5 ms·

AMD hasn't scaled up their production to meet that demand, and neither has Nvidia; they're both wary of a crash causing demand to stop as miners sell their burn

AMD hasn't scaled up their production to meet that demand, and neither has Nvidia; they're both wary of a crash causing demand to stop as miners sell their burned-out GPUs. Right now that makes things hard for everyone not a miner, but if AMD ends up being the only option for gamers while Nvidia still sells to miners, they might actually gain from the mess.

philjohn 9y ago
Memory shortages have also played their part.
paulmd 9y ago
The thing to remember about NVIDIA is that their production levels are much higher to begin with due to their overwhelming marketshare. They have a lot more inventory in the channel at any given time, which gives them more of a buffer when GPU mining heats up. I actually think AMD's timing on increasing production is really questionable here. Barring another major runup in Bitcoin prices (altcoin profitability follows Bitcoin prices), which seems unlikely at the moment, network difficulty will continue to increase and mining profits are going to continue to decline. That means miners are going to taper off purchases, and some may even be dumping their rigs. So AMD is essentially increasing production at the exact moment when they may already have a problem with the market being flooded. The time when they needed to increase production was 4 months ago, it's too late now. On top of that, NVIDIA may be pushing out their new GPUs within the next few months, implying another step in efficiency which puts AMD's performance/efficiency even further behind. They are already about a generation behind (roughly competitive with Maxwell), despite being on similar nodes. I'd be prepared for NVIDIA to make a minimum of a half-gen step (30%), if not a full-gen step (60%). AMD's timing seems exquisitely poor here. They didn't want to bet on crypto, and now it's too late. Fortune favors the bold.
sangnoir 9y ago
> they're both wary of a crash causing demand to stop as miners sell their burned-out GPUs Offtopic FYI: GPUs that are in use 24/7 would probably in a better shape than those that encounter more thermal (on/off) cycles. There's obviously more wear and tear on the moving part, but fans are easier to replace than mechanically stressed silicon/PCBs.
- Animats 9y ago
  Do they fail from thermal stress or electromigration?
- Yizahi 9y ago
  a) it is not at all certain since both processes do damage to board and chips and not a single person in this thread knows what will be more severe for each small part of the board or connections. b) have you actually tried replacing monstrous coolers on moderns cards? Or even buying spares? It is rare, expensive, hard to disassemble and hard to fit back. Some cards even use glue stickers on memory thermo-interface, you can potentially tore chips away.
  - Fnoord 9y ago
    Regarding B I just checked iFixit for Radeon and found 2 (two!!) guides for graphics cards. One for reapplying thermal paste, and one for fixing a noisy fan with I think tape. All the other guides are for laptops specifically. If you know any other good written guides (pref not videos) available to clean and/or replace fans on graphics cards I'd like to learn about them.

philjohn 9y ago

Memory shortages have also played their part.

paulmd 9y ago

The thing to remember about NVIDIA is that their production levels are much higher to begin with due to their overwhelming marketshare. They have a lot more inventory in the channel at any given time, which gives them more of a buffer when GPU mining heats up. I actually think AMD's timing on increasing production is really questionable here. Barring another major runup in Bitcoin prices (altcoin profitability follows Bitcoin prices), which seems unlikely at the moment, network difficulty will continue to increase and mining profits are going to continue to decline. That means miners are going to taper off purchases, and some may even be dumping their rigs. So AMD is essentially increasing production at the exact moment when they may already have a problem with the market being flooded. The time when they needed to increase production was 4 months ago, it's too late now. On top of that, NVIDIA may be pushing out their new GPUs within the next few months, implying another step in efficiency which puts AMD's performance/efficiency even further behind. They are already about a generation behind (roughly competitive with Maxwell), despite being on similar nodes. I'd be prepared for NVIDIA to make a minimum of a half-gen step (30%), if not a full-gen step (60%). AMD's timing seems exquisitely poor here. They didn't want to bet on crypto, and now it's too late. Fortune favors the bold.

sangnoir 9y ago

> they're both wary of a crash causing demand to stop as miners sell their burned-out GPUs Offtopic FYI: GPUs that are in use 24/7 would probably in a better shape than those that encounter more thermal (on/off) cycles. There's obviously more wear and tear on the moving part, but fans are easier to replace than mechanically stressed silicon/PCBs.

Animats 9y ago
Do they fail from thermal stress or electromigration?
Yizahi 9y ago
a) it is not at all certain since both processes do damage to board and chips and not a single person in this thread knows what will be more severe for each small part of the board or connections. b) have you actually tried replacing monstrous coolers on moderns cards? Or even buying spares? It is rare, expensive, hard to disassemble and hard to fit back. Some cards even use glue stickers on memory thermo-interface, you can potentially tore chips away.
- Fnoord 9y ago
  Regarding B I just checked iFixit for Radeon and found 2 (two!!) guides for graphics cards. One for reapplying thermal paste, and one for fixing a noisy fan with I think tape. All the other guides are for laptops specifically. If you know any other good written guides (pref not videos) available to clean and/or replace fans on graphics cards I'd like to learn about them.