8GB VRAM not enough (and 10 / 12)

BFG10K · Jul 13, 2021

8GB

Dying Light 2, Battlefield 2042, Avatar Frontiers of Pandora, Call of Duty Modern Warfare 3, Starfield, Forza Horizon 5, Horizon Forbidden West, Alan Wake 2, The Last of Us, Tekken 8, FC 24, Detroit Become Human, Death Stranding, Alone in The Dark, Overwatch 2
3060 12GB higher 1% low than 4060 8GB https://www.youtube.com/watch?v=gIfz57wmywg
Cyberpunk 2077, Diablo 4, Hogwarts Legacy, Horizon Zero Dawn, Microsoft Flight Sim 2020, Witcher 3, Watch Dogs Legion skip to 4K + DLSS for each game https://www.youtube.com/watch?v=tPiaWWd0xGc
Ratchet & Clank, Plague Tale Requiem, Last Of Us, Jedi Survivor, Far Cry 6, Hogwarts Legacy, Forza Horizon 5, Cyberpunk 2077 https://www.youtube.com/watch?v=Wr3MckDxMfE&pp=ygUTMTZnYiB2cyA4Z2IgNDA2MCB0aQ==
Far Cry 6, Godfall, Hogwarts Legacy, Forza Horizon 5, Forspoken, Doom Eternal, Watchdogs Legion https://youtu.be/fJc--C01P90?t=268
Last Of Us, Resident Evil 4, Callisto Protocol, Plague Tale Requiem, Halo Infinite, Forspoken https://youtu.be/2_Y3E631ro8?t=429
Last Of Us, Hogwarts Legacy, Resident Evil 4, Forspoken, Plague Tale Requiem, Callisto Protocol https://youtu.be/Rh7kFgHe21k?t=288
Hogwarts Legacy, Last Of Us, Spider Man Remastered, Doom Eternal, Forza Horizon 5 https://youtu.be/R943CbDTq_s?t=1473
Horizon Forbidden West 4060 has issues @ 1080p + DLSS unless you switch off frame generation and reduce settings to high https://youtu.be/3AyKvI23VGw?t=233
Horizon Forbidden West 3070 requires reducing textures 2 notches https://youtu.be/xTwEGy6HHKo?t=351
Deathloop https://youtu.be/lu_XHyO-6zY?t=918
Ghost Recon Breakpoint https://youtu.be/iUGppQVpXzU?t=529
Resident Evil 2 https://youtu.be/Aa-gsgRjMII?t=796
Talos Principle 2 https://youtu.be/K6FATQAuEwI?t=337
Spider Man Remastered https://youtu.be/vRVR_plVSlc?t=1060
Witcher 3 https://youtu.be/vRVR_plVSlc?t=1240

Horizon Forbidden West 3060 is faster than the 2080 Super despite the former usually competing with the 2070. Also 3060 has a better 1% low than 4060 and 4060Ti 8GB.

Resident Evil Village 3060TI/3070 tanks at 4K and is slower than the 3060/6700XT when ray tracing:

Company Of Heroes 3060 has a higher minimum than the 3070TI:

10GB / 12GB

Plague Tale Requiem 3080 10GB tanks if you enable ray tracing, would be fast enough if it had more VRAM because if you stop moving, the framerate stabilizes to >60FPS https://youtu.be/7zjrMww3fAc?t=496
Hogwarts Legacy 4K + RT, 4060Ti 16GB is faster than 4070 12GB https://youtu.be/X-h_R488Mq8?t=191

Reasons why still shipping 8GB since 2014 isn't NV's fault.

It's the player's fault.
It's the reviewer's fault.
It's the developer's fault.
It's AMD's fault.
It's the game's fault.
It's the driver's fault.
It's a system configuration issue.
Wrong settings were tested.
Wrong area was tested.
Wrong games were tested.
4K is irrelevant.
Texture quality is irrelevant as long as it matches a console's.
Detail levels are irrelevant as long as they match a console's.
There's no reason a game should use more than 8GB, because a random forum user said so.
It's completely acceptable for the more expensive 3070/3070TI/3080 to turn down settings while the cheaper 3060/6700XT has no issue.
It's an anomaly.
It's a console port.
It's a conspiracy against NV.
8GB cards aren't meant for 4K / 1440p / 1080p / 720p gaming.
It's completely acceptable to disable ray tracing on NV while AMD has no issue.
Polls, hardware market share, and game title count are evidence 8GB is enough, but are totally ignored when they don't suit the ray tracing agenda.

According to some people here, 8GB is neeeevaaaaah NV's fault and objective evidence "doesn't count" because of reasons(tm). If you have others please let me know and I'll add them to the list. Cheers!

jpiniero · Dec 12, 2023

New benchmark video shows how much better asset streaming latency is using DirectStorage with GPU decompression over not using it... but using the shaders to do the decompression hurts the frame rate. A fixed function block would def help.

blckgrffn · Dec 12, 2023

jpiniero said:
New benchmark video shows how much better asset streaming latency is using DirectStorage with GPU decompression over not using it... but using the shaders to do the decompression hurts the frame rate. A fixed function block would def help.

Oh man, like the old days of physx can we off load to a secondary card? 😂

I have a short stack of 6600’s…

coercitiv · Dec 13, 2023

jpiniero said:
New benchmark video shows how much better asset streaming latency is using DirectStorage with GPU decompression over not using it... but using the shaders to do the decompression hurts the frame rate. A fixed function block would def help.

It does not hurt performance when used as intended. No matter how you look at it, it's still a net gain over what we had before. The point of using DirectStorage is to either shorten time spent in the loading screen or asynchronously load assets over a relatively long period of time for games that aim to avoid loading screens. So when the GPU can stream 8-10x faster than the CPU you can either shorten an 8 second loading screen to a 1 second transition animation, or you can stream with the same speed as with a CPU but with minimal load on the GPU and zero load in the CPU.

A fixed block would be even more efficient so may become tempting as assets grow in size. I can see it make perfect sense in the consoles. It would probably also introduce limitations, such as supporting only certain compression methods. For example, Nvidia is looking into beating traditional compression with some type of ML hybrid method.

blckgrffn said:
Oh man, like the old days of physx can we off load to a secondary card? 😂

I have a short stack of 6600’s…

Imagine the joy of transferring many GB of uncompressed textures over the PCIe bus

blckgrffn · Dec 13, 2023

coercitiv said:
Imagine the joy of transferring many GB of uncompressed textures over the PCIe bus

Haha, yes, I know. It just struck me funny how we seem to go through these cycles where everything is different but seems like flavors of the same.

Hopefully we'll see some algo agnostic decompression hw block, but if AMD were to add the same hardware as featured in the consoles to their cards - or a version thereof - so long as it was DirectStorage compliant it seems like whatever it is would quickly become the standard to measure other solutions by.

SolidQ · Dec 19, 2023

Seems we're still far from 4k native...

BFG10K · Dec 31, 2023

Happy 10th birthday, 8GB VRAM!

Sapphire's Vapor-X R9 290X 8GB: The More, The Merrier?

We've waited patiently since we saw this card demoed at CES in the beginning of 2014. Sapphire has finally released the 8GB version of it's Vapor-X R9 290X. Does more memory make us merrier?

www.tomshardware.com

Here's to another ten more amazing years of 8GB, all thanks to Papa Jensen for allowing us to run "correct settings".

Saylick · Dec 31, 2023

BFG10K said:
Happy 10th birthday, 8GB VRAM!

Sapphire's Vapor-X R9 290X 8GB: The More, The Merrier?

We've waited patiently since we saw this card demoed at CES in the beginning of 2014. Sapphire has finally released the 8GB version of it's Vapor-X R9 290X. Does more memory make us merrier?

www.tomshardware.com

Here's to another ten more amazing years of 8GB, all thanks to Papa Jensen for allowing us to run "correct settings".

What’s old is new, and what’s new is old.

igor_kavinski · Dec 31, 2023

Saylick said:
What’s old is new, and what’s new is old.

Sure, I've got no problem if they create a universal 8GB standard which mandates that games nust work at 60 fps with ultra quality graphics. Then put the good games following the standard on a Hall of Fame list and the offenders on the Hall of Shame list. That should force most PR sensitive developers to optimize their games properly.

BFG10K · Jan 3, 2024

First result from 2024, Talos Principle 2.

1440p DLSS upscaled, 4060TI 16GB is 18% faster than 8GB.

Fanatical Meat · Jan 4, 2024

I don’t get why this is even a discussion. I believe the spot price for 8GB of GDDR6 is like $20-something dollars. It just doesn’t cost all that much more to add it.
I know this is a huge simplification I am making however if it’s designed to have the larger memory it isn’t a big deal.
Just add the 8GB and charge $25 to $50 more to preserve margins.
Problem is the smaller memory is designed to encourage yet another expensive upgrade a few years down the line.
This is the problem.

SteinFG · Jan 19, 2024

3GB GDDR7 chips should be out by the time RTX 50 launches, and I hope Nvidia will just give every tier 1.5x VRAM, without anything stupid, so 5060 - 12, 5070 - 18, 5080 - 24, 5090 - 36. But If they're reducing bus width again next gen, the general sentiment will be very bleak.

TESKATLIPOKA · Jan 19, 2024

SteinFG said:
3GB GDDR7 chips should be out by the time RTX 50 launches, and I hope Nvidia won't be stupid and just give every tier 1.5x VRAM (5060 - 12, 5070 - 18, 5080 - 24, 5090 - 36). But If they're reducing bus width again next gen, the general sentiment will be very bleak.

It would be great If Nvidia gave us 1.5x Vram more per tier, not sure why you think that's not enough.
Problem would be If they reduced bus-width thanks to higher transfer rates from GDDR7.

SteinFG · Jan 19, 2024

TESKATLIPOKA said:
It would be great If Nvidia gave us 1.5x Vram more per tier, not sure why you think that's not enough.

I think that wouldbe great, rephrased my comment to make it clear

BFG10K · Jan 19, 2024

TESKATLIPOKA said:
It would be great If Nvidia gave us 1.5x Vram more per tier, not sure why you think that's not enough.

I wouldn't be surprised if they're still milking 8 /12 next generation except at the highest tiers. With a virtual monopoly in GPU space NV have no reason to do otherwise.

coercitiv · Jan 20, 2024

BFG10K said:
I wouldn't be surprised if they're still milking 8 /12 next generation except at the highest tiers. With a virtual monopoly in GPU space NV have no reason to do otherwise.

They can increase it a bit even with lower bus widths. A possible scenario can be 5060 96bit/9GB, 5060 TI 128/12GB, 5070 160/15GB. AFAIK there will also be 2GB GDDR7 chips, so they could also make 5060 96/12GB if the market is willing to pay. These SKU predicitons are not my cup of tea, but it should be obvious they have more room to maneuver with both 3GB and 2GB chips as options.

Overall I would expect the most cost effective SKUs to come out the gate, both in terms of die size (so bus width) and BoM. Even more room for the SUPERs!

TESKATLIPOKA · Jan 20, 2024

BFG10K said:
I wouldn't be surprised if they're still milking 8 /12 next generation except at the highest tiers. With a virtual monopoly in GPU space NV have no reason to do otherwise.

This is also a possibility, but I think they will go for 3GB modules If available.

coercitiv said:
They can increase it a bit even with lower bus widths. A possible scenario can be 5060 96bit/9GB, 5060 TI 128/12GB, 5070 160/15GB. AFAIK there will also be 2GB GDDR7 chips, so they could also make 5060 96/12GB if the market is willing to pay. These SKU predicitons are not my cup of tea, but it should be obvious they have more room to maneuver with both 3GB and 2GB chips as options.

Overall I would expect the most cost effective SKUs to come out the gate, both in terms of die size (so bus width) and BoM. Even more room for the SUPERs!

96-bit bus means 3 memory modules, so you would need 4GB(32Gbit) modules for 12 GB Vram or 6x 16Gbit modules in clamshell.
Don't think many would pay for the extra Vram, If Nvidia will once more ask an unreasonable price for It like they did with 4060Ti 16GB.

GDDR7 would allow a significant increase in BW, so Nvidia won't need to increase L2 cache and could also decrease controller width.
128-bit 16-18gbps GDDR6 -> 256-288GB
128-bit 30-32gbps GDDR7 -> 480-512GB/s
96-bit 30-32gbps GDDR7 -> 360-384GB/s

The nextgen *107 chip could have 32SM:128TMU:48ROP paired with 24MB L2 and 96-bit 30gbps GDDR7. With 10% higher clocks It will be faster than 4060Ti.

coercitiv · Jan 20, 2024

TESKATLIPOKA said:
96-bit bus means 3 memory modules, so you would need 4GB(32Gbit) modules for 12 GB Vram or 6x 16Gbit modules in clamshell.

Except my post mentions 96bit / 9GB config for 3GB chips. The 96/12GB config is mentioned for 2GB chips, obviously in clamshell "if the market is willing to pay".

TESKATLIPOKA · Jan 20, 2024

coercitiv said:
Except my post mentions 96bit / 9GB config for 3GB chips. The 96/12GB config is mentioned for 2GB chips, obviously in clamshell "if the market is willing to pay".

Yeah, I saw that later, that's why I added clamshell but didn't change the first part.

Unless GDDR7 starts at much lower speeds, It's very likely Nvidia will reduce memory width because It's not needed. In desktop they could go for clamshell, but in laptop they would be limited to 9GB. Successor to 4050 could be once more limited to 6GB using 64-bit width and 2 modules.

jpiniero · Jan 20, 2024

TESKATLIPOKA said:
Unless GDDR7 starts at much lower speeds

Kind of figure that any low end desktop Blackwell will use GDDR6(X). Despite that I'd expect 7 to be 96-bit and 5 to be maybe less than 192-bit.

igor_kavinski · Feb 14, 2024

Custom upgraded 22GB 2080 Ti available for $499: https://www.ebay.com/itm/256351082761

Excellent VRAM/dollar ratio or what?

psolord · Feb 15, 2024

igor_kavinski said:
Custom upgraded 22GB 2080 Ti available for $499: https://www.ebay.com/itm/256351082761

Excellent VRAM/dollar ratio or what?

View attachment 93859

Wow 22GBs. I bet with the new RTX Chat tool, it will be able to speak french too!

BFG10K · Feb 16, 2024

igor_kavinski said:
Custom upgraded 22GB 2080 Ti available for $499:

The more [VRAM] you buy, the more you save!

igor_kavinski · Feb 16, 2024

igor_kavinski said:
Custom upgraded 22GB 2080 Ti available for $499: https://www.ebay.com/itm/256351082761

Excellent VRAM/dollar ratio or what?

View attachment 93859

Proof that WTFtech scours these forums: https://wccftech.com/nvidia-geforce-rtx-2080-ti-gpus-equipped-22-gb-vram-ai-market-costs-499-us/

That author pretending that HE found the listing on eBay after I posted about it.

igor_kavinski · Feb 16, 2024

This is Unreal Engine 5.3

The future we deserve but denied to us thanks to being limited to 8GB at the entry level.

MrTeal · Feb 16, 2024

Entry level? The 4060 Ti firmly midrange; you could buy a Series X at Walmart before the holidays for less than you can buy an 8GB 4060 Ti.

8GB VRAM not enough (and 10 / 12)

Lifer

Lifer

Diamond Member

Diamond Member

Diamond Member

Senior member

Lifer

Diamond Member

Lifer

Lifer

Lifer

Senior member

Platinum Member

Senior member

Lifer

Diamond Member

Platinum Member

Diamond Member

Platinum Member

Lifer

Lifer

Golden Member

Lifer

Lifer

Lifer

Diamond Member