Question How much vram is too much vram ?

[marees]

This is a joke thread !!

Please don't comment 🤪🙏

Need a thread for how much vram is too much vram 😉

 

[marees]

Timestamped. AMD will launch a 16gb or 24gb budget card
& can go upto 64gb on budget cards

 

[marees]

Lpddr6 / lpddr5x for 'budget' gpus to be shared with medusa halo & medusa premium

 

[marees]

Still 2 years out for the release. Fingers crossed. In Lisa Su we trust

(AMD has had spectacular failures previously such as bulldozer)

 

[marees]

If you had these 3 options (for an entry level GPU) then which one are you buying 🤔

• 10050xt 32gb = $350 (~9060xt 16gb)
• 10050xt 24gb = $300 (~9060)
• 10040xt 16gb = $250 (~3060 12gb in raster)

My revised estimations / guesstimates (no LLMs used)

Imagine this line-up (in 2027)

• AT0
  • 10090xt+ — Multiple models starting at $1500 plus and huge vram like Radeon VII or titan
• AT1
  • 10080xt — scrapped (Lisa Su took her toys & went home)
• AT2 (gddr7)
  • 10070 xtx 24gb = $700 (~5080)
  • 10070 xt 20gb = $600 (~5070 ti)
  • 10070 gre 16gb = $500 (~5070 super)
• AT3 (lpddr6)
  • 10060 xt 24gb = $450 (~5070)
  • 10060 16gb = $400 (~5060ti 16gb)
• AT4 (lpddr6/lpddr5x)
  • 10050xt 32gb = $350 (~9060xt 16gb)
  • 10050xt 24gb = $300 (~9060)
  • 10040xt 16gb = $250 (~3060 12gb in raster)

 

[marees]

Good point below. Let us see what happens 🤔

I know what you are thinking of. But does that work today as good as in the past? Anybody can pull out ChatGPT to ask for the better GPU (and might get the correct answer - or not).
For example, the megapixel race has pretty much ended. Many new phones and cameras get released with fewer pixels than their predecessor. People either gained more knowledge (more pixels != more quality), simply don't care because good enough or are not interested in technical details.

Higher VRAM amounts on lower end parts make the more expensive ones less attractive as well.

 

[SamMaster]

To be honest, manufacturers have been giving less and less performance difference between tiers and generations in comparison to past releases, so really they are manufacturing their own problem, and performance of lower end cards has reached a point that 8GB is artificially limiting them.

 

[Mopetar]

How much is too much? Whatever amount makes the PC too large to fit under my desk.

Of course I could always get a bigger desk 🤔

 

[marees]

MLID demands 36 gb to 40 gb for PS6 handheld

Claimed Sony PS6 handheld console specs promise a miracle of next-gen, cutting-edge processor architectures at a price that's barely enough for today's hardware

It's not the season to be jolly yet, so how about a season of silly suggestions instead?

www.pcgamer.com

Watched the vid and analysed it. This might be a little crazy but there are multiple issues (*) with it and this needs to be adressed:

The following is not in order.

*Calls out @Kepler_L2 for speculating that CDNA5 WGP mode could be brought to RDNA 5, when it's tech media being irresponsible and careless and reporting on it as fact. Also MLID isn't exactly a saint and overhypes AMD products. N2X and +7ghz Zen 6 on consumer come on.

Shares renderer of die

Reveals up to date specs

Shares BOM estimates for all PS6 gem SoCs. Used $20K wafer prices for N3P. Without AMD markup.
And lists likely PS5 Canis handheld and home console price ranges and Orion (PS6) price range.
My extrapolated die size for PS6 Orion around 270-280mm^2 or -10% vs launch PS5 on N7.

*The pricing estimate for PS5 SoC at launch is bizarre. Like N16 N7 had excellent ramp and d0 reduction curve. D0 below 0.1 in mid to late 2020. N7 stable at ~$10K/wafer since forever which would put SoC price at ~$60 not $105.

*No emphasis on retailer and distributor pricing overhead. This would sour selling at at lower price estimate even further.

Thinks Canis needs 24-36GB based on dev feedback

Reiterates claims about Sony wanting to bury PS4 for good.

*Game devs think console needs well above 30GB to run nextgen games, base this on current factors and doesn't include forward looking functionality. A few examples:
#1: LLMs in 2030s will be nothing like the ones in 2025. Look at token/$ and accuracy progress for mini-versions of LLMs, it's very fast. Devs could get away with tiny models in the future.
#2: GPU workgraphs (50-70X reduction in scratch buffers)
#3: Smarter and lower footprint BVH frameworks
#4: Neural compression for assets and textures. NTC is just the beginning everything will prob be AI compressed going forward.
#5: Smarter asset loading (Sampler Feedback streaming+ etc...)

With the rendering side changes at iso-fidelity VRAM usage can shrink >10X. Leaves plenty of room for the LLMs and Cerny def knows this since AMD is spearheading GPU workgraphs and they're collaborating on Project Amethyst.

*Not sure but it sounds like he thinks they'll go 40-48GB on PS6 Orion and Xbox Magnus. For Orion this requires 4GB x 5 clamshell over 160bit bus pr 4GB x 6 clamshell over 192bit bus. This isn't happening.This would end budget gaming for good and the nextgen consoles. $/GB isn't dropping remotely fast enough to make this a reality.

 

[Fallen Kell]

Please realize that the "*Game devs think console needs well above 30GB to run nextgen games" is talking about not just the graphics card but the platform as a whole, typically as a shared/unified memory architecture in these kinds of designs, where the CPU and GPU use the same memory pool. I don't doubt that it will take that, as we see most PC's have 16-32GB RAM and then anywhere from 8-32GB of GPU memory. And a PC that is able to run 4k games typically has at least 32GB RAM and 16GB GPU memory (so already 48GB and that is for current generation games at 4k).

 

[marees]

AMD rumored to use lpddr5x or lpddr6 based GPUs at the low end

Why is AMD using lpddr now for dGPU (timestamped video in quoted post below that raises this question)

Possible reasons

1. They are now faster — newest lpddr stuff will be as fast the gddr5 in base PS5
2. They are now not as slow as GDDR competition — they will only be half as slow as GDDR — which can be compensated by doubling the bandwidth
3. Improved caching — larger L1, L2 cache plus extra Infinity cache
4. They will not be capacity constrained — the latest & greatest gddr stuff always came with crippling capacity constraints which hopefully will not be the case with lpddr
   1. More vram for AI — professional cards can go upto 512gb vram
   2. More vram for gaming — marketing advantage like more megapixels in a camera
   3. No more excuses for low vram at budget — traditionally budget GPUs had less bus width such as 128gb (for cost & power saving) resulting in less gddr RAM. this will not be the case anymore with lpddr
5. Can be shared with laptop APU
   1. Medusa premium — shares 24 CU AT4 (10050xt) — this is a drop in replacement for strix point & will be used for all windows PC gaming handhelds, starting 2027-2028
   2. medusa halo — shares 48 CU AT3 (10060xt) — premium tablets / laptops / NUCs & edge inference use case, maybe a few $2500 handhelds too

Why is AMD using LPDDR for dGPUs now...and not before? (Timestamped video)

 

[igor_kavinski]

Can't wait to see the slim 10050 XT card with 32GB or even 64GB LPDDR6 for AYE AYE and LLM use in homelab setups!

 

[marees]

MLID continues to demand that next gen PS consoles (including handheld) should have 24gb to 40gb+ ram

Will Sony oblige ?

 

[igor_kavinski]

48GB would be just fine and dandy.

 

[Mopetar]

24 GB is likely sufficient for consoles unless they're trying to sell 8K gaming as an option. If they created a separate memory for the OS to use so that games could have the full 24 GB that would be enough for most titles.

More than that doesn’t seem feasible without making a much larger bus and that adds a lot to the cost. Expect both MS and Sony to lean on DirectStorage-like technologies and either add or enhance what they already have as a way to get around the limited memory capacity.

 

[DavidC1]

I play Expedition 33(UE5) with 8GB GTX 1080, a Pentium 10th Gen and 8GB system memory. I tried it originally on 4GB RX 470, and I knew immediately it was an issue. It does work on low, but certain parts really annoyed me. The 8GB RX 470 worked to remove that issue. 8GB system memory in the beginning would cause game crashes.

Getting extra 8GB of memory, but other than slower loading times, I don't get crashes anymore. Gets 20-35 fps which is acceptable. Med-High settings and 1440p with no upscaling. 8GB more system memory coming soon. I wish I could get better performance but I cannot complain about a $70 cdn card that I fixed.

 

[marees]

Additionally, with 16GB of VRAM, it could be positioned as an affordable GPU for running local AI applications. 🤔🤔🤔


Here's a look at the specs.




Read more: https://www.tweaktown.com/news/1077...e-radeon-rx-7700-with-16gb-of-vram/index.html

 

[igor_kavinski]

Not bad if one puts two of these in an AI rig.

 

[marees]

Memory is costly. 24gb is more than enough — Digital Foundry

 

[Golgatha]

3Gb chips on every tier would be sufficient currently.

e.g. 24GB on a 256bit memory interface, 18GB on a 192bit interface etc.

 

[marees]

3Gb chips on every tier would be sufficient currently.

e.g. 24GB on a 256bit memory interface, 18GB on a 192bit interface etc.

You mean gddr 7

But next medusa halo (10060xt) & medusa premium (this is used in xbox handhelds / series S, 10050xt) comes with lpddr RAM

AMD could use vram as a marketing tactic. (Hint the newly revealed 7700 16gb)

 

[Golgatha]

You mean gddr 7

But next medusa halo (10060xt) & medusa premium (this is used in xbox handhelds / series S, 10050xt) comes with lpddr RAM

AMD could use vram as a marketing tactic. (Hint the newly revealed 7700 16gb)

Whatever (higher density or more chips) gets us to that amount of VRAM or more would be sufficient.

 

[Mopetar]

3Gb chips on every tier would be sufficient currently.

e.g. 24GB on a 256bit memory interface, 18GB on a 192bit interface etc.

So you're telling me that NVidia will cut busses down to 96 or 64-bit because some people will insist that 8 GB is more than enough VRAM.

 

[jpiniero]

So you're telling me that NVidia will cut busses down to 96 or 64-bit because some people will insist that 8 GB is more than enough VRAM.

As long as nVidia feels like the sub-$500 market is worth pursuing... this is what you are gonna get.

 

[marees]

There ends the dream (of excess vram)

https://twitter.com/x/status/1974053425043140871

https://twitter.com/x/status/1974048796519649441