I apologise for my curt response, but your "
What absolute nonsense!" did get my back up.
I'll carry on the discussion/disagreement in a civil manor.
To be fair I don't even think TMSC is the biggest issue here, I'll get to that point at the bottom.
I know you never explicitly stated that Nvidia make their own GPU's but that doesn't stop your claim being nonsense. In fact its the very reason it is nonsense because otherwise NVidia could ramp up production of their current GPU's without it being necessarily at the expense of other work from different companies (assuming in such a scenario that Nvidia only manufactured their own product)
I didn't remotely say they made their own GPUs, I still can't work out how you came to that conclusion when I even mentioned competing again other high end goods and smartphones. Which obviously Nvidia don't make.
Moving production of their most profitable(as in % per unit sold) products like their professional cards wouldn't be an option, so I don't know why you'd bring that up. I'm pretty sure they get top priority.
So care to argue why Nvidia couldn't have more Pascal GPU's made if they wanted to? (even if this was at vastly increased costs to themselves)
Why would they need increase the costs vastly if the fabs have spare capacity as you say?
For example if I am talking about Nvidia GPU's saying it would be unwise for Nvidia to increase production and you say 'but production is already running at maximum' that implies that you are saying that the company making the GPU's (i.e. TSMC) are making Nvidia GPU's at their maximum capacity....
Which they are not as shown so you were wrong!
Are you seriously going to suggest that Nvidia could not pay TSMC more money to even just increase production by a relatively modest percentage? What with Nvidia being a long standing and rather large partner of TSMC?
Do you seriously think that Bitman can book up all the spare slots for TSMC fab production for the foreseeable future when their entire business model relies on selling a product the demand for which could evaporate over night?
I did not imply that if TSMC are at their maximum capacity of making GPUs if they stopped making products for other people. I said the FABs are at maximum capacity, so of course that would include all their other customers work.
I probably shouldn't have mentioned paying more as that probably has nothing at all to do with it and is a red herring.
I'm pretty sure like any respectable business once a supply order has been taken and confirmed they would not bump it because someone offered more money long term partner or not.
The Bitmain order from what I have read does indeed look like they have taken everything they can get in January/February having already taken everything they could get in November and December. These will only be short term orders. Speed is of the essence for them. This isn't about the foreseeable future, but now, I'm sure there is available capacity in the future at TSMC if you order now.
So in answer to you question, not I don't think bitmain has booked all the spare slots for the foreseeable future (if that means after February), but this is about what is available now.
But going back to my original point that I think the fab is probably at full capacity and you don't. So why would you even bring up pay more as an option. This doesn't match your stance or has that change now?
@ Nano2k thanks for the heads up......
article is here..........
@Sasahara 'Production is at maximum capacity?'
'When asked about increasing card production, we were given a few primary answers:
- Betting on cryptocurrency is a big bet, because the mining market has not proven stable, and has proven unpredictable. To order on current demand doesn’t mean those cards show up tomorrow – they show up in a few months, and if cryptomining dies down in that time, that’s a big bag to be left holding. The manufacturers are forecasting months ahead (quarterly, actually), not weeks ahead.
- NVIDIA and AMD could likely produce more GPUs, but board partners need to actually place an order for them. We’ve seen some uninformed content creators online accuse nVidia and/or AMD of undersupplying the market. Well, nVidia and AMD need a customer to sell those GPUs to – that’d be the board partners. If they’re not ordering more, nVidia and AMD aren’t going to make more. Simple as that.
- New supply is showing up weekly, but it’s selling out fast. The best bet is to show up at a local retailer and ask when the next shipment comes in, then go there that day.
- There is concern about over-production, especially if mining falls enough that the second-hand market becomes flooded, outstripping the ability of first-parties to make money.'
'Board partners seem to think that, if the partners placed more orders for cards, nVidia (at least) could sustain more orders.'
Oh dear!
So Nvidia could (likely) increase the production of their GPU's but only if their customers (primarily the AIB's) were ordering the GPU's which the are not as they realise they could easily be left holding lots of stock which they will find hard to sell.
So what I was saying then? Nvidia could (likely) up production but wont because its unwise to do so! QED
Yes they think in months ahead because they can't just turn the taps on and instantly increase production. They have to wait for the fabs TSMC and RAM suppliers which is probably the most significant bottleneck in the supply chain, to have free capacity. TSMC have order books and waiting lists and Bitmain will be a big chunk of that, and no they would just cancel a contract because a bigger player asked.
So I stand by my statement the
FABs are at maximum capacity.
Now obviously it would not take years to get some extra production. But even a month delay isn't an option because the demand is now. If there is as you say, spare capacity they wouldn't need to pay more money?
It's also telling that in the article they did 6 months ago, it mentioned it was too risky to increase production.
But in the 3rd quarter, aka the quarter after that initial article, nvidia and their partners increase shipments by 29.5%. So someone took some risks and production was increased.
So we can take what Steve gets told and what actually happens with a little pinch of salt. He was probably told in good faith.
Nothing in the article either confirms or denies if TSMC is at maximum capacity. But as a general rule the cutting edge fabs are always run close to maximum capacity. It the sign of an economic down turn if they are not.
Now all of that may be irrelevant anyway, because the real problem is likely to be the supply of GDDR5(x). It's not just a TSMC issue.
There is a industry wide shortage of ram production. I'm pretty sure you know that.
Some of the VRAM production was repurposed to make smartphone memory. This is what cause the jump in price and lowered the amount available.
Do you agree that RAM production FABs are at maximum capacity and this would limit GPU production?
