• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

New AMD Polaris based GPU

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
jpiniero said:
If anything, AMD is small fish when it comes to TSMC's 7 nm production. The issue isn't capacity per se, it's more that the wafer costs are insane at 7 nm, and early on it's even worse. That's why you are likely going to only see very high margin products from AMD (and nVidia) until the middle of the year at the earliest... and that includes Ryzen.
Click to expand...

But are 7nm prices really that high? For one thing- Apple's chip is ~83mm2, and is estimated to cost $72 to produce. Ryzen chip would be bigger, but also, Apple probably does not do much binning.
Another point of reference- AMD were estimating 2x cost to produce a 200mm2 chip on 7nm, compared to 14nm. Considering how much smaller 7nm Ryzen would be, and better yields on a smaller chip- that also suggests reasonable pricing.
 
neblogai said:
But are 7nm prices really that high? For one thing- Apple's chip is ~83mm2, and is estimated to cost $72 to produce. Ryzen chip would be bigger, but also, Apple probably does not do much binning.
Click to expand...

Apple doesn't do much binning but you have to think they are getting a substantially better deal than AMD is though... if only because of the volume. I would think they would want to sell even the busted dies as Epyc early on.

I imagine Vega 20 is going first because the margins on that are even better than Epyc.
 
jpiniero said:
Apple doesn't do much binning but you have to think they are getting a substantially better deal than AMD is though... if only because of the volume. I would think they would want to sell even the busted dies as Epyc early on.

I imagine Vega 20 is going first because the margins on that are even better than Epyc.
Click to expand...
OK, I'm confused. We have the claim made often that TSMC will not have enough capacity for AMD to fab all their die there and now Apple can get a better pricing deal because they have the volume whereas AMD does not.

Seems contradictory.
 
There isn't much volume left for everyone who isn't Apple because Apple - who have much more money than anyone else right now! - have first call on the capacity and need a lot of it.
 
Qwertilot said:
There isn't much volume left for everyone who isn't Apple because Apple - who have much more money than anyone else right now! - have first call on the capacity and need a lot of it.
Click to expand...

Apple stockpiles chips for a launch for months. It's not like a graphics card launch where in most cases they are shipping once the cards are assembled from the first batch released from the fab.

Besides, the A12 isn't the only TSMC 7 nm phone chip being released in 2018 - looks like the Mate 20's launch is on Tuesday.
 
Yeah pretty sure that rx590 card is meant to be released for the Chinese market and is not gonna be the refresh that is coming or at least I hope this is the case.I thought the rx590 is a rebrand of the rx580 with higher clock rate and the refresh will be branded as rx6xx and will be refined 12nm. where as the rx590 is still 14nm to sell off the remaining supply and make room for the up coming GPU's for this next year.
 
Last edited:
GodisanAtheist said:
-After the sometimes absurd levels of kowtowing to NV from reviewers during this latest launch, is such a thing really so outlandish?

There is a long and storied history of driver level "optimizations" in the video card space. If NV can enable variable rendering as the default in drivers they'd likely get away with it for a while since image quality comparisons aren't really a thing in modern reviews. Everyone just assumes the final rendered product is exactly the same between all vendors, and that no one is taking any shortcuts.
Click to expand...

kinda wish there was a reference software renderer available just to see what stuff is *supposed* to look like vs. how the graphics cards are rendering.
 
ElFenix said:
kinda wish there was a reference software renderer available just to see what stuff is *supposed* to look like vs. how the graphics cards are rendering.
Click to expand...
This is something I've often wondered about. Are images rendered by both GPU designers equivalent? I for one have noticed differences in effects depending on which card I used. Who's right? Is any? Certainly in the past, there have been discrepancies resulting in performance differences.
 
Charlie22911 said:
These guys need to stop! A cut down RX580 is not an RX580. This isn’t as bad as NVIDIA with their outright deceptive low end parts, but it is leading down the same path.
Click to expand...
Agreed, the Nvidia camp isn't any better with there crazy prices and its not gonna get any better any time soon.
The upper mid-range card I was gonna get the RTX 2070 is now expected to be released with a $599-$699 price point when the msrp was supposed to be like $499..
I think if AMD can do a refresh and not a rebrand of the 580 and match gtx 1070 performance then that will be good for everybody.
 
Justinbaileyman said:
I think if AMD can do a refresh and not a rebrand of the 580 and match gtx 1070 performance then that will be good for everybody.
Click to expand...

A 10% increase they are saying will not catch up to the GTX 1070. I said the 12nm parts won't do much, and they'll keep using 8GT GDDR5. GTX 1070 will also overclock better as AMD was never good on clocking. It's a refresh with higher clocks. The might need to jack up power usage again too, just like the RX 570/580.
 
cortexa99 said:
https://translate.google.com/translate?hl=en&sl=auto&tl=en&u=https://www.chiphell.com/thread-1918935-1-1.html

Polaris 30 will use TSMC 12nm, would be released next month 1H
Click to expand...

That would be quite strange. The way I understand it, it would cost a lot to port Polaris to TSMC.So if investing large sums- it would not make much sense to release Polaris, but an upgraded architecture instead. Also, add WSA fees. In contrast, 12nm at GloFo should be painless, cost very little, and should be a natural choice for a refresh of an old architecture.
 
Last edited:
So if I’m reading the rumors correctly, it looks like we have a RX 580 that’s being built on a 12nm process (GF or TSMC still unknown for sure). It appears (but is unconfirmed) that the memory clocks remains unchanged. The name is (probably) the RX 590. If true I would expect it to be released in the next month or two so as to be available for the Christmas buying season.

The name would make sense. I get the feeling that they are only coming out with this one card so it would be lame to call it the RX 680 as that would imply there should be a 670, 660, etc.. Of course AMD can be pretty damn lame (RX 570 rebadged as a RX 580 2080SP) so anything is possible.

Honestly I believe these rumors because AMD needs to do something. Nvidia just released their high end products and AMD won’t have anything to respond with until sometime next year. Nvidia will soon release their midrange cards kicking AMD further to the curb. AMD needs something to keep them in the public eye. They pulled that RX 560X BS (regular RX 560 just with newest drivers) just to give the OEM’s something new to put on their spec sheet. Releasing an updated RX 590 card would actually give them something real. It would keep people talking about their GPU’s, even if it’s just to say how lame it is. Sometimes bad news is still good news as it reminds people you still exist.

Concerning the fab, I doubt the TSMC rumor. Since GF 12nm is most likely an improvement on the current 14/16nm process, the work to migrate it would not be unreasonably large. The amount of work required to move to a new TSMC process would be much larger. Having said that however, AMD is going to have to do it anyway. They are moving everything to TSMC anyway and migrating RX 580 to TSMC may be useful since they have to learn new processes anyway. I’m no expert on processes so I’m just guessing here.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top