Home
Artists
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register
Partychan Telegram ThePornDude
Home Artists Posts Import Register

Broken Silicon (Video): Intel Arrow Lake vs AMD Zen 5, Nvidia DLSS 4, RDNA 3 & RTX 4000 Legacy | Vex (Patreon)

Published:
2024-03-04 22:15:53
Imported:
2024-03
Tags:

Content

Vex joins to discuss the legacy of RTX 4000 and RDNA 3, Intel ARL, and future DLSS.


0:00 What inspired Vex into making YT videos?

7:37 What's driving growth in opinionated tech channels?

19:46 GPU Competition in 2024, and the Legacy of RDNA 3 & Lovelace

30:45 DLSS 4 – “Dynamic Frame Generation” could be the Holy Grail

44:24 AMD vs Nvidia Software, Modern Cooler Quality, RX 7900 GRE

56:32 Intel Arrow Lake vs Zen 5 - Can ARL take back mindshare?

1:16:56 Is the PC Gaming Market Healthy in 2024?

1:26:05 Recent PlayStation Layoffs amidst Helldiver 2’s Success

1:37:57 What games have we been playing?



Check out Vex on YouTube: https://www.youtube.com/@vextakes

RDNA 4 “Not Cancelled” Explanation: https://youtu.be/siG6TJtLztc?si=FROpjCKXz_tsPDNV

Navi 48 Details: https://youtu.be/Hbx4AUcQ5do?si=DCBxep_gksdaGaUR

Files

AD FREE Broken Silicon 247

@vextakes joins to discuss the legacy of RTX 4000 and RDNA 3, Intel 15th gen, and the future of DLSS. 00:00 Meeting Vex, how and why he started 7:37 What could be driving the recent tech world growth, Switching to the 6800XT 19:46 The state of GPU competition, why isn’t AMD more aggressive 30:45 RDNA 4 and AMD innovation, The ideal DLSS/FSR 44:24 AMD and Nvidia experiences, hardware quality improvements, following the GRE trend 56:32 Intel vs AMD competition, upgrading paths 1:16:56 Comparing the health of the PC market 1:26:05 Are games driving GPU sales, layoffs and their consequences 1:37:57 What games have we been playing Check out Vex on YouTube: https://www.youtube.com/@vextakes RDNA 4 “Not Cancelled” Explanation: https://youtu.be/siG6TJtLztc?si=FROpjCKXz_tsPDNV Navi 48 Details: https://youtu.be/Hbx4AUcQ5do?si=DCBxep_gksdaGaUR

Comments

Thomas Burchsted

I will say Steve from GN will definitely say if a product is crap or poor value lol.

capn_hector

Haha, the cooler is the cheapest part of the card. A whole big 3090 or 4090 cooler might be $50-75 all-up - that’s why partners like big coolers, it’s a cheap way to segment their cards that doesn’t cost them much actual money. A whole raijentek Morpheus II is like $80 MSRP at retail, and they have to make all their markup and shipping costs on that one item. And while that doesn’t include fans… vendors aren’t using Noctua either, they’re the cheapest shit fans they can get. Also, running at higher power generally yields higher margins because people will pay for performance and they won’t pay for efficiency. So yeah you save $5-10 on the cooler, but then you have to sell the card for $100 less. You spend like half the episode suggesting “if they made the cooler cheaper they could make a 4070 that’s 10% slower for $400 instead of $500” and that’s not supportable by the cost you’ve given there. That’s literally 20% of the cost of the card, the cooler would have to cost like 40% of the total msrp to produce before you could get those kinds of cuts and it simply doesn’t work like that. Again, just look at the raijintek Morpheus II. That’s a better cooler than most 4070s have, it’s rated for 360W, from a boutique manufacturer selling the cooler standalone. How do you get $100 of cost savings out of a product that would cost $100 *total at retail* with 2 cheap fans included? The numbers don’t remotely add up. There is no one magic bullet to reduce costs by >25%, or AMD would definitely have pulled that lever already. Things just cost more in a post Moores law and post pandemic era.

capn_hector
>>131675888

Also just in general, there is a meme these days that “40-series is pushed too hard” and that’s not really supportable either. As far as I can tell it springs from comparisons that amount to “GTX 670 and 1070 used less power for the same size die” and yeah, that’s cause dennard scaling ended in the early 2000s and power density climbs every time you shrink now. A 300mm2 die on 4nm just pulls more power than a 300mm2 die on 40nm did, and that’s counterintuitive but that’s how it works now. You’re getting 10x more transistors but the transistors aren’t 10x more efficient, so it pulls more power. Further, the whole idea is in direct contradiction to the meme that “nvidia wanted to release a 900W TGP 4090” (kopite7kimi) and that nvidia somehow “backed down at the last minute” (lol nice footwork there). Like they backed down but it’s also still “pushed really hard”? Contradictory. (And yea I’ve seen the horizontal PCB card, that’s not proof of anything in particular other than nvidia is looking at PCB sag imo. And I don’t doubt partners had thermal samples that went to 600w or whatever - why wouldn’t they have those? But it doesn’t mean that’s the product that was gonna ship.) Ada of all generations is not pushed particularly hard. Really Turing, Maxwell, and pascal all are pushed about the same imo. (And there is a related problem that the voltage wall is getting steeper and steeper on newer nodes… so there really isn’t a *point* to pushing super hard anymore. Undervolting is where it’s at now.) But the problem here is this means you can’t even measure things like “how far is it from the voltage wall” as a proxy, because those things have changed over time too…