Daniel Pina

1K posts

Daniel Pina

Daniel Pina

@PinaJr

Katılım Haziran 2009
152 Takip Edilen44 Takipçiler
Daniel Pina
Daniel Pina@PinaJr·
@carygolomb I'm not a gamedev but I'd say lots? You'd need a lot more context per-object (texture, transparency, etc.), context for light sources, much more context for faces and facial expressions. GenAI for fairly homogeneous textures is one thing, but this would be a whole other beast.
English
1
0
3
66
Cary Golomb
Cary Golomb@carygolomb·
@PinaJr How much more detail do you think is needed on the base to compensate
English
1
0
1
422
Cary Golomb
Cary Golomb@carygolomb·
This is not today. But certainly someday you can imagine this eventuality What's the frame budget on the top part? Gotta crazy fast render. How long did it take to make? Gotta be way faster To be clear, I really don't want this at all and I hope I'm wrong.
Sergiy Galyonkin@galyonkin

What I really want from DLSS5:

English
45
7
202
16.5K
Daniel Pina
Daniel Pina@PinaJr·
@IanCutress Assuming this actually happens, it won't invalidate people's opinions on GTC's demo reels for DLSS5. Fact is lighting and tonemapping were all over the place (and yes, by looking like all other "DSLR ultrarrealist" prompts), and people *can* criticize DLSS5 on those merits.
English
0
0
4
319
𝐷𝑟. 𝐼𝑎𝑛 𝐶𝑢𝑡𝑟𝑒𝑠𝑠
𝐷𝑟. 𝐼𝑎𝑛 𝐶𝑢𝑡𝑟𝑒𝑠𝑠@IanCutress·
If GTA6 launches as a DLSS5 native title.... it'll still sell 50 million copies half the people will have DLSS5 running most of them won't realise it here's betting most of them enjoy the game and marvel at the next gen graphics
English
75
4
188
52.6K
Daniel Pina
Daniel Pina@PinaJr·
@AncientGameplay People do want ultra realistic lighting in many games. It's just that ultra realistic lighting isn't supposed to look the same everywhere. Nvidia went all in on the "DSLR ultrarrealism" prompt people use on genAI image/video. Makes us wonder if the model can do otherwise.
English
0
0
1
162
Ancient Gameplays
Ancient Gameplays@AncientGameplay·
It is just me or the DLSS 5 shots just make me feel "odd" or "unsettled". They're just over the top, and to be honest ALL games shown so far, have the same type of lighting applied, which makes them all feel the same... Same tone, same lighting, everything the same... What I see is NVIDIA controlling game developing here... And in most shots? The "normal" game looks much better with DLSS 4. Artistically speaking. People don't want ultra realistic lighting and faces in all games.
English
58
23
441
17.4K
Daniel Pina
Daniel Pina@PinaJr·
@SebAaltonen Hate to be the pedantic one, but the PS4 was 1.84 TFLOP/s. Nowadays it sounds silly to argue about 0.4 TFLOPs, but it's almost 30% difference😉 Also, PS4's OS tasks took a whole 3GB? That's gigantic!
English
0
0
3
513
Sebastian Aaltonen
Sebastian Aaltonen@SebAaltonen·
Overwatch originally shipped on PS4: - 1.4TFLOPs/s GPU (upper mid tier phone today) - 5GB RAM (for games) PS4 ran Overwatch flawlessly at locked 60Hz. Blizzard knows how to write optimized code. My 9 year old PC GPU (AMD Vega 64) ran Overwatch at stable 144Hz (2560x1440).
Vadim Yuryev@VadimYuryev

Alright now THIS is impressive. A $500 MacBook with excellent build quality that's being marketed as a web browsing laptop for students.. is running games like Overwatch at 60FPS?!

English
22
36
1.4K
78.9K
Daniel Pina
Daniel Pina@PinaJr·
@carygolomb If both consoles come within the same form factor and power budgets, it's reasonable to assume the PS6 with a smaller chip can afford higher clocks.
English
1
1
2
355
Cary Golomb
Cary Golomb@carygolomb·
We still don't know clocks, or fundamental uArch of PS6 vs Xbox Helix iirc, Sony boosted clocks on GPU in the 11th hour for PS5. Going to 2.23ghz closed the compute gap with Xbox Series X cROPs on PS5 is 22% higher!! than Xbox Series X AND dROPs are over 2x!! on PS5 over XSX
English
19
4
80
11.4K
Daniel Pina
Daniel Pina@PinaJr·
@TDevilfish @Olrak29_ Leaving out support for FSR4 that increases IQ for GPUs and iGPUs released as recently as last week (AI 400 with RDNA3.5), when the perfectly capable and stable INT8 version was leaked a year ago, is anything but basic product segmentation. It's not normal.
English
1
0
2
133
Gray
Gray@Olrak29_·
As AMD is radio silent about official FSR4 on RDNA 3, there's nothing you can do about it
English
16
7
177
24.3K
Daniel Pina
Daniel Pina@PinaJr·
@SebAaltonen @lafaiel I'm not sure gamers "should avoid Strix Halo" altogether. Many would kill for e.g. a single-CCD + 40CUs Halo handheld with unmatched performance at 30W with linear fps-per-watt performance up to ~65W to use in docked mode. Neither dGPUs nor PTL-H can reach that capability.
English
0
0
0
50
Sebastian Aaltonen
Sebastian Aaltonen@SebAaltonen·
@lafaiel Is the test using hardware ray-tracing? Strix Halo is RDNA 3.5. Doesn't have the latest RT tech. Big improvement in RDNA4. Another reason why gamers should avoid Strix Halo is the lack of fp8 WMMA ops, which are needed in FSR4.
English
2
0
12
1.8K
INIYSA
INIYSA@lafaiel·
Is the Blender version used here HIP-optimized? AMD AI Max 395's 8060s is way more powerful than the M5 GPU, yet it's still slow. Seems to show results similar to the Blender 4.5 benchmark?
INIYSA tweet media
English
11
3
86
8.5K
Jacob Freeman
Jacob Freeman@GeForce_JacobF·
GeForce On - GDC Update is tomorrow!!! Tune in Tuesday, March 10 at 8 AM PT for some new updates on RTX games, features and more! 😀 youtube.com/watch?v=bqraHG…
YouTube video
YouTube
English
17
14
104
25.1K
Omer Cheema
Omer Cheema@OmerCheeema·
PC OEMs are complaining about Intel Pantherlake supply issues. While Intel claims it is prioritizing servers hence PC chips are deprioritized, PTL is the only product on 18A. PTL shortage is probably driven by low yields. AMD will gain share. Source: PC OEMs.
English
22
10
125
61.2K
Daniel Pina
Daniel Pina@PinaJr·
@carygolomb Take away RAM and NAND's current prices due to shortages and there's zero reason to believe a small tower with a cut-down N33 made on cheap 6nm and a tiny HawkPoint2 would cost more than $400 in BoM + assembly. Put a 30% markup in there and there's your $500-600.
English
1
0
3
457
Cary Golomb
Cary Golomb@carygolomb·
Xbox Helix doubles down on Steam Machine needing to be $500-600 If it's any higher, it will have a short shelf life.
English
71
8
213
18.7K
Daniel Pina
Daniel Pina@PinaJr·
@SebAaltonen @SaifuddinAmri__ @grok There's no way they could fit a 750mm^2 large GB202 with 512bit GDDR7 engineered for >500W into a laptop. GB203 using 3GB chips clamshell would get 48GB. I'm not sure dGPUs in laptops make sense for local AI inference though. SoCs with 256bit LPDDR seem more viable IMO.
English
0
0
3
111
Sebastian Aaltonen
Sebastian Aaltonen@SebAaltonen·
@SaifuddinAmri__ @grok 24GB limits your model size a lot. It's a shame that Nvidia doesn't bring RTX Pro 6000 96GB equivalent GPU to laptops.
English
2
0
16
2.2K
Daniel Pina
Daniel Pina@PinaJr·
@phoronix GMA500 was PowerVR SGX, right? So this was Intel's first ever solution carrying a GPU with (slow) GPGPU compute abilities lol.
English
0
0
0
99
Phoronix
Phoronix@phoronix·
Intel GMA500 "Poulsbo" Driver Still Seeing New Open-Source Activity In 2026 Intel Atom Poulsbo still makes me cringe. phoronix.com/news/Intel-GMA…
English
2
1
43
3.7K
Daniel Pina
Daniel Pina@PinaJr·
@Mareeswj @carygolomb @MajesticBowler AFAIK there's no such thing. Also, planning for a "low-cost videogame SoC" would be a fruitless endeavor considering they can't really make a low-cost handheld anyway due to RAM prices. Just look at Intel dragging out the release of Panther Lake G3 for the same reason.
English
0
0
1
45
JoiningUnrelatedDots
JoiningUnrelatedDots@Mareeswj·
@carygolomb @PinaJr @MajesticBowler what do you guys mean the z-series is dead ? I expect medusa premium to be z3e (possible launch at CES 2028) x.com/PinaJr/status/…
Daniel Pina@PinaJr

@carygolomb @MajesticBowler Perhaps that's the main reason why the Z-series is apparently dead. Selling cheaper SoCs while expecting recouped costs from for each new driver certification just resulted in no updates because OEMs couldn't care. And then people became angrier at AMD than at Asus and Lenovo.

English
1
0
0
20
Daniel Pina
Daniel Pina@PinaJr·
@malliberal Gosto de tudo menos da música de fundo. Tende a diminuir um bocadinho os argumentos quando estes são bons, na minha opinião.
Português
0
0
0
398
MAL
MAL@malliberal·
Doente com cancro. Incapacidade de 60% reavaliada em 52%. AT retira-lhe benefícios fiscais. Cinco (5!) acórdãos do Supremo Tribunal Administrativo decidem CONTRA a AT em outros processos similares. Tribunal Constitucional pronuncia-se contra a AT. O que faz a AT, conseguem adivinhar?
Português
25
39
327
15.3K
Daniel Pina
Daniel Pina@PinaJr·
@carygolomb @MajesticBowler Perhaps that's the main reason why the Z-series is apparently dead. Selling cheaper SoCs while expecting recouped costs from for each new driver certification just resulted in no updates because OEMs couldn't care. And then people became angrier at AMD than at Asus and Lenovo.
English
1
0
1
105
Cary Golomb
Cary Golomb@carygolomb·
@MajesticBowler Yea, it's generally why I mostly prefer "standard" APUs from AMD Just better off getting 7840U instead of Z1E..effectively both the same thing, yet only one gets mainline drivers.
English
2
0
5
260
Daniel Pina
Daniel Pina@PinaJr·
@carygolomb The big problem started last year when @AMD started to block driver sideloading in their newer GPU drivers. Sounds a bit like they're trying to punish their handheld partners for not paying for new driver certification, but ends up being a shot in the foot for their popularity.
English
0
0
3
156
Daniel Pina
Daniel Pina@PinaJr·
@SebAaltonen @atla_ Models that run on consumer HW are scaling quicly while big server-only frontier / foundation models are becoming stagnant due to data incest. At the same time, the market is boosting server HW while asphyxiating consumer HW. Things are about to get fun.
English
0
0
2
148
Sebastian Aaltonen
Sebastian Aaltonen@SebAaltonen·
@atla_ Not beating $200/month big/slow frontier models. Beating the fast models. And beating GPT 4o, which was the best big model one year ago. That's impressive. People finally started optimizing their LLMs. Brute force didn't scale.
English
4
0
46
3.3K

Keşfet

@carygolomb @IanCutress @AncientGameplay @SebAaltonen @TDevilfish @Olrak29_ @lafaiel @VideoCardz