Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

AMD Confirms RX 480 At $199 USD, Other APU & Polaris Announcements

ManuelGomes said:
RX500 coming up soon?!
Click to expand...

Why shouldn't it been soon? It is just a speed bump and it has been almost a year.

8 pin connector finally on 580.
Too bad about the mem speed being the same, bump in core clocks though. 570 gets both.
No renaming to Polaris 20 after all? We'll see.
Click to expand...

If all "Polaris 20" is consists completely of narrow fab process optimization tweaks ( and perhaps better binning) to support a very incremental bump in clock speed of the same architecture ..... it doesn't deserve a new microarchitecutre name. In that context, "Polaris 20" is seems more so a 'cover story' as to why the product number is being bumped from 400 to 500. Like one of the comments said at the first line, this more appropriately would be labeled by adding a '5' at the end of the existing product 475 then replacing the '4xx'.

Do the marketing and sales folks really think they are fooling very many folks with this stunts? (I'm sure there bundled OEM parts contexts where this might work, but add-in-cards for sale in retail market??? ). Long term, this kind of smoke blowing doesn't really help AMD's reputation.
 
deconstruct60 said:
Do the marketing and sales folks really think they are fooling very many folks with this stunts? (I'm sure there bundled OEM parts contexts where this might work, but add-in-cards for sale in retail market??? ). Long term, this kind of smoke blowing doesn't really help AMD's reputation.
Click to expand...
Nvidia marketing is better than AMD, and they can get away with some "questionable" things from the past(GTX 970 3.5 GB Fiasco, for example). From AMD perspective, they have an excuse to charge the same amount of money for one year old tech. Secondly, they do not want to overshadow the Vega launch, which is real deal for them, this year. Its the same type of launch on GPU side as Ryzen is on CPU side, for them.
 
koyoot said:
Nvidia marketing is better than AMD, .....
Click to expand...

Nvidia has rebadged stuff also, but there real strength is in just have more products targeting different markets.


. From AMD perspective, they have an excuse to charge the same amount of money for one year old tech.
Click to expand...

I bet a substantive fraction of that 'extra' money the RX 500 brings in goes toward subsidizing discounts on the Rx 400 that they just turned into "last years tech". The bargain hunters will demand a discount and still get approximately the same tech. Unless AMD has very carefully managed the deployed inventory this probably doesn't generate much additional profit.

The folks primarily buying on 'tech' and not the name on the box will still look at this as "last years tech". The only folks drawing in are the ones looking the "400" and "500" on two boxes and figuring the "500" is worth paying a little more.

That "last years tech is worth less money" is actually a learned behavior that is reinforced by doing gimmicks like this. Gimmicks like this make that erosion worse , not better, over the long term.


Secondly, they do not want to overshadow the Vega launch, which is real deal for them, this year. Its the same type of launch on GPU side as Ryzen is on CPU side, for them.
Click to expand...

Vega is not going to fill this role any time soon. So far the Vega has same "RX" major product name label.

https://www.techpowerup.com/231102/amd-names-radeon-vega-product-line-as-simply-radeon-rx-vega

Churning the 400 to 500 isn't going to block much of anything at all for "RX Vega xxxxx" or simply "Vega" (no numbers because it is only one entry in a product line up.). Highly doubtful there is any overlap at all between Vega and the RX 500 product line. So how could it possibly overshadow???

Again this just erodes the who notion that there is some "top to bottom" refresh coming any time soon. They are bumping to RX 500 probably because it will be an intermediate-long amount of time until there is a RX 600.

Vega to fill the top end . Polaris 10 to fill the mid range desktop and perhaps a bump at the lower end to fill lw end desktop end the most of the mobile space. Add-in-card market. [ There may be some very chopped down version of NCU (iVega') in an CPU-APU offering later but wouldn't be anywhere in the same zone performance wise. the memory subsystems are likely different. ]
 
deconstruct60 said:
Vega to fill the top end . Polaris 10 to fill the mid range desktop and perhaps a bump at the lower end to fill lw end desktop end the most of the mobile space. Add-in-card market. [ There may be some very chopped down version of NCU (iVega') in an CPU-APU offering later but wouldn't be anywhere in the same zone performance wise. the memory subsystems are likely different. ]
Click to expand...
Lisa Su, actually in investor conference few months ago have said that AMD will have "sort of top to bottom launch of Vega architecture GPUs". ;)

Well what this actually means is that there are two Vega GPUs, big one, and smaller one. Big one for 500-700$ market, and smaller one for 350-500$ market.

And there are also APUs with Vega architecture, so this is what actually she meant by "sort of..." ;)
 
koyoot said:
Lisa Su, actually in investor conference few months ago have said that AMD will have "sort of top to bottom launch of Vega architecture GPUs". ;)
Click to expand...

Have a direct quote? There is a difference between top-to-bottom 'new/refreshed' GPU product line up and one that is 100% based on Vega. The latter there is nothing that suggest that I've found. top end? Yes.

"...
Meanwhile, Su said that the company is also on track to ship Vega graphics chips in the second quarter. Those chips could help the company gain high-end graphics and gaming computer customers. Vega will also be targeted at machine intelligence applications, a mainstay for both Intel and Nvidia.

“We are returning to the high end of the market, where we have not been in years,” Su said. ... "
https://venturebeat.com/2017/01/31/...-products-are-on-track-for-big-2017-launches/


CPU processor line up ...
"... As we introduce the next families you will see positioning but the end result is that you will see a top to bottom stack with a processor for everyone. At every price we will offer more performance. You will be able to see that in your own testing. .. ."
http://www.anandtech.com/show/11177/making-amd-tick-a-very-zen-interview-with-dr-lisa-su-ceo

That's "... as we introduce next families... ". That is not this year with one single micro-architectural design.

Well what this actually means is that there are two Vega GPUs, big one, and smaller one. Big one for 500-700$ market, and smaller one for 350-500$ market.
Click to expand...

That still leaves the sub $350 market which is not top-to-bottom. That is actually primarily just mostly the top. Which is exactly what the direct quote addresses above. The RX 480 launched in the $199-240 zone.... it never was anywhere near $350. And neither is there Vega product.

http://www.anandtech.com/show/10446/the-amd-radeon-rx-480-preview

So how can possibly overshadow a product $100 cheaper than ( that is -31% price difference than: $350 -> $240. Going other way is bigger percentage blocker: +45% )? It can't do product differentiation with a 45% price difference (and performance diff ) then should just quite the marketing business altogether because without a clue.
 
koyoot said:
Lisa Su, actually in investor conference few months ago have said that AMD will have "sort of top to bottom launch of Vega architecture GPUs". ;)

Well what this actually means is that there are two Vega GPUs, big one, and smaller one. Big one for 500-700$ market, and smaller one for 350-500$ market.

And there are also APUs with Vega architecture, so this is what actually she meant by "sort of..." ;)
Click to expand...

Given that we haven't heard any details/leaks on this second Vega chip, I'm starting to think its simply the Raven Ridge APU/GPU. A "sort of top to bottom launch of Vega" can mean a lot of things. For instance, it could mean its a top(Vega 10) and bottom(Vega APU) launch.
 
If I remember correctly, RX4x5 was indeed reserved for a refresh.
Right, here it is:
upload_2017-4-2_22-22-45.jpg


We never saw the 490, or any of the 4x5.
 
Stacc said:
Given that we haven't heard any details/leaks on this second Vega chip,
Click to expand...

the $350-400 vs. $500-700 could also be there is no separate die. That range could be solved with a smaller set of working core groups coupled with smaller HBM2 memory allocated. As an illustrate example a die with nomial 4096 compute cores. "Big' with 4096 working and stacks that add up to 8-16GB of HBMv2 'memory cache' and 'small" with 3072 working and HBMv2 stack(s) that add up to 4GB. That would more so be playing with configuration ( working sections of die and memory stacks added to the interposer. ) than a new die layout.

Depending upon yields a decent chunk of the 3072 working units would be culled from the defects would have had to throw away anyway. ... so lower price. Less HBMv2 memory so again lower price.

I'm starting to think its simply the Raven Ridge APU/GPU. A "sort of top to bottom launch of Vega" can mean a lot of things. For instance, it could mean its a top(Vega 10) and bottom(Vega APU) launch.
Click to expand...

but is Vega APU ( little to no HBMv2 and DDR4 for bulk of RAM access ) really going to be "Vega" or just the NGU core subset?

"... Meanwhile it’s very interesting to note that with Vega, AMD is calling their on-package HBM stacks “high-bandwidth cache” rather than “VRAM” or similar terms as was the case with Fiji products. ... "
http://www.anandtech.com/show/11002/the-amd-vega-gpu-architecture-teaser/3

If trying to fix 2-4 Zen cores and a chopped down set of NGU GPU cores onto the same die then the 'cache' and some of the fancy interposer stuff may go. That would be a very different performance curve. Even if just leave something like a 1GB cache around. the "vega" high end card is going to set a perception that this iGPU isn't going to match.

That is more so AMD using "vega" interchangeably with "NGU core". If AMD does that I think they are screwing up the marketing message; it won't play well. If going to put an explicit "Vega" tag on the high end card then that branding isn't really going to scale well.
 
deconstruct60 said:
but is Vega APU ( little to no HBMv2 and DDR4 for bulk of RAM access ) really going to be "Vega" or just the NGU core subset?
Click to expand...

It is GFX IPv9, with every single one improvement, and hardware feature, including High Bandwidth Cache Controller. It is actually essential for Vega GPUs.
 
koyoot said:
None. Its Apple's own custom chip. But they can protect themselves from any lawsuit, by licensing AMD Intellectual Property, which will be cheaper than licensing the GPU tech from Imagination.
Click to expand...
Basically that's what I meant. Apple needs some foundation IP to start with. MY guess is it is AMD's.
 
Last edited:
  • Like
Reactions: Synchro3
ManuelGomes said:
....
Intel is prepping KBL-G with integrated discreet GPU with possibly HBM2. I wonder if this is AMD based? Guess not (yet), it's GT2.
Click to expand...

'Integrated discreet' is a bit of an oxymoron. Those are actually opposites adjectives.

What wouldn't be surprising is if Intel is switching from eDRAM to HBMv2 RAM that the GPU switch over to the same die as the HBMv2. Intel's packages with eDRAM now have one CPU-GPU die and one eDRAM die inside them. This would just shift the GPU to the other die. The interposer only need span the GPU and the HBMv2 memory.

That would help with moving to 10nm because minus the GPU ( or respectively CPU) they could produce smaller dies ( with higher yield at new levels. ). So just package together more dies more often across the product line. The CPU is loosing access to the eDRAM cache (which was more shared ) but with the GPU transistor count decrease Intel could bump the L2 ( or add L3 cache ) for the CPU/GPU to continue to share. (it would be smaller than the eDRAM cache, but don't have to take total hit if GPU RAM cache goes somewhat high latency and to a small extent 'private' because a bit less inclined to use from CPU side. )

Little of that really "needs" AMD GPU tech. Intel already had mechanisms for shared use of the eDRAM. It may be faster to leverage some of the HBM as a cache stuff if pretty thoroughly patented.
 
The CPUs in question have 65 and 100W TDP. Current mobile/BGA Intel chips have 45W TDP. So that kinda shows you what we have to expect. Intel Mobile CPU cores are very efficient. GPU subsystem alone appears to be adding additional 50W of power, to the package.

Interesting prospect.
 
Zarniwoop said:
https://techcrunch.com/2017/04/03/apple-gpu-tech/

How much AMD IP will be in the Apple GPU?
Click to expand...

This link was posted in article that is one the front page

http://www.realworldtech.com/apple-custom-gpu/

It isn't like Apple hasn't pragmatically already started on this . From David Kanter analysis last Fall a big chunk o the internals of the current A-series GPU is custom Apple implementation already. Apple has already been doing the "computational" part and evidently their own Metal drivers.

It does beg the question of just how 'clean' this is of ImaginTech's patents since the "front end" (fixed logic ) is compatible. The assembly commands of the shaders are different, but still coupled to same fixed function stuff.


This probably has close to zero impact on the Mac line-up though. There isn't enough volume in the Mac line-up (let alone some smaller subsection of it) to do a custom, Mac only design. It is illustrative though in that if folks are a slacking in Apple's eyes in getting requirement demands done, they they will kick a contractor to the curb. AMD needs to get their stuff together. [ Nvidia already appears to have been benched too.
Nvidia: OpenCL ... pfft we're going to backburner that....
Apple: OK bye. ]
 
  • Like
Reactions: koyoot
koyoot said:
324afb9bf3ab4f11bb19b5409e43cbd2bf494db4904f2dde2d80efd4c59bd442.jpg

1.5 GHz RX 580. That is impressive.
Click to expand...
Where can I buy these RX580 cards with Apple's proprietary EFI firmware?
[doublepost=1491683158][/doublepost]
ManuelGomes said:
Titan Xp?! Will it work only on Windows XP? Kidding :)
Support for Mac coming it seems:
https://blogs.nvidia.com/blog/2017/04/06/titan-xp/
Full fat GP102, nice!!
Click to expand...
At least they realized that having two very different cards with the same name wasn't a good idea.

Maybe they should rename the "Titan X (Maxwell)" to "Titan Xm" and bundle a satellite radio subscription.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back