The AMD Execution Thread [2007 - 2017]

Status
Not open for further replies.
That might help explain why Polaris achieves the same transistors/mm2 gain from the supposedly more dense GF 14nm process as Pascal gets from the TSMC 16nm process.
I don't think it's more dense, not really, they are roughly the same, it's just different nomenclature between different foundries.

image.jpg
 
New Actually it is denser
That is not proof. That is an example of a specific application. It may be that Apple needed to target specific power consumption or performance characteristics, and die size is a derivative of those. For different requirements, it may be possible that the density advantage would swap around. Of course, since I'm an economics teacher, I may be completely wrong :)
 
That is not proof. That is an example of a specific application. It may be that Apple needed to target specific power consumption or performance characteristics, and die size is a derivative of those. For different requirements, it may be possible that the density advantage would swap around. Of course, since I'm an economics teacher, I may be completely wrong :)

Well, A9 is as close as we can get to compare both processes

Anyway, this is more a philosophical comparison, AMD can source at both foundries TSMC and GlobalFoundries (with samsung tech at 14nm). As you said between AMD and Nvidia are more differences than just the process
 
That's interesting, but as I understand, it doesn't change anything about the Wafer Supply Agreement, does it? Let's hope AMD can shove a ton of semi-custom chips at GloFo, so they have the ability to look elsewhere for products aimed at more competitive markets.
 
That's interesting, but as I understand, it doesn't change anything about the Wafer Supply Agreement, does it? Let's hope AMD can shove a ton of semi-custom chips at GloFo, so they have the ability to look elsewhere for products aimed at more competitive markets.
Just need to supply a chip for a new portable Nintendo for pokemon catching and they're golden.

In other news, I've seen some investor speculation that AMDs stock price increases may be related to a potential acquisition.
 
AMD confirmed today that they have finished testing and are now prepared for production use of Samsung's 14nm fabs. That doesn't mean they're actually using Samsung yet, but they have the option ready to go. GloFo's 14nm process is a licensed copy of Samsung's, so there should be no design changes needed. Even if they don't use Samsung, the mere option of an alternative fab will give AMD significant bargaining power with GloFlo on future pricing and wafer allocation.

Probably coincidentally, AMD's stock rose 15% more today, up 30% since Thursday's quarterly financials release.

Ha, I'm late to the party, this news was already shared in the forum today.
 
Last edited:
The Samsung/GF arrangement was termed copy-smart, which apparently brings the processes into closer alignment without reaching the level of synchronicity that can put a single chip design through more than one fab like Intel's copy-exact.

It seems like it wouldn't be as drastic a shift as it was when AMD made some lateral moves between TSMC and GF at 28nm, but it sounds like AMD's attention will still be split between TSMC and GF regardless of the additional burden of characterizing with Samsung.
The quote in the article indicates that AMD would be leveraging the Samsung/GF agreement to provide 14nm capacity. The charitable interpretation that AMD may very well want is that means AMD can leverage it when GF is a disappointment. Alternately, the capacity agreement might only play a role if it AMD's volumes exceed GF's capacity or the limits of the WSA, even should the silicon be a little underwhelming.
 
AMD has also changed the criteria it uses for IP numbering and moved some items that used to be externally visible out of the driver-level tables.

This doesn't sound quite right. What I said previously was that historically any non-trivial change in IP implementation also required non-trivial* changes in the driver support, so the issue of whether the IP block version tracked implementation or driver programming was moot. Polaris was the first time where we had non-trivial but largely back-compatible implementation changes and so the distinction became important for the first time as well.

* non-trivial in this context defined as "needing a separate copy of the IP block handler in the driver rather than being easily handled by runtime switches in the existing code"

This means AMD has gone from some externally unknown methodology to another externally unknown one.

I would say we are consistently applying the same externally unknown methodology in both cases, just dealing with some new use cases.
 
Hmm why would they use the Polaris design then? If Vega is much better at what ever its going to do an architecture level for power and performance, there would be no need to Polaris's design to be shifted to 14nm.
Perhaps Polaris was needed for AMD's console 'refresh' design wins. Vega may include significant architectural changes which make perfect emulation of GCN2 difficult. This is pure speculation however.
 
Bloomberg reports on AMD Polaris 11 getting the higher-end Macbook Pro design win.

http://www.bloomberg.com/news/artic...-plan-first-pro-laptop-overhaul-in-four-years

Apple is using one of AMD’s "Polaris" graphics chips because the design offers the power efficiency and thinness necessary to fit inside the slimmer Apple notebook, the person said.
I wouldn't be surprised if the lower z-height they claimed for P11 was actually something that apple asked for.

[sarcasm]That terrible design win that is sure to be so bad for AMD because apple is definitely giving them only $2.5 per GPU so AMD is losing money on this.[/sarcasm]
 
I would not expect wonders from Vega, AMD needed Polaris to make the smaller Chips competitive again, but Vega simply hangs on the availability of HBM2. Sure they might have improved the architecture a lot.
 
Ok so it finally happened: PCI-SIG withdraws PCIe certification for reference design Radeon RX480 cards.

...Or so Nordic Hardware claims anyway. Not sure from where they got their info, there's no source reference in their news post.
 
lol it makes you wonder how it was certified in the first place if they're now withdrawing that certification (and it's true).
 
Ok so it finally happened: PCI-SIG withdraws PCIe certification for reference design Radeon RX480 cards.

...Or so Nordic Hardware claims anyway. Not sure from where they got their info, there's no source reference in their news post.
I had heard someone mention the same thing. But I've never seen it on the list to begin with (admittedly, it's not as if I regularly check).
 
Status
Not open for further replies.
Back
Top