NVIDIA 4000 Series

[gpuerrilla]

They are like Apple,wanting to skimp on hardware as much as possible to increase margins.

And have a nice proprietary ecosystem to ringfence.

 

[CAT-THE-FIFTH]

And have a nice proprietary ecosystem to ringfence.

But,but consoles are PrOpRiEtArY,Apple is PrOpRiEtArY!

 

[Calin Banc]

Nvidia is so anti vram, they're willing to come up with new texture compression algorithms

It does look kinda cool - so you can have a texture that looks closer to native 4k but only takes up the space in vram that a native 1080p texture would. Main issue seems to be the processing time since it runs on the TMU - it takes the GPU 3 times longer to render these compressed textures

I was thinking it runs on the AI cores in the GPU. If it runs on the TMUs it will bad for AMD (if will be somehow open for them too), I think they're using that for RT, right?
As for vRAM... well, consoles and especially PS5 went nuts with the storage to avoid more of it as well (30s of gameplay in RAM with previous gen, 1 second now, so you don't need as much); I wouldn't be to tough on them...

But, in some interview from MLID there were talks of about 20% or so, performance lost on a 4090 to use direct storage. Add some loss from this as well and we'll have a great deal of penalty for... nothing. Strange times ahead...

 

[gpuerrilla]

But, in some interview from MLID there were talks of about 20% or so, performance lost on a 4090 to use direct storage. Add some loss from this as well and we'll have a great deal of penalty for... nothing. Strange times ahead...

Is this factoring the driver overhead already present?

 

[CAT-THE-FIFTH]

I was thinking it runs on the AI cores in the GPU. If it runs on the TMUs it will bad for AMD (if will be somehow open for them too), I think they're using that for RT, right?
As for vRAM... well, consoles and especially PS5 went nuts with the storage to avoid more of it as well (30s of gameplay in RAM with previous gen, 1 second now, so you don't need as much); I wouldn't be to tough on them...

But, in some interview from MLID there were talks of about 20% or so, performance lost on a 4090 to use direct storage. Add some loss from this as well and we'll have a great deal of penalty for... nothing. Strange times ahead...

If it means they can sell you less hardware for more money,it's totally worth it for them. Then when the new generation of hardware comes along,optimisation of the algorithms will be for the next generation,or a new version will only support newer hardware,so the older generation will start to have more issues.

Also it makes me wonder how long before all these features,start getting locked behind subscription models. So buy a new dGPU,and after the first two years you have to pay for support,etc.

It's happening even with cars,ie,certain features you need to buy a subscription too:

BMW Responds to Fury Over Heated Seats Subscription Fee

The automaker charges for the use of pre-installed hardware in some markets, and people still aren't happy about it.

www.thedrive.com

Edit!!

The consoles are dedicated I/O controllers so there isn't that penalty for I/O. Makes you wonder WTF,Microsoft,AMD and Intel are doing on the platform side. We shouldn't be needing the GPU to handle I/O in the first place.

 

[Calin Banc]

Is this factoring the driver overhead already present?

Don't know. Basically, someone from nvidia was saying they could do on the GPU what is being done on consoles, same decompression/streaming, but will need about 32gb vRAM card, then (actually before in the video), it was mentioned the 20% penalty for direct storage, perhaps extra on top to avoid stutters in UE5 and such... Ah, and 64GB DDR5

Is a video @humbug posted a while back.

1:04:40 - Direct Storage, do what PS5 does. 20% when streaming and rendering the game (then adding, you'll still have stuttering, so you'll need to lose 30-40% to remove stutter) and boom! Half the performance is gone.
1:06:45 - RTX I/O aka Direct Storage talk
1:29:39 - he mentioned the texture thing, maybe done on Tensor cores, not TMUs, so he knows at least a bit of stuff beforehand

With that said I'm still wondering why on earth they've went with that crazy storage... to save vRAM I guess.

Oh, and if you want to decode on the CPU you need 9 Ryzen 2 Cores for PS5 and 2 Cores for Xbox ) Nuts!

While on PC we have all sorts of cool demos which most likely won't run on current consoles due to lack of raw power (until a refresh is coming at least), are there games with some nice graphics on consoles and gameplay enable by that ultra-giga fast storage? I know about Ratchet and Clank, but overall doesn't look like something that couldn't be done with the "regular" tech.

 

[Calin Banc]

If it means they can sell you less hardware for more money,it's totally worth it for them. Then when the new generation of hardware comes along,optimisation of the algorithms will be for the next generation,or a new version will only support newer hardware,so the older generation will start to have more issues.

Also it makes me wonder how long before all these features,start getting locked behind subscription models. So buy a new dGPU,and after the first two years you have to pay for support,etc.

It's happening even with cars,ie,certain features you need to buy a subscription too:

BMW Responds to Fury Over Heated Seats Subscription Fee

The automaker charges for the use of pre-installed hardware in some markets, and people still aren't happy about it.

www.thedrive.com

Edit!!

The consoles are dedicated I/O controllers so there isn't that penalty for I/O. Makes you wonder WTF,Microsoft,AMD and Intel are doing on the platform side. We shouldn't be needing the GPU to handle I/O in the first place.

TBF, that's pushing tech forward (speaking about the texture stuff), doing more with less should be applauded. So far DLSS has worked fine for me, so fingers crossed they don't get your ideas!

As for the I/O, I was wondering the same. Why not having dedicated hardware for streaming/decompression in the GPUs (since it's used for gaming) with these new cards? Hopefully will come later on (as in with the next gen of cards), as of now it looks like a total waste of performance otherwise.

 

[gpuerrilla]

Edit!!

The consoles are dedicated I/O controllers so there isn't that penalty for I/O. Makes you wonder WTF,Microsoft,AMD and Intel are doing on the platform side. We shouldn't be needing the GPU to handle I/O in the first place.

Good spot @Calin Banc I just watched the timestamp. Yeah so lots of vram is needed eh... gonna age well this.

I was about to ask what is happening with RTX IO when you posted.

 

[CAT-THE-FIFTH]

TBF, that's pushing tech forward (speaking about the texture stuff), doing more with less should be applauded. So far DLSS has worked fine for me, so fingers crossed they don't get your ideas!

I am just very cynical about all this,especially when the accountants and stock market lot get involved. The current upselling from Nvidia and AMD is most likely because of this. If engineers had their way we would be getting something closer to Pascal.

As for the I/O, I was wondering the same. Why not having dedicated hardware for streaming/decompression in the GPUs (since it's used for gaming) with these new cards? Hopefully will come later on (as in with the next gen of cards), as of now it looks like a total waste of performance otherwise.

That would mean more die area,and even higher dGPU costs. Not sure why this isn't integrated into the standard I/O on motherboards/CPUs. Seems weird we need to use dGPUs for all of this.

Good spot @Calin Banc I just watched the timestamp. Yeah so lots of vram is needed eh... gonna age well this.

I was about to ask what is happening with RTX IO when you posted.

Don't worry! You can buy another card for I/O.

 

[gpuerrilla]

Don't worry! You can buy another card for I/O.

 

[CAT-THE-FIFTH]

Don't know. Basically, someone from nvidia was saying they could do on the GPU what is being done on consoles, same decompression/streaming, but will need about 32gb vRAM card, then (actually before in the video), it was mentioned the 20% penalty for direct storage, perhaps extra on top to avoid stutters in UE5 and such... Ah, and 64GB DDR5

Is a video @humbug posted a while back.

1:04:40 - Direct Storage, do what PS5 does. 20% when streaming and rendering the game (then adding, you'll still have stuttering, so you'll need to lose 30-40% to remove stutter) and boom! Half the performance is gone.
1:06:45 - RTX I/O aka Direct Storage talk
1:29:39 - he mentioned the texture thing, maybe done on Tensor cores, not TMUs, so he knows at least a bit of stuff beforehand

With that said I'm still wondering why on earth they've went with that crazy storage... to save vRAM I guess.

Oh, and if you want to decode on the CPU you need 9 Ryzen 2 Cores for PS5 and 2 Cores for Xbox ) Nuts!

While on PC we have all sorts of cool demos which most likely won't run on current consoles due to lack of raw power (until a refresh is coming at least), are there games with some nice graphics on consoles and gameplay enable by that ultra-giga fast storage? I know about Ratchet and Clank, but overall doesn't look like something that couldn't be done with the "regular" tech.

Just shows that Microsoft,AMD,Intel and Nvidia have been more worried about counting their profits instead of bothering to innovate on the PC. We went from 768MB/1GB VRAM on mainstream dGPUs in 2009 to 6GB/8GB in 2016. So by now we should have been at 48GB/64GB at the same rate of improvement.

There was no reason the companies couldn't have worked together to actually introduce similar functionality on PC,if they didn't want to increase the amount of VRAM. The sad reality is I/O is still used in a backward way on PC,and most PC games just treat it like a faster HDD.

The consoles at least bothered to address it by trying to use the storage in a different way.

The storage isn't crazy fast - the PS5 storage maxes out at 5500MB/S,and Mike Cerny suggests the WD SN850 or Seagate FireCuda as expansion drives. The official Sony expansion drive uses a WD SN850. The XBox Series X uses an SN530 which maxes out at 2400MB/S.

 

[muon]

Steam Hardware & Software Survey

New month on steam hardware survey.

Basically need to ignore last month as there must have been an issue as certain cards spiked up, but are now back down to the normal trend level e.g. 3060, 2060.

So comparing to February 23, in April 23:

RTX 4090 has gone from 0.31% to 0.43%
RTX 4080 has gone from 0.20% to 0.29%
RTX 4070 Ti has gone from 0.18% to 0.41%
RTX 4070 not on main list yet

Sales have slowed down for the 4080 and 4090, I would say.

 

[Dicehunter]

1:06:45 - RTX I/O aka Direct Storage talk

I remember RTX I/O being a selling point of the 3000 series, Nearly 3 years on and it's still vapourware.

 

[Calin Banc]

Just shows that Microsoft,AMD,Intel and Nvidia have been more worried about counting their profits instead of bothering to innovate on the PC. We went from 768MB/1GB VRAM on mainstream dGPUs in 2009 to 6GB/8GB in 2016. So by now we should have been at 48GB/64GB at the same rate of improvement.

There was no reason the companies couldn't have worked together to actually introduce similar functionality on PC,if they didn't want to increase the amount of VRAM. The sad reality is I/O is still used in a backward way on PC,and most PC games just treat it like a faster HDD.

The consoles at least bothered to address it by trying to use the storage in a different way.

The storage isn't crazy fast - the PS5 storage maxes out at 5500MB/S,and Mike Cerny suggests the WD SN850 or Seagate FireCuda as expansion drives. The official Sony expansion drive uses a WD SN850. The XBox Series X uses an SN530 which maxes out at 2400MB/S.

I was reading stuff like below, but I don't know how they reached that almost 10GB/s. however, multiplatform games still need to be design around the limitation of the weakest link which is the Xbox - at the very least. I'm not aware if the storage has some special characteristics besides the speed, like a constant read rate that is higher than the usual drives can provide.

How to Run PS5 SSD Speed Test [100% Working Solutions]

Are you unsure which PS5 SSD to select or what speed your PS5 is currently capable of? Following this post's guidelines, you can run PS5 SSD speed tests and get the fastest drive for your games.

www.easeus.com

PS5's PCIe 4.0 x 4 NVMe interface allows it to read uncompressed data at speeds of up to 5.5 GB/sec, and its powerful Kraken compression technology allows it to read compressed data at speeds of 9.9 GB/sec.

I remember RTX I/O being a selling point of the 3000 series, Nearly 3 years on and it's still vapourware.

Sadly, Direct Storage is missing almost completely and since they've made it a windows 11 thing... At the end of the day, cards can support this, but without gaming and OS implementing is just a checkmark indeed.

 

[CAT-THE-FIFTH]

I was reading stuff like below, but I don't know how they reached that almost 10GB/s. however, multiplatform games still need to be design around the limitation of the weakest link which is the Xbox - at the very least. I'm not aware if the storage has some special characteristics besides the speed, like a constant read rate that is higher than the usual drives can provide.

How to Run PS5 SSD Speed Test [100% Working Solutions]

Are you unsure which PS5 SSD to select or what speed your PS5 is currently capable of? Following this post's guidelines, you can run PS5 SSD speed tests and get the fastest drive for your games.

www.easeus.com

Sadly, Direct Storage is missing almost completely and since they've made it a windows 11 thing... At the end of the day, cards can support this, but without gaming and OS implementing is just a checkmark indeed.

Thanks - I forgot about the custom compression technologies! So its even worse for us on PC!

 

[gpuerrilla]

$2100 on the scalper market @tommybhoy !

 

[Dicehunter]

$2100 on the scalper market @tommybhoy !

EVGA definitely got out just in time.

 

[Jay85]

Wow… £20 reduction, round of applause

 

[Calin Banc]

Thanks - I forgot about the custom compression technologies! So its even worse for us on PC!

We'll see. They still have to optimize for Xbox on multiplatform games and that's much weaker, defiantly doable for plenty of PCs.

Time will tell, I guess.

 

[CAT-THE-FIFTH]

We'll see. They still have to optimize for Xbox on multiplatform games and that's much weaker, defiantly doable for plenty of PCs.

Time will tell, I guess.

If they are using their own custom decompression algorithms it might be closer to a decent PCI-E 4.0 SSD then. The consoles actually didn't skimp too much on the SSD type - they use TLC drives with DRAM cache,instead of cheaper DRAMless QLC drives. The TLC drives have much better write consistency as they get filled. The only thing is a lot of cheap gaming SSDs are QLC drives,where write speeds can be more of an issue.

But even before that we need Microsoft to work better with hardware companies to try and incorporate better platform support. If they are not bothering to spend money on increasing VRAM amounts,then something has got to give. Otherwise how can you push features like RT,whilst hampering the dGPU without a critical resource? It makes me wonder whether the generation after this will be the one to look out for. Makes you wonder what we might have got without a Pandemic and Mining...I suspect a different landscape.