GPUs & CPUs & Enthusiast hardware: Questions, Discussion and fanboy slap-fights - Nvidia & AMD & Intel - Separe but Equal. Intel rides in the back of the bus.

Gerelatha said:
war thunder’s RT implementation is just as bad as their DLSS implementation, it’s a standout example of how not do do both of those things
Click to expand...
True. It's a good example of how these techs are not simply "toss it in, turn it on" as some imply.

Who knows, maybe when every single game is on UE5,6,7, whatever it'll be different.
 
Max Weber said:
Chinese company DeepSeek released a new AI model. It's free, it's open source, and more relevant to Nvidia specifically, it requires a lot less resources while outperforming most other models.
Click to expand...
Hahaha, didn't they move the doomsday clock up because of it? I'm hearing a lot of doom talk because of it. You can gurantee that AMD...will do nothing to capitalize on the moment, especially given the scam that is the 5090. They've always capitalized on not capitalizing.
 
  • Informative
Reactions: Flaming Dumpster, Brain Problems and Tri-Tachyon'sClown
Tri-TachyonRepresentative said:
Hahaha, didn't they move the doomsday clock up because of it? I'm hearing a lot of doom talk because of it. You can gurantee that AMD...will do nothing to capitalize on the moment, especially given the scam that is the 5090. They've always capitalized on not capitalizing.
Click to expand...
How long until the doomsday clock is forced to resort to fractional seconds? I'm guessing that in a couple years they will be 0.05 seconds from midnight
 
  • Like
Reactions: Another Char Clone and Tri-Tachyon'sClown
The Mass Shooter Ron Soye said:
AMD claims RX 7900 XTX outperforms RTX 4090 in DeepSeek benchmarks

AMD released instructions for running DeepSeek on Ryzen AI CPUs and Radeon GPUs
Click to expand...
Here is the link to the actual instructions: https://community.amd.com/t5/ai/exp...distilled-reasoning-models-on-amd/ba-p/740593
Sadly, their solution is Windows-exclusive and is only available for AMD's more recent hardware.
If your system doesn't meet those requirements, you can always use llama.cpp or koboldcpp for inference with the cross-platform Vulkan backend: https://github.com/LostRuins/koboldcpp
 
  • Informative
  • Like
Reactions: Tri-Tachyon'sClown and The Mass Shooter Ron Soye
Pattern Juggler said:
Wouldn't surprise if the scalpers are left holding the bag on the 5080s. The 5090s are all getting funneled to China and Russia, but I just don't think there's enough demand to justify the 5080 prices.
Click to expand...
The 7900 XTX is about 10% less perf than the 5080, with more VRAM. Seems to be still on the market, don't know if it's discontinued yet. The 9070 XT won't be very far behind the 5080 either.

It's likely that Nvidia will refresh the 5080 with 24 GB at some point, using the 3 GB GDDR7 modules already being used in the laptop 5090. The laptop 5090 is basically the 5080 with a few less CUDA cores, using the same GB203 die.
 
Last edited:
  • Like
Reactions: TheRedChair
Just Some Other Guy said:
Yeah I don't do RT. Turned it on in the few games I play that has it such as Hogwarts but wasn't blown away by it.
Click to expand...

Just Some Other Guy said:
I think War Thunder has it? If it did, I didn't find it remarkable and their DLSS is garbage anyways.

The spectrum of quality for RT implementation is still far too wide atm, unfortunately.

The real winning strategy is to not play AAA slop.
Click to expand...
Buggered the quote, but yeah, War Thunder has it, or at least the mobile version does. Remember it being mentioned in 2023 at the iPhone event.

As for me, I've tried it a bit. It's not bad, but I'd rather sacrifice RT for more/smoother framerates.
 
TheRedChair said:
I fucking can not stand this bullshit.
Click to expand...
Why, did you want to become 5090 Chad within a week of launch?

For gayming you can just use some lesser card, turn down settings, turn on upscaling, etc. DLSS4 Super Resolution will work on all RTX cards, and multi-frame gen might be added to 40 series. For AI, AMD's Strix Halo or Nvidia's Project Digits are probably going to be better at around $2k-3k, which is what most 5090s could be selling for during its lifespan.

The 5080 16 GB is irrelevant, other than being slightly faster and more power efficient than the 4080 Super.

geckogoy said:
I see the globalist hyperspecialization strategy is panning out well now that all chips are made on one tiny island
Click to expand...
It's not responsible for 5000 series shortages, and TSMC is about to start production in Arizona.

If Intel doesn't screw up 18A, companies including Nvidia could start using that. That node should at least be better than the TSMC N4P that Blackwell is being made on, even if it's not as good as TSMC N2/A16/etc. Nvidia is rumored to be using Intel 3 this year as well.
 
  • Agree
Reactions: Brain Problems
The Mass Shooter Ron Soye said:
It's not responsible for 5000 series shortages, and TSMC is about to start production in Arizona.

If Intel doesn't screw up 18A, companies including Nvidia could start using that. That node should at least be better than the TSMC N4P that Blackwell is being made on, even if it's not as good as TSMC N2/A16/etc. Nvidia is rumored to be using Intel 3 this year as well.
Click to expand...
The 5000 series is TSMC only, how is it not TSMCs fault
 
geckogoy said:
The 5000 series is TSMC only, how is it not TSMCs fault
Click to expand...
It's Nvidia's fault for paper launching the card. They could have built up supply and launched later.

TSMC makes plenty of chips for Nvidia. Nvidia would rather sell a B200 AI accelerator for $40,000 than two RTX 5090s for $4,000.
 
  • Agree
  • Like
Reactions: milk, Wooo, Brain Problems and 1 other person
TheRedChair said:
I'll put this here as it is important. The 5000 series is pretty much limited/vaporware at launch.
Click to expand...
this is why I’m glad retailers in my country typically allow back orders, managed to snag a back order on a 5090 right as they went live so I’m probably pretty close to first in line and the site says it’s expected by the end of feb
reminds me of the ampere situation
 
  • Like
Reactions: N Space
The Mass Shooter Ron Soye said:
It's Nvidia's fault for paper launching the card. They could have built up supply and launched later.

TSMC makes plenty of chips for Nvidia. Nvidia would rather sell a B200 AI accelerator for $40,000 than two RTX 5090s for $4,000.
Click to expand...
Seems like at the end of the day there was simply not as much production as hoped for and nvidia decided to launch on the pre-defined schedule anyways rather than delay
 
geckogoy said:
Seems like at the end of the day there was simply not as much production as hoped for and nvidia decided to launch on the pre-defined schedule anyways rather than delay
Click to expand...
ampere was a similar paper launch though I’ve heard blackwell will have faster restocks than ampere did, remains to be seen I guess
 
You must log in or register to reply here.
Back