New Polaris 30 Respin at 12nm to be released?

They really should not bring out a other refresh with a 5 to 10% increase in performance , what they need to do is bring out a GFX card that is as fast as a gtx1080 but under 300$(and that should be the price in stores) while not eating more power then a vega56 , that would sell like hoot cake.

 

The problem is that it's older than you and not good enough for 1440p.

 

You guys don't have to agree but I tell only from my experience with amd products. Someday my space heater will be faster then intel & nvidia. maybe then I can recommend buy it used amd instead of recommend buy intel & nvidia.

also I buy 2 used rx cards for 100 dollars both guys say they buy that for VR but it cant run it so they buy 1080ti. haters gonna hate. I test not much improvement over 1050ti and lost nvidia feature set. why waste mfg on this just make more vega 56. there is no magical source just other gamers. what is craigslist? apparently I can buy your credit cards too. git gud.

 

Roadmaps, market segments, profitability. That's what determines what product will come out or will be canned.
Does your 5~10% improved Polaris obstruct some incoming product on roadmap in target performance segment? No.
Will it be more profitable? Yes.

 

your market is based of a trend of miner? surely. but their rx on used market.
an it blocking vega 56 refresh midrange
will it be profit for amd? no everyone buys nvidia.

 

Why not. AMD have already stated there will be a Polaris refresh.

 

To be honest with you, my only complaint on AMD drivers is the shitty OpenGL for Windows, and the lack of setting vsync modes.

Everything else seems much better at the moment.

 

I don't think AMD have to abide by that contract any more, since GloFo has pulled out of the 7nm race.
I understood it as that they are free to do what they want now.

 

Is that a good cooler? No idea. Just because it cost a lot doesn't mean it's any good...

Don't know what you mean by this. I didn't pick any fight, this is just a fact.

No idea what you mean there. The drivers are almost always out for any major launch and even then it doesn't really make any meaningful difference to update in 99% of cases. For either nVidia or AMD cards.

I don't understand how you can't see that this is indeed a good move for AMD to make to keep its business fresh and money rolling in. The mid-tier market is the most valuable and this card will sit there just fine and probably outperform anything nVidia has in the same pricetier. Fanboying over who has the best performing card at the top is not important for this discussion.

 

They have to use GloFO because of this
https://www.guru3d.com/news-story/amd-changes-wafer-supply-agreement-with-globalfoundries.html

 

actually, you have this backwards.
the facts are that RTX and the 20xx line are rushed out just to steal headlines and be the shiny object for a few months.
I own Nvidia products, both gpus and stock. i love their solutions when they're not full of sh*t.
ray tracing has amazing potential and dlss even more so.
however, as i've said from the announcement, this is a rush job, completely out of character with Nvidia's pattern of development.
Nvidia pulled an Intel, just without the mini-fridge.
there is no stock to be had (in numbers relative to their last two releases), these are low yield gpu's which is directly reflected in pricing and it's all just to steal the thunder for c.e.s. and AMD.

AMD has economies of scale at every point of production, mainly because they created scalable designs at high yield, outperforming Nvidia and Intel.

you do not have to take my word for it...AMD was the #1 performing stock on the S&P 500 over the last year

 

The AMD drivers on Linux have both AMD's own drivers and the open source drivers which have been worked on and fine tuned over several years, lots of the compatibility workarounds, profiles and hacks in the Windows driver can be taken out and the OGL driver essentially rebuilt though it's a monumental task and even for AMD's own team the current OGL core took years from the info we do have available on the last rewrite they did for it.
(Started out really fast but got progressively slower with more compatibility hacks, workarounds, various other layers and code and that's probably been added to since, that was for DOOM 3 or at least the info available regarded Doom 3's performance and rewrite of the OGL core in the drivers and how it might have been faster and better to update the existing code instead.)

I do hope to see the improvements from the Linux driver and Windows driver eventually being merged and unified but that's also months of work or more for Vulkan and Windows.
OGL is also inherently single threaded I believe so AMD is following the specs and standards though both vendors have a ton of specific extensions (Same with Vulkan though it's only gotten started here as development is continuing on.) though from what I'm hearing AMD's Linux drivers or at least the open source ones are doing pretty well against NVIDIA now but both AMD and NVIDIA have also increased performance and compatibility even further in both OGL and VLK lately.
(Going by projects such as DXVK and AMD's Reddit section which frequently posts updates regarding Linux code commits they're both doing a lot of updates here particularly for Vulkan now but OGL isn't going anywhere anytime soon.)

Far as Cemu goes I'd imagine their work with a Vulkan API for 2019 is going to happen faster than AMD rewriting and updating the OGL driver components in the Windows driver and that will hopefully improve emulation GPU and CPU performance and allow for some of the existing speedup hacks to also apply to AMD GPU's.


EDIT: From my understanding of it at least, regardless of API the driver code is a huge piece of work with tons of profile specific settings, outright hacks sometimes even needed due to how broken the game itself is or software and then all sorts of other things plus all these extensions and more.
(Meaning any major overhauls or rewrites have to be done carefully and it'll also be a long term project well over a year or more for completion.)

4.6 still doesn't have full compliance in the Windows driver I believe though it's not missing much and there's not many games coming out using OGL 4.5 or newer either although for applications and work related tasks this can also be a bigger hindrance.
(And emulators as a very special piece of software pushing GPU and CPU pretty heavily particularly for newer systems.)

 

As far as I know, the far goalpost for GCN, (any revision), is 4096 stream processors, 256 TMUs, and 64 ROPs. That is as "fat" as it can be. This is Fury X, Vega 64 territory. I think we can agree that Polaris won't be buying a home anywhere near that neighborhood anytime soon.

My view is, if you shrink Polaris to 12nm, you can either invest that edge into clocks, or architecture. I don't think anything meaningful can be accomplished by trying to do both. To put it into gaming parlance, it's like you have just leveled-up. How do you want to use your newly-acquired skill points? Choose wisely.

If you choose higher clocks, with the same underlying architecture, I would imagine there are already examples of liquid-cooled Polaris 20 performance that would show us what could be achieved, and what might be plausible to expect at 12nm on air. 1500MHz? Higher?

Alternatively, if you choose to play-around with the stream processors, TMUs, etc., that immediately becomes more costly. And, AMD would need to tread carefully. You don't want to introduce something that in any way muddies-up your existing line of graphic cards, from a price/performance standpoint.

I agree that if AMD chooses to let the upcoming holiday season slip by without something to entice people to purchase their existing products, that isn't only lost revenue, but lost mindshare, as well. You can do the "free games" thing, and/or shave prices as much as possible. But, nothing gets attention, as well as sales, like a truly new product.

Simply bumping clocks upwards as a result of 12nm invites the "Ack! Rebadge!" reaction. Especially since this is how the transition from the RX400-to-RX500 series is already perceived.

If I were at the controls at RTG, (don't see that happening), I'd opt for tinkering around with the Polaris internals. I wouldn't want to completely re-engineer the wheel, so, I'd dive into the existing portfolio of GCN designs, and re-imagine one or two as Polaris. For the "lesser" product, I'd consider the basic design of the Hawaii Pro: 2560 stream processors, (40 CUs), 160 TMUs and 32 ROPs. For the upscale product, let's try something from the Firepro archives; the Hawaii XT GL: 2816 streamers, (44 CUs), 172 TMUs and 32 ROPs. What would this achieve, and how fast could it go? I have no idea. That is tonight's homework project for all of you.

(The true geeks reading this will point out that the Hawaii family has 64, not 32, ROPs. As far as I know, AMD has never produced a 256-bit memory bus graphics card with more than 32 ROPs. But, I'm not advocating the use of the older Hawaii template literally, but simply as a possible way to mitigate the start-from-scratch expense of a totally new design.)

Seems to me that while you're creating a new mask, you may as well bolt-on that GDDR5X or GDDR6 memory controller, too.

For marketing purposes, I'd insure a designation that indicates a truly new product, but minimizes throwing shade on products I'm intending to keep in production. Like, the "RX something-something Pro and XT" might work. I wouldn't want any more PCIe power cables hanging off my new product than existing RX580 cards, nor would I want any reasonable person to legitimately be able to whine about thermals and noise. (There will always be whiners, regardless.) And, if I was considering a reference model, as RTG Person-In-Charge, I would visit a level of rage of epic magnitude upon anyone who suggested we use yet another "blower" cooler. Clean out your desk and hit the 'pike.

 

Getting new Architecture is not about deciding how many TMU/ROP/SPs it will have. That's easy part driven by economics.
If AMD was to touch anything inside Polaris 30, it should be something that delivers clock to clock improvement, or area improvement.
Vega failed on both mentioned. But it seems that some things may have been finished for Vega 7nm and maybe Some things will be "reused" somehow in Navi.
So, if there is better and compatible functional block, it would be sensible to put it into Polaris 30. But Vega had nothing complete to give. So, I think Polaris 30 => Power efficiency used for higher clock.
And sure GDDR5X is good idea.

 

Yes, i understand that.
But GloFo pulling out of 7nm changes all that, as it is GloFo that is pulling out.
This contract becomes null and void if there is no future in GloFo high-end foundry business, which is what AMD wants.
AMD are now free to do what they want, without having to pay GloFo any money.
They can use TSMC, or anyone else, without repercussion.
thats how I understand it anyway.

 

My understanding was that they can use other fabs for 7nm, but still have to use Global Foundries for nodes that they do make. That is until there contract expires.

 

My understanding was that the ending of further nodes would be a breach of contract in some way, and that they were out, or had renegotiated it.

 

I think the specific wording is in the contract but also depends on what was the cause of the failure of GloFo to move to 7nm. If AMD and GloFo both failed to come up with an efficient 7nm process, then it cannot really be said to be breach of contract if they both failed to deliver on it if it is was a joint project. I am also interested to see whether this will impact the wafer payments AMD must make if buying from someone else - in this case TSMC, but most importantly, that share deal at $5.98 a share up until 2020. Unless there is a required hold time for their stock e.g. 10 years, that is an INSANE deal for them considering the current market price of AMD shares.