• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Polaris architecture – GCN 4.0

CAT-THE-FIFTH said:
The 32NM and 28NM plants GF use are located in Dresden and New York:

https://en.m.wikipedia.org/wiki/GlobalFoundries

This covers not only AMD CPUs but also the SOCs in the PS4 and XBoxOne too. All are assembled in Asia.

AMD produced most of its CPUs at the fab in Dresden and they have WSA considerations too and Samsung had a head start over TSMC and has capacity which is not contested by a billion other companies too. WSA is really an important consideration here. It's one of the main reasons AMD has lost so much money through penalty clauses and inventory writedowns(since they need to produce a fixed number of chips based on how much capacity GF expects them to buy).

You also need to consider that Samsung fab is probably producing chips for iPhones which are being assembled in China.

Also using a low power process for mainstream chips will be advantageous due to the target market and also since it will be using GDDR5,so they cannot rely on power saving measures with HBM2.

You need to consider shipping costs are a tiny percentage of overall costs. FFS, a while ago I saw some Young's Frozen Cod which was caught in the Atlantic, packed in China and then shipped back here!

But we are probably going to disagree on this so best keep it at that.
Click to expand...

Per my prior post, that's fine for much higher volume CPUs but not for GPUs, especially as the latter are frequently far bigger and more costly (and therefore much more expensive to insure against loss or damage). Smaller batches and higher costs and margins, and higher fluctuations in price and demand = they don't want a complex supply chain for GPUs.

Maybe they could do it with some small super high volume OEM parts. Not the kind of packaged product that you or I are likely to buy though.
 
WCCFTech are now reporting availability (probably for ulp OEM parts) as happening in 2 months, so perhaps that does explain GF.
 
You would ship an entire wafer of chips, a wafer is a fixed size and a fixed cost to process a new one(or thousands of them as would be being shipped) so there is absolutely no cost difference in the slightest. If a wafer costs $4k and you ship 10k wafers worth of parts to a packaging plant, it doesn't matter how many chips or what the final sale price would have been but the cost of the wafers, which is identical.

The supply chain would be identical in either case, if you can't have a complex supply chain for gpus you couldn't for CPUs either, you either know exactly how to deal with it or don't. You either ship millions of parts a year or you don't. Once you ship millions, adding extra millions is effectively easy, the whole supply chain is already in place.

in reality, shipping would be incredibly cheap and swift via air travel. A full boxed GPU becomes fairly expensive to ship in mass on planes, but the silicon chips in trays, you can fit probably 100mil + worth of chips into a single cargo plane and the cost of shipping that number would be pretty damn insignificant. The time for shipping would be reduced from 4-8 weeks to a couple of days max. Shipping silicon chips isn't a problem. Intel wouldn't have plants all over the world, nor would AMD have gotten permission to build a fab in the states, nor would ATIC(forget the name actually, might be something else) have bought out the fab side of AMD and created glofo if building chips in the USA was such a huge disadvantage. Neither would IBM have made chips there.
 
Last edited:
pmc25 said:
WCCFTech are now reporting availability (probably for ulp OEM parts) as happening in 2 months, so perhaps that does explain GF.
Click to expand...

I was reading through the preview of GNC 4 on Anandtech and the part were they mentioned the split between TSMC and GF was certainly an eyebrow raiser.

I wonder if there doing this to manage capacity demands on the new smaller process or if AMD are being forced to use Global Foundries due to contractual obligations to buy a certain amount of wafers from them. Either way it's certainly a very unorthodox move by AMD.
 
drunkenmaster said:
You would ship an entire wafer of chips, a wafer is a fixed size and a fixed cost to process a new one(or thousands of them as would be being shipped) so there is absolutely no cost difference in the slightest. If a wafer costs $4k and you ship 10k wafers worth of parts to a packaging plant, it doesn't matter how many chips or what the final sale price would have been but the cost of the wafers, which is identical.

The supply chain would be identical in either case, if you can't have a complex supply chain for gpus you couldn't for CPUs either, you either know exactly how to deal with it or don't. You either ship millions of parts a year or you don't. Once you ship millions, adding extra millions is effectively easy, the whole supply chain is already in place.

in reality, shipping would be incredibly cheap and swift via air travel. A full boxed GPU becomes fairly expensive to ship in mass on planes, but the silicon chips in trays, you can fit probably 100mil + worth of chips into a single cargo plane and the cost of shipping that number would be pretty damn insignificant. The time for shipping would be reduced from 4-8 weeks to a couple of days max. Shipping silicon chips isn't a problem. Intel wouldn't have plants all over the world, nor would AMD have gotten permission to build a fab in the states, nor would ATIC(forget the name actually, might be something else) have bought out the fab side of AMD and created glofo if building chips in the USA was such a huge disadvantage. Neither would IBM have made chips there.
Click to expand...

That's a good theory, except for one thing. They never ship wafers, only the chips once broken up.
 
Kaapstad said:
17 billion on full fat Pascal

Your turn, how many on full fat Polaris ?
Click to expand...

Can you find Nvidia saying that anywhere? I didn't watch the recent press thing because, auto drive cars = boring. 17mil is a guesstimate that has been around for ages based off taking the current big Maxwell and taking that times the transistor density improvement of 16nm. That number has been around for ages and maybe it's been confirmed but was just a guess initially. The same way people state that big Polaris is between 15-18bil transistors. Nvidia tend to stick with 500-550mm^2 for their big GPU, AMD have only done a big gpu once since the 2900xt and largely only because 28nm was around so long.

With 10nm closer it's more likely for AMD to go back to a 370-450mm^2 core for 14/16nm and we'll be at 10nm before you could bring out a new gen at 14/16nm.
 
You must log in or register to reply here.
Back
Top Bottom