Rumour: Radeon R9 480/480X 3DMark 11 Scores

[Bacon1]

I'd assume the 480/X will launch at $200-220 and $250-280. It could be less but based on AMD's claim that they want VR capable cards under $300 they'll most likely stay closer to $300 for the top Polaris 10 card.

390/970 performance will be $200ish, Furyish performance would be $300ish.

There is no way to sell cards otherwise due to the older cards getting price cuts to that location anyway. 290 have sold for $200ish once 390 came out and these chips are cheaper for them to make. Same with fury, its a pricey chip. I wouldn't be surprised for the big Polaris to hit its speeds if they want to price something @ $300-350 range.

Top end card price is dropping regardless, they can't keep selling fury @ $550+ once 1080 launches, so those cards will either drop to $400ish or just be outright replaced with a Polaris card.

They are not only trying to bring VR to more people, but also have to compete against old cards that will be getting replaced.

 

[Sonikku]

God I wish they'd just ship already so we can stop the constant theorycrafting. I'm kicking around a 7770 still due to repeated advice to pass on a used 290x in favor of Polaris and I just want it here already.

 

[ShintaiDK]

Do you have a link? The most recent one I have seen (That was released to everybody) was from Q1, which was almost two months ago. The most recent call with investors was private, and no data from that was released officially. So maybe it really is 2017 still, just have not seen any updated docs on that.

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-irhome

 

[topmounter]

Why in the world do you think price / performance should be similar to the previous generation after a node shrink?!

Sorry, I don't understand your logic at all. I don't think 480x with all bells and whistles will be 200 USD, more like 275, but I expect 480 msrp to be 200-225, and I fully expect AMD to take meaningful share with these entrants.

I agree. I'd expect 480/x pricing to be consistent with that of the 380/x generation. I've seen nothing that makes me think P10 has any aspirations of being a 490/x tier card.

I do think that with nVidia's pricing of the 1070 has left the door open for the 480x to be a $279 card at launch if it can deliver Fury Air / 90% of the 1070's performance at FHD/QHD resolutions.

 

[Stuka87]

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-irhome

I am aware they have an investor relations page. However, there are no roadmap updates on that page since march. This was before we got all the rumors of Vega being pushed up to October, which AMD did not deny. And it should be noted that they did respond to this rumor by saying that Polaris was not being pushed back to October as well, and that they still fully intended to release Polaris on schedule.

 

[Face2Face]

God I wish they'd just ship already so we can stop the constant theorycrafting. I'm kicking around a 7770 still due to repeated advice to pass on a used 290x in favor of Polaris and I just want it here already.

At least you're patient. I just picked up a used cheap GTX 970 just so I can enjoy FO4 @ 1440p, the 7970 GE isn't cutting it. Anyway, I figured I might as well wait until July or so until everything settles down. Then I'll sell the 970 or put it in my kids build.

 

[krumme]

The 150 to 280 market is where the meat is. Next to laptops that is most important. p10 p11 better have far better access to that market whatever it takes. If amd can not capitalize on the console wins now in this market with the solid grib on dx12 perf they never will.

Be warned. If this gpu is highly efficient and performs like 390x and can do so in a mobile setup mobile will surely eat all it can and there will only be scrapes left for desktop and it will keep prices high. Ofcource. This is no 5850 situation guys. Desktop is last. First mobile soc then laptops then desktop (for new nodes).

 

[tajoh111]

If 14nm wafers cost $5K.

P10 yields [full die]

83/wafer @ 0.5 defects/cm^2 $60
101/wafer @ 0.4 defects/cm^2 $50
123/wafer @ 0.3 defects/cm^2 $41
152/wafer @ 0.2 defects/cm^2 $33

And you get all the cut die for free. If you assign a value to the cut, partially defective
die, this cost goes way down. The room certainly exists for $199 - $ 249 P10 models.

Wafers costs 4400 or 5k for TSMC and samsung after equipment expense. Actual selling price to companies is around 8k-9k. There are rumors of some discounts but nothing close to cost and these are mostly for contracts with big volume phone company soc's, and the rumors mention these companies as well.

Since GF is paying royalty fees and is a separate entity compared to samsung, it is not going to get the same discount. Hence costs are higher than 5k per wafer. Probably 8k with 7k as a best case scenario.

Too low a cost, kills AMD margins. Volume means nothing if margins are none existent.

Videocards with normal gddr memory have a cost somewhere between 50-60 dollars to make. So add this cost in and if AMD starts selling these cards at under 250, they start losing money elsewhere for their lower bins. Cut down cards are only about 10-20% cheaper to produce for the chip itself in the BOM. So if a cut down chip retails for 60-100 dollars less, then margin become console like very quickly without the volume of console chips.

 

[sirmo]

CPU/GPU packaging also isn't free. And that part of it has a yield (defects) as well.

 

[Det0x]

AMD just updated their investor presentation, it clearly shows Vega in 2017.

Do you have a link? The most recent one I have seen (That was released to everybody) was from Q1, which was almost two months ago. The most recent call with investors was private, and no data from that was released officially. So maybe it really is 2017 still, just have not seen any updated docs on that.

I'd like to see it too.

Me three.

Arachnotronic, don't just throw out unsubstantiated claims as others do. Show us your information.

Are we getting the link soon Arachnotronic ?

 

[krumme]

Wafers costs 4400 or 5k for TSMC and samsung after equipment expense. Actual selling price to companies is around 8k-9k. There are rumors of some discounts but nothing close to cost and these are mostly for contracts with big volume phone company soc's, and the rumors mention these companies as well.

Since GF is paying royalty fees and is a separate entity compared to samsung, it is not going to get the same discount. Hence costs are higher than 5k per wafer. Probably 8k with 7k as a best case scenario.

Too low a cost, kills AMD margins. Volume means nothing if margins are none existent.

Videocards with normal gddr memory have a cost somewhere between 50-60 dollars to make. So add this cost in and if AMD starts selling these cards at under 250, they start losing money elsewhere for their lower bins. Cut down cards are only about 10-20% cheaper to produce for the chip itself in the BOM. So if a cut down chip retails for 60-100 dollars less, then margin become console like very quickly without the volume of console chips.

That makes absolutely no sense. Gf cost might be high but not for the reasons given here. What is payed in licence is saved in dev cost. Ofcource. And more too as the dev cost is shared.
The entire idea of something getting more expensive because you buy it instead of making it yourself is very popular but frankly pretty stupid.

 

[Arachnotronic]

Are we getting the link soon Arachnotronic ?

Sorry my bad, go to AMD's investor page but it's the product roadmap slide deck. Slide #3.

http://phx.corporate-ir.net/Externa...WxkSUQ9LTF8VHlwZT0z&t=1&cb=635918349639602935

 

[Sweepr]

I am aware they have an investor relations page. However, there are no roadmap updates on that page since march. This was before we got all the rumors of Vega being pushed up to October

That rumour comes from a forum post, just like Polaris pushed back to October.

 

[krumme]

Sorry my bad, go to AMD's investor page but it's the product roadmap slide deck. Slide #3.

http://phx.corporate-ir.net/Externa...WxkSUQ9LTF8VHlwZT0z&t=1&cb=635918349639602935

Lol. It s bit of a stretch. Its one name per year on slide 3. Ofcource vega is pushed to 2017 here as all would have to be in 2016 and 2017 would be empty. Its not excactly techinical documentation we have here but marketing ppt and it shows.
What about referring to slide 5 where the granularity is like 5 words for gfx for 2016 including hbm 2.

 

[Sweepr]

Here's a closer comparison using the same CPU, similar memory:

Polaris 10 (67DF:C7): 18.060
http://www.3dmark.com/3dm11/11263084

R9 390X: 18.865
http://www.3dmark.com/3dm11/10039529

 

[Bacon1]

That rumour comes from a forum post, just like Polaris pushed back to October.

Rumours that you just love to post about

http://forums.anandtech.com/showthread.php?t=2472724

http://forums.anandtech.com/showthread.php?t=2472645


Threadcrapping and trolling are not allowed
Markfw900

 

[Sweepr]

Rumours that you just love to post about

http://forums.anandtech.com/showthread.php?t=2472724

http://forums.anandtech.com/showthread.php?t=2472645

Just because you select the ones you believe doesn't mean they're more than rumours.

According to forum chitter-chatter picked up by Germany's 3DCenter.org, AMD will bring forward its VEGA GPU architecture launch to October this year.

 

[Bacon1]

Just because you select the ones you believe doesn't mean they're more than rumours.

Are you talking about yourself? I try to avoid all this rumour BS but it seems to infect every single thread on the forums.

 

[nurturedhate]

Sorry my bad, go to AMD's investor page but it's the product roadmap slide deck. Slide #3.

http://phx.corporate-ir.net/Externa...WxkSUQ9LTF8VHlwZT0z&t=1&cb=635918349639602935

That slide say's it's from March 2016.

 

[dark zero]

Polaris 10 maxed out is a little better than a GTX 980/390X, slightly slower than Fury/Fury X.

Now we know why AMD is probably going to price Polaris 10 at $299-ish -- at $379, 1070 should be faster, and obviously 1080 goes unrivaled.

My prediction is that R9 390/390X/Fury/Fury X all get EOL'd in favor of Polaris 10. Much better cost structure (no HBM, smaller die, narrower bus width), as well as probably much better perf/watt.

Agreed with all the R 300 tier, but Fury, Fury X and Fury Nano are gonna receive a big discount, but still be alive until Vega comes.

Similar thing with nVIDIA. Once 1080 and 1070 comes, every Maxwell above 960Ti are going to being EOL'd and the rest are gonna get a discount.

 

[casiofx]

Kinda disappointed if the retail cards is not 1500Mhz+, it is 14nm already. Even 28nm Fury Nano have good performance with some low TDPs.

They should make the Polaris 10 480X with 150-180 watts to push the clocks higher. Lower clocked Polaris 10 can be made into 480.

 

[swilli89]

Are you talking about yourself? I try to avoid all this rumour BS but it seems to infect every single thread on the forums.

Eh yeah that Sweepr guy loves to post negative news about AMD, really makes the atmosphere around here kinda stink to be honest. Find myself spending more and more time at other forums because I come here and its just people like him spamming any little tidbit of negative rumors or gossip. It gets old :\


insulting other members is not allowed
Markfw900

 

[Arachnotronic]

That slide say's it's from March 2016.

Probably is, but it wasn't added to the deck until May.

 

[Magee_MC]

Sorry my bad, go to AMD's investor page but it's the product roadmap slide deck. Slide #3.

http://phx.corporate-ir.net/Externa...WxkSUQ9LTF8VHlwZT0z&t=1&cb=635918349639602935

Slide 5 in that deck shows "Computing and Graphics Roadmap: 2016"

Under High Performance GPUs it lists 2nd Gen High Bandwidth Memory.

Since Vega is the only chip so far that AMD has said will definitely get HBM2, it sounds to me that Vega could launch in 2016 and go through 2017 until Navi is released in 2018.

Either way it's less than definitive.

 

[Arachnotronic]

Slide 5 in that deck shows "Computing and Graphics Roadmap: 2016"

Under High Performance GPUs it lists 2nd Gen High Bandwidth Memory.

Since Vega is the only chip so far that AMD has said will definitely get HBM2, it sounds to me that Vega could launch in 2016 and go through 2017 until Navi is released in 2018.

Either way it's less than definitive.

That slide is from 2015