language pack on PS5 is another 1GB you have to manually install from the game menu. The Steam install comes with all of them and you have to go into the game folders and delete them yourself. Or use a repack.
It's an interesting technology, but we just have no games that actually use it to know how useful it really is. It might be great, or it might turn out like VRS that has become rather useless. As for Mesh Shaders, PS5 has Primitive Shaders which are almost identical, just being slightly less programmatically flexible.
I never claimed it is not useful, just that most games don't really need it. The problem is that the really good version (GPU enabled) is a bit inflexible at the moment and a bit of a pain to implement (apparently).
Physiognomonics
Member
Textures are not compressed.
Zathalus
Member
Based on?
Negotiator
Banned
winjer
Gold Member
Thebonehead
Gold Member
5090 / 9800X3D / Crucial T705