I'd rather see them make more PS5's with that precious limited TSMC silicon, but sure AMD you do you. I can be patient.

They're a for-profit business, they need to focus on whatever generates the highest margins. After years of Intel blowing billions to suppress them AMD still needs to grow financially.

JAB CreationsThey're a for-profit business, they need to focus on whatever generates the highest margins. After years of Intel blowing billions to suppress them AMD still needs to grow financially.

This goes through satisfying the customers' needs and requests. Focusing on high margins doesn't necessarily improve a company's financial situation.
AMD needs to find a way to make as many chips as the actual, real demand says.

ARFThis goes through satisfying the customers' needs and requests. Focusing on high margins doesn't necessarily improve a company's financial situation.
AMD needs to find a way to make as many chips as the actual, real demand says.

Getting Navi 22 out the door should be a good step towards this - the smaller die size means dramatically higher numbers of good dice per wafer, so there'll be more GPUs on shelves from the same number of wafers. Navi 21 is ~18x29mm (~520mm²), so let's assume this is more like 18x18mm= 324mm² (the PR die shots/renders look very square, and given memory controllers and other non-CU parts aren't halved, it will be more than half the size of that. Comparing the two in awafer yield calculator with known parameters for TSMC 7nm (0.09 defects/CM², standard scribe line and exclusion zone sizes) gives the following output for a 300mm wafer:
Navi 21: 96 complete dice per wafer; 35 with some defect(s), 61 defect-free.
Navi 22: 164 complete dice per wafer; 41 with some defect(s), 123 defect-free.
That's a 2x increase in good dice, and likely the defective ones can be stocked for later binning into lower-end cut-down products once there is sufficient supply. Given what AMD has said about current yields from these GPUs, it seems likely that the defect rate is even below 0.09, or that they have sufficient mitigations in place to be able to use a portion of defective dice for the top SKU as well (hence them not launching a 6700 non-XT (yet)).

Given that this is still at the very upper reaches of consumer pricing ($479 is solidly high-end still, even if it's lower than $579 and $649), there are good chances that supply will actually be all right - though that obviously depends on miners and scalpers. At least Navi 2X is less attractive to miners than Ampere due to the lower memory bandwidth, so that's something. It should perform somewhere around the RTX 3070, after all, so it'll be a very high performance GPU.

Of course the real kicker will be once they get Navi 24 out the door. If that's a ... let's say 28CU die (assuming a cut-down Navi 22 SKU with ~34-36 CUs is slated to arrive at some point) with 64MB of IC and a 128-bit memory interface, that could be a much smaller die again. If they get it down to around 200mm² - let's say 18x11mm, as rectangular dice tend to yield better from what I've read - that would again double per wafer outputs: 282 total dice/wafer; 45 w/defects, 237 defect-free. That would make for a fantastic laptop GPU at reasonable clock (<2GHz), as well as a very attractive 6500 series of GPUs. Hopefully those could deliver some real value in the~$200 segment on desktops as well - if they clock as high as the 6700 XT, a 28CU 6500XT could come pretty close to the 5700 XT in real-world performance.

JAB CreationsThey're a for-profit business, they need to focus on whatever generates the highest margins. After years of Intel blowing billions to suppress them AMD still needs to grow financially.

That's just accepting the fundamental logical fallacy of the current dominant US(-originated) mode of economics: that margins are somehow not connected to actually delivering attractive products that people want and building customer trust over time. If you have the clout you can squeeze them without doing so, but only for a short while before things come crashing down. If AMD needs to grow financially, they need to do so by building said growth on a solid base of attractive products, high sales volumes, satisfied customers and a solid reputation. They have some attractive products currently, but mostly ones out of the reach of 90% of GPU buyers, meaning they're not getting high sales volumes, (high volumes of) satisfied customers, or a solid reputation from it. There's still a ways to go, and expanding to lower reaches of the market is a necessity for this to happen. The current supply crunch makes the relatively slow launch pace of this generation understandable, but they need to hit the $300 and $200 price points within the next few years if they don't want to risk alienating and disillusioning the vast majority of their customer base.

sure... also desktop 6800 cards are "coming soon"...

I thought they were going to use the Navi23 and 24 chips for laptops?

At $479 for the cheapest 6000-series card, AMD desperately need something to fill the $200-400 price points!

yawn...

Chrispy_I thought they were going to use the Navi23 and 24 chips for laptops?

At $479 for the cheapest 6000-series card, AMD desperately need something to fill the $200-400 price points!

Yeah, they really need to get moving on smaller dice. I'm thinking they're "meeting" the supply squeeze by waiting a bit longer than normal with those though. Nothing egregious yet - a normal cadence would have been something like October - January - March, but instead we're seeing october - march - ???, and with the March launch still solidly in the high end in terms of price. Hopefully the last step will be May-June and not later, and they'll flesh out more of the stack at that point. It might be that they are already producing Navi 24, but are stockpiling it for mobile use before they launch desktop SKUs.

AMD have known the situation of the market since Nvidia premiered the 3000 series in September and if they have been uncapable to increase production capacity in 6 months there is some thing wrong with their business model.

They can blame what ever they want but it is still their responsibility to meet demand in the market, if they fail they did some thing wrong. They can try to blame the pandemic or what ever but it is still up to them to find ways around the problem.

As an example in the auto industry lots of car manufacturers are now experiencing component shortages EXCEPT FOR HYNDAI. BUT HOW COME HYNDAI IS NOT AFFECTED. I will tell you why, they always asses the market regularly to adapt their purchase quantity's for components and when the detected that shortages where coming they increased purchase volumes and are now the only car manufacturers without component shortage while all other struggles getting component.

The take away lesson here is that if you do not adapt quickly it is your own fault, you can blame what ever but it is still your own fault for sitting on your hands when action was needed to avoid undersupplying the market demand.

fynxeruncapable to increase production capacity in 6 months there is some thing wrong with their business model.

You really think TSMC's not booked to the tee? Heck AMD themselves are prioritizing zen3 chips right now!

www.techpowerup.com/279301/tsmc-reportedly-auctioned-off-excess-capacity-at-a-15-20-price-premium

fynxerAMD have known the situation of the market since Nvidia premiered the 3000 series in September and if they have been uncapable to increase production capacity in 6 months there is some thing wrong with their business model.

They can blame what ever they want but it is still their responsibility to meet demand in the market, if they fail they did some thing wrong. They can try to blame the pandemic or what ever but it is still up to them to find ways around the problem.

As an example in the auto industry lots of car manufacturers are now experiencing component shortages EXCEPT FOR HYNDAI. BUT HOW COME HYNDAI IS NOT AFFECTED. I will tell you why, they always asses the market regularly to adapt their purchase quantity's for components and when the detected that shortages where coming they increased purchase volumes and are now the only car manufacturers without component shortage while all other struggles getting component.

The take away lesson here is that if you do not adapt quickly it is your own fault, you can blame what ever but it is still your own fault for sitting on your hands when action was needed to avoid undersupplying the market demand.

That isn't a valid comparison - AMD isn't buying components from a supplier like Hyundai is doing; they ARE the supplier. There is no indication that AMD hasn't booked all TSMC capacity they could get their hands on. They could of course re-spin some of their chip designs for a different foundry or node, but that's not something you get done in 6 months from scratch. And of course that again assumes other foundries have capacity, which there is no indication of. And besides, the Hyundai tactic is a short term tactic at best - it alleviates short term shortages, but once that initial inventory boost is spent they'll be just as short on components as everyone else.

"Coming soon", I've been hearing this too often. But more than often, it will end up with disappointing availability.

Curious, barely any RDNA2 cards sold on week 10 at mindfactory, it used to be between 500-1k, closer to 1k, with RDNA2 cards as the most popular. Ampere dropped tiers.

Total Gaming GPU Sales Week 10
Nvidia Units 1760 = 72.42%
Radeon Units 670 = 27.58%

Radeon Top 5 Selling Brand Line

1. RX 580 = 520 Units.
2. RX 6900XT = 90 Units.
3. RX 6800 = 40 Units
4. RX 6800XT = 20 Units.
5. Nothing else was restocked this week.

Nvidia Top 5 Selling Brand Lines!

1. RTX 3060 12GB = 820 Units.
2. RTX 2060 Super = 200 Units
3. RTX 3070 8GB = 150 Units.
4. GT 710 = 150 Units.
5. RTX 2060 6GB = 130 Units.

fynxerAMD have known the situation of the market since Nvidia premiered the 3000 series in September and if they have been uncapable to increase production capacity in 6 months there is some thing wrong with their business model.

NV has entire Samsung 8nm to itself, yet you pick up AMD who shares TSMC with a bunch of guys AND OWN PRODUCTS (2 major consoles, CPUs) to bash.

How do you think "increase of production capacity" looks like, when you depend on third parties for silicon business?