Digital Foundry Article Technical Discussion [2021]

[goonergaz]

Again hard drives have always gotten faster. We've had sata ssd's multiple times faster than the hard drives available in last gen systems for what 5 or 6 years before last gen systems even came out. Faster storage isn't an new thing. But what is different this generation is that ram didn't go up in size like it used too. So now everyone is trying to hype up faster storage speeds. If the systems came with 24 or 32 gigs of ram no one would be gushing over these ssds which also exist in computers and have for years.

I’d love to hear generations where you can load the whole dataset in a couple of seconds outside cartridges

Maybe you can list each generation advancement to prove your comments…but we both know the increase has never been this big.

We didnt have SSD's implemented by default with high efficiency controllers and games developed around that idea before. Actually you should thank consoles because they bring up solutions that can benefit the PC space where brute force is the norm and games are made around the idea that there are weak systems they need to support with simple downgrades.
Trying to spin the SSD in consoles to sound like nothing more than a storage to load games faster is very far from reality.
Even if you had 32GB of ram on PS5 a fast SSD opens up hugely what can be done with those 32GBs

Indeed, why have one without the other - it’s so restrictive…I really hope devs embrace the advantages these SSDs and I/Os bring.

 

[Phantom88]

 

[milk]

Yea it just seems odd , oh this drive suddenly got faster on consoles and its totally better than having this other storage that is multiple times faster.

It's indeed odd to favor a cheapo NVME SSD over a premium PCIE-5 SSD.

But if you are talking about a cheapo NVME vs a premium HDD, than its not odd at all. Even if very fast HHDs do exist, they are advertising sustained spead of concurrent data. Not random reads. Let us never forget that. Seek-time latency is atrocious to HDD performance, while irrelevant to SSDs. The magic of SSD as a baseline is not the raw speed (while that is nice too) but being able to fetch data from storage with absolute granularity, being able to read thousands of small files from all over the place all the time. No HDD can do that, no matter what it's top-speed is.

 

[babybumb]

Is the PS5 resolution right? nxgamer and elanalistadebits have it at 1800p average

 

[chris1515]

Is the PS5 resolution right? nxgamer and elanalistadebits have it at 1800p average

The resolution is right. Digitalfoundry is much more precise than other channel.

 

[chris1515]

It's indeed odd to favor a cheapo NVME SSD over a premium PCIE-5 SSD.

But if you are talking about a cheapo NVME vs a premium HDD, than its not odd at all. Even if very fast HHDs do exist, they are advertising sustained spead of concurrent data. Not random reads. Let us never forget that. Seek-time latency is atrocious to HDD performance, while irrelevant to SSDs. The magic of SSD as a baseline is not the raw speed (while that is nice too) but being able to fetch data from storage with absolute granularity, being able to read thousands of small files from all over the place all the time. No HDD can do that, no matter what it's top-speed is.

He talks about RAM but you need to fill RAM maybe he like wait 1 minutes and 28 second in Death Stranding against 3 seconds on PS5. First fast storage means better usage of RAM and less loading time. Even if you have 256 GB of RAM it means you can have the full game in memory you need to load data into it and a long first loading. Having only 16 GB of RAM and 13 or 12 GB of RAM for Games but being able to replace in 1 second is much clever.

 

[BRiT]

DF Written Article @ https://www.eurogamer.net/articles/digitalfoundry-2021-kena-bridge-of-spirits-tech-review

 

[Phantom88]

Is the PS5 resolution right? nxgamer and elanalistadebits have it at 1800p average

Alex theorized that most likely nxgamer is counting static images instead of active gameplay scenarios where its nowhere close to 1800p.

 

[Silent_Buddha]

It's indeed odd to favor a cheapo NVME SSD over a premium PCIE-5 SSD.

But if you are talking about a cheapo NVME vs a premium HDD, than its not odd at all. Even if very fast HHDs do exist, they are advertising sustained spead of concurrent data. Not random reads. Let us never forget that. Seek-time latency is atrocious to HDD performance, while irrelevant to SSDs. The magic of SSD as a baseline is not the raw speed (while that is nice too) but being able to fetch data from storage with absolute granularity, being able to read thousands of small files from all over the place all the time. No HDD can do that, no matter what it's top-speed is.

Well, it's not completely irrelevant, but seek times are massively lower. So, while random reads for an HDD with 4kb transfer size will drop below 1 MB/s they are very roughly 100-200x faster for an SSD (50-140 MB/s depending on drive, 88 MB/s for the SN750). That's still nowhere near sequential read speeds for an SSD or even the difference between the average random read latency where SSD is very roughly about 1000x faster.

Basically there's some unavoidable overhead associated with random reads beyond just the average latency for accessing the data location. Hence, random reads on SSD are still massively slower than sequential reads.

That said, read speed will increase with larger file transfer sizes. However, it'll always be significantly slower than sequential read speed. Interestingly, larger file transfer sizes also lead to higher latency for reads.

And where SSDs particularly excel is at IOPs, however, outside of datacenter applications IOPs will generally be quite low (like say for games) since consumer workloads don't have massive requests for data on nearly the same level as say a frequently accessed data store serving up web pages.

So with high enough read requests and larger transfer sizes, you can get up to half of the sequential read speed of an SSD like the SN750. But that's unlikely to ever happen for any consumer workload.

Regards,
SB

 

[milk]

Well, it's not completely irrelevant, but seek times are massively lower. So, while random reads for an HDD with 4kb transfer size will drop below 1 MB/s they are very roughly 100-200x faster for an SSD (50-140 MB/s depending on drive, 88 MB/s for the SN750). That's still nowhere near sequential read speeds for an SSD or even the difference between the average random read latency where SSD is very roughly about 1000x faster.

Basically there's some unavoidable overhead associated with random reads beyond just the average latency for accessing the data location. Hence, random reads on SSD are still massively slower than sequential reads.

That said, read speed will increase with larger file transfer sizes. However, it'll always be significantly slower than sequential read speed. Interestingly, larger file transfer sizes also lead to higher latency for reads.

And where SSDs particularly excel is at IOPs, however, outside of datacenter applications IOPs will generally be quite low (like say for games) since consumer workloads don't have massive requests for data on nearly the same level as say a frequently accessed data store serving up web pages.

So with high enough read requests and larger transfer sizes, you can get up to half of the sequential read speed of an SSD like the SN750. But that's unlikely to ever happen for any consumer workload.

Regards,
SB

Interesting. I was under the wrong assumption that seek times were low enough on SSD to be near-zero in real world performance. Guess not...

 

[see colon]

I’d love to hear generations where you can load the whole dataset in a couple of seconds outside cartridges

Whole dataset or just filling RAM? Because if we are talking whole datasets, PS5 would take 40+ seconds to load COD:Warzone's full dataset, and that's a last gen game.

If we are just talking ram, PS2 can fill it's ram from a DVD in about 6 seconds with sequential reads, about half a second from HDD.

 

[iroboto]

Uhh can we branch the IO/SSD posts to the IO/SDD thread?

I can see the arguments but the technical side of the discussion should happen elsewhere and not here.

 

[Deleted member 13524]

DF Written Article @ https://www.eurogamer.net/articles/digitalfoundry-2021-kena-bridge-of-spirits-tech-review

I read this and I think the .ini mention for TAAU is really cool.

On the othet hand, it's a bummer that only RTX cards are mentioned in the PC performance paragraphs. The game has no RT and it doesn't support DLSS. Reportedly it runs great on Radeons, older Geforces and non-RTX Turing that most gamers own.
This is a subject that should be taken into consideration more often because most PC gamers have no means to upgrade their 3-4 year old GPUs, either because there's no stock or because they're just not affordable. People thinking they need at least a RTX 2060 to play this might feel pressured into being scalped, which isn't great.

From this DSOGaming comparison, the game plays great even in a Maxwell 980Ti.

https://www.dsogaming.com/pc-performance-analyses/kena-bridge-of-spirits-pc-performance-analysis/


One interesting thing is that the game seems to run in 3 main + 4 lesser taxing threads. It's most probably a result of its 8th-gen lower common denominator. That also means the game doesn't really hurt from running in a 4-core/8-thread CPU, so the Steam Deck might run this pretty well.

 

[Deleted member 11852]

I’d love to hear generations where you can load the whole dataset in a couple of seconds outside cartridges

And cartridges also don't solve this problem. The Switch uses either carts or NAND for storage or games and games sure as hell don't load instantly. Even much smaller, simpler games on the DS and 3DS didn't load instantly.

Because the other side of the coin is even if you can load data at ridiculous speeds, the CPU still needs to be able to process that data in order to build the game world, setup whatever environmental simulations need to be running - enemies, pedestrians, card, birds, animals, weather, time, lighting.. whatever - lots and lots.

 

[chris1515]

And cartridges also don't solve this problem. The Switch uses either carts or NAND for storage or games and games sure as hell don't load instantly. Even much smaller, simpler games on the DS and 3DS didn't load instantly.

Because the other side of the coin is even if you can load data at ridiculous speeds, the CPU still needs to be able to process that data in order to build the game world, setup whatever environmental simulations need to be running - enemies, pedestrians, card, birds, animals, weather, time, lighting.. whatever - lots and lots.

Exactly currently CPU code is the bottleneck on R&C Rift Apart and they need to continue to improve CPU game engine to be able to push higher speed.

 

[PSman1700]

I read this and I think the .ini mention for TAAU is really cool.

On the othet hand, it's a bummer that only RTX cards are mentioned in the PC performance paragraphs. The game has no RT and it doesn't support DLSS. Reportedly it runs great on Radeons, older Geforces and non-RTX Turing that most gamers own.
This is a subject that should be taken into consideration more often because most PC gamers have no means to upgrade their 3-4 year old GPUs, either because there's no stock or because they're just not affordable. People thinking they need at least a RTX 2060 to play this might feel pressured into being scalped, which isn't great.

From this DSOGaming comparison, the game plays great even in a Maxwell 980Ti.

https://www.dsogaming.com/pc-performance-analyses/kena-bridge-of-spirits-pc-performance-analysis/


One interesting thing is that the game seems to run in 3 main + 4 lesser taxing threads. It's most probably a result of its 8th-gen lower common denominator. That also means the game doesn't really hurt from running in a 4-core/8-thread CPU, so the Steam Deck might run this pretty well.

In the same vein, there are many more PS4 owners out there than PS5 users (aswell as scalpers/no availability), so should PS5 performance analysis be omitted or have a backseat as opposed to 2012/2013 hardware, since its running 'quite well' on ps4? I think game reviews and analysis are always done on current hardware.
Also there are over 20 million RTX users out there? Quite a big market and relevant to performance-analysis. What they should have included are AMD gpus since they cover quite a chunk of the market too.

I doubt the availability of say a 980Ti is all that easy to find either

 

[Dictator]

I read this and I think the .ini mention for TAAU is really cool.

On the othet hand, it's a bummer that only RTX cards are mentioned in the PC performance paragraphs. The game has no RT and it doesn't support DLSS. Reportedly it runs great on Radeons, older Geforces and non-RTX Turing that most gamers own.
This is a subject that should be taken into consideration more often because most PC gamers have no means to upgrade their 3-4 year old GPUs, either because there's no stock or because they're just not affordable. People thinking they need at least a RTX 2060 to play this might feel pressured into being scalped, which isn't great.

From this DSOGaming comparison, the game plays great even in a Maxwell 980Ti.

https://www.dsogaming.com/pc-performance-analyses/kena-bridge-of-spirits-pc-performance-analysis/


One interesting thing is that the game seems to run in 3 main + 4 lesser taxing threads. It's most probably a result of its 8th-gen lower common denominator. That also means the game doesn't really hurt from running in a 4-core/8-thread CPU, so the Steam Deck might run this pretty well.

I think what we do is different than what you may desire.
I retired primarily covering the RX 580 and GTX 1060 before the GPU shortage, including covering an older i5 8400. Time moves on, and so does DF. On top of that, a DF vid like this last one covers a lot more diverse topics than the average PC GPU round up and requires a lot more diverse work. The video examines the graphical presentation, compares the graphics between versions, does the performance for 2 console versions and 3 graphical modes, talks about the console graphics vs. PC in precise terms of what the settings are on PC, adds info about how to tweak the game to get a similar visual experience on PC without losing too much in visuals ("optimised settings"), tests the differences in large terms between DX11 and DX12, tests loading between versions, and then wraps it all up in a long form produced video essay with a script and production totalling 4 days. I got Kena on all systems on Tuesday at noon (post game-launch) and finished the video on Friday at 19. To say the least, covering anymore would explode the conditions of producing anything in a timely manner.
Beyond that, I do not want my coverage of PC stuff to just be covering tons of GPUs and graphs and whatever. PC outlets have been doing that for 30 years almost and other outlets do that well... we do something different.

 

[goonergaz]

Whole dataset or just filling RAM? Because if we are talking whole datasets, PS5 would take 40+ seconds to load COD:Warzone's full dataset, and that's a last gen game.

If we are just talking ram, PS2 can fill it's ram from a DVD in about 6 seconds with sequential reads, about half a second from HDD.

Layman talk for everything the game needs to run, and yes, there will be small games - I specifically talking about ‘big games’ that will require a lot of loading/storage access…I don’t recall PS2 loading times being consistently a maximum of 6 seconds, but I may be wrong. Either way that’s still slower than PS5 which can get you into game in a couple of seconds no?

 

[PSman1700]

Layman talk for everything the game needs to run, and yes, there will be small games - I specifically talking about ‘big games’ that will require a lot of loading/storage access…I don’t recall PS2 loading times being consistently a maximum of 6 seconds, but I may be wrong. Either way that’s still slower than PS5 which can get you into game in a couple of seconds no?

Transformers PS2 had large worlds with no loading and zero pop-in, aswell as no draw distance limiter (though with LOD levels), amazing looking for the ps2 and 60fps. You could fly pretty fast through the environments. Where those worlds loaded in its puny 32mb memory?

 

[chris1515]

Layman talk for everything the game needs to run, and yes, there will be small games - I specifically talking about ‘big games’ that will require a lot of loading/storage access…I don’t recall PS2 loading times being consistently a maximum of 6 seconds, but I may be wrong. Either way that’s still slower than PS5 which can get you into game in a couple of seconds no?

PS2 was far from loading time of PS5. It was slow but better than PS4.