Thinking that Navi will be the first real "Raja" card we see from AMD

Zion Halcyon

2[H]4U
Joined
Dec 28, 2007
Messages
2,108
Reposting this (I was the original author) from reddit. It was a thought that occurred to me as I stupidly took a thread here off topic, and then used the ensuing vacation Kyle was nice enough to give me to do some digging on AMD, Raja, Arctic Islands conversion to Polaris and Vega, and put it all together - I apologize for the obscene length.


--------------------------------------------------------------------------------------------------------------------------------------------------------


Forgive the editorial format of this as I just want to get these thoughts out in light of Polaris launching, the competition with NVidia, the reviews coming in, and frankly, my thoughts on where AMD is going from here.

I'll do you all a favor and get the TL|DR portion out of the way first:

I think we will not see real, head to head competition with Nvidia until Navi, because I believe that is going to be the first real graphics chip that is 100% innovated by Raja Koduri. And AMD is holding down the fort with Polaris and Vega as best they can until then.

Now, on to the meat:

Since before Polaris launched, there has been a significant debate of what is going on behind the scenes. HardOCP, among other things, did correctly predict in an article that "AMD has a problem on its hands, as both these products have come up significantly short of where these were supposed to land. But that is OK for AMD, it will simply send Chris Hook out to fall on the sword and tell a story of that was the plan all along....to produce brand new parts much slower than its last high end GPUs. In the simplest terms AMD has created a product that runs hotter and slower than its competition's new architecture by a potentially significant margin. " (http://www.hardocp.com/article/2016/05/27/from_ati_to_amd_back_journey_in_futility/#.V7KGD4WcFaQ)

Editorializing aside (and ignoring other claims in the article that aren't pertinent to this thread), Polaris runs about the same as an AMD 390, and while it is actually a VERY nice card for the price that has HUGE gains in Vulkan, it still isn't very competitive compared to NVidia, actually borderline fails to live up to the VR Ready claims marketing put forward in terms of quality based on tests with actual games and simulations, and generally has been a nice card for the price, if underwhelming.

It would be easy for some to write off AMD for that, but there is a historical reason for this, and for that, we need to go back to pre Raja, and realize Polaris and the upcoming Vega are actually rebranded Arctic Islands chips already in the pipeline when Raja joined in 2013.

Prior, under Hector Ruiz's mishandling of AMD, he shut down both processor and GPU R&D which led to a severe lack of innovation and a string of AMD failures that are not easily recovered from, and given the turnaround time for a new GPU to be designed to sold, even the upcoming Vega was likely partially already in the pipeline when Raja came on board, meaning when he and Lisa Su started up their R&D when they took over the respective parts of the company, there likely was a limited amount they could do.

Staying on point with the video cards, Greenland (now Vega) will likely be the last "Hector Ruiz" era card and will have some engineering tweaks done, as this was always intended to be the "flagship" of this generation of cards (albeit underperforming ones compared to their NVidia brethren).

Hence, there is a reason they priced these cards to sell and marketed them how they did (genius really) - they need to make their way through and keep capital coming in.

And really, for the price, while they may not be world beaters, they are great little card for the current PC standard - 60FPS at 1080p with all the bells and whistles.

I guess my overall point is this - don't be surprised if Vega disappoints compared to NVidia's offerings - in fact I am almost counting on it. But also understand one doesn't dig out of a Hector Ruiz-sized hole overnight, or even in 2-3 years.

AMD has better leadership, and even if they have to put lipstick on a pig for a couple of generations, they are SMART, and pricing aggressively to do it so that the baseline budget gamer should more than feel they are getting their money's worth.

It's not because of the cards, but because of the leadership that AMD is rallying in the stock market, and even if they have to work with some hindering engineering from the lean years of AMD, they are selling it and working it out.

Which means, tune out all the people who say "AMD can't afford to keep playing cheap or they will go under." - They are grabbing market share and know what they are doing, even if they have one arm and both legs tied behind their back at the moment to do it.

But if you are looking for a chip to finally trade blows with NVidia (and let's not kid ourselves - NVidia will have Volta at the ready when Vega is in full swing), don't look to Greenland/Vega, barring some real technical wizardry or frankly, AMD hitting the engineering lotto given the era it came from.

Nope, keep your eyes peeled for Navi - which was not a part of the old "Arctic Islands" architecture, and will be the very first card Raja himself is responsible for from start to finish.

Maybe Zen turns it around on the processor side, but when it comes to graphics - all eyes on Navi, and it might not be fair to judge them a minute sooner before that.
 
One footnote - I know my post paints Raja as a savior of sorts, and it was not my intent to say I totally believe he can. But I do think he has a legit, pre-built excuse if Vega tanks. However, once Navi is released into the wilds, he has no more excuses, and not all the PR in the world will be able to save him.
 
Well Navi is a completely different approach, small GPU's working together, we have seen this strategy before and it did work with AMD, when their GPU was performance comparative to nV's,

That is the problem, they still need performance to be comparative. Doesn't matter if its one GPU, two or more gpu's acting as one gpu, so the small die approach works better with new nodes, where yields and volume production will be an issue, as those two problems alleviate, the need for smaller dies gets less too as the cost and time to market will end up equalizing to a certain degree.
 
Well Navi is a completely different approach, small GPU's working together, we have seen this strategy before and it did work with AMD, when their GPU was performance comparative to nV's,

That is the problem, they still need performance to be comparative. Doesn't matter if its one GPU, two or more gpu's acting as one gpu, so the small die approach works better with new nodes, where yields and volume production will be an issue, as those two problems alleviate, the need for smaller dies gets less too as the cost and time to market will end up equalizing to a certain degree.

It'll be fascinating to be sure to see if Raja can finally get a card out that warrants him having that ego of his ;)
 
Reposting this (I was the original author) from reddit. It was a thought that occurred to me as I stupidly took a thread here off topic, and then used the ensuing vacation Kyle was nice enough to give me to do some digging on AMD, Raja, Arctic Islands conversion to Polaris and Vega, and put it all together - I apologize for the obscene length.


--------------------------------------------------------------------------------------------------------------------------------------------------------


And really, for the price, while they may not be world beaters, they are great little card for the current PC standard - 60FPS at 1080p with all the bells and whistles.


Which means, tune out all the people who say "AMD can't afford to keep playing cheap or they will go under." - They are grabbing market share and know what they are doing, even if they have one arm and both legs tied behind their back at the moment to do it.

480x can't do all the bells and whistles at 1080P.

Not sure if that other statement is true either...

It's hard to be optimistic with AMD since for years we heard, "AMD will be great when...."
 
480x can't do all the bells and whistles at 1080P.

I'd say close enough. You can always bump up AA or what not to an absurd enough setting to bring any card to its knees but for most all games there is very little appreciable IQ gain to be found at 1080p vs what the 480x can provide.
 
Hey he said all the bells and whistles. It can't even hit 60fps avg with hair works off at 1080p in TW3.

image.jpg


Don't get me wrong, it's good value, but don't lead some poor smuck into thinking he can max out every game. Far from it.
 
Hey he said all the bells and whistles. It can't even hit 60fps avg with hair works off at 1080p in TW3.

View attachment 6745

Don't get me wrong, it's good value, but don't lead some poor smuck into thinking he can max out every game. Far from it.


Sorry for speaking too loosely. But as the other guy said, pretty much close enough. And perhaps its wrong of me to assume, but I would think you would have a certain aptitude to even think to look for this site, let alone be an avid reader of it's forums.
 
Hehe, that neg func dude over there is a riot. I love reading his posts where he just makes stuff up. I did reply to set the record straight, but I am sure facts are not what he is concerned with.
 
Sorry for speaking too loosely. But as the other guy said, pretty much close enough. And perhaps its wrong of me to assume, but I would think you would have a certain aptitude to even think to look for this site, let alone be an avid reader of it's forums.

Yeah close enough is one way to put it. It might not hit 60 average in TW3 but the analysis here did not ding it:
The Witcher 3 1080p - AMD Radeon RX 480 Video Card Review
 
That's the problem with fanboys. You can praise their favorite product 99% of the time, but they'll focus on that 1% where you say something that isn't "OMG IT'S AMAZING!!!".

Indeed. Just told the nutter that 4chan was this way ->

Seriously, that guy has issues, and not going to waste time fighting with crazy.
 
Well I wont be setting myself up for disappointment this time around no matter how awesome the forthcoming leaks make vega look.
 
It's hard to be optimistic with AMD since for years we heard, "AMD will be great when...."

Well, AMD was last great on 2013 with the 7xxx series, so yeah, it has been "years" as in 2014, 2015, and so far 2016, were they have been price competitive but power has been iffy.

Honestly i love AMD but the power consumption difference does matter if you are trying to get mobile designs, i am damn glad that they have gotten the consoles to give them a small yet steady revenue but they should be aiming for so much more, and i hope that they are. I can rant a bit but maybe they need to either learn how to do a proper power gating of parts of the chips, or get GCN onto the chopping board, it has been a mighty fine chipset family but it needs to evolve, fast...

I expected it to have evolved for this iteration, sure it would be hard a new node + new chip family but dayumn to me this was slightly lackluster.

TLDR: I hope that Zion is right, because i was expecting more from Polaris on the power consumption front.
 
Well, AMD was last great on 2013 with the 7xxx series, so yeah, it has been "years" as in 2014, 2015, and so far 2016, were they have been price competitive but power has been iffy.

Honestly i love AMD but the power consumption difference does matter if you are trying to get mobile designs, i am damn glad that they have gotten the consoles to give them a small yet steady revenue but they should be aiming for so much more, and i hope that they are. I can rant a bit but maybe they need to either learn how to do a proper power gating of parts of the chips, or get GCN onto the chopping board, it has been a mighty fine chipset family but it needs to evolve, fast...

I expected it to have evolved for this iteration, sure it would be hard a new node + new chip family but dayumn to me this was slightly lackluster.

TLDR: I hope that Zion is right, because i was expecting more from Polaris on the power consumption front.


We also need not forget that in the R&D department, AMD are severely lacking compared to their direct competitors, Nvidia and Intell. I think we tend to overlook this important fact. The cards are heavily stacked against AMD and therefore as it stands now, the chances that AMD can reverse this slow downward trend is extremely low unless a new player invests heavily into AMD and their gpu division. R300 type scenarios are always possible but extremely unlikely given this situation. Atm they simply dont have the money to invest where they need to.
 
What every they did in the past prior to AMD's R&D cut backs, were already in the pipeline line, we don't know what the R&D cut backs have done to AMD's graphics division yet, as we haven't see any products with the cut back R&D as of yet, if I'm not mistaken, R&D wasn't cut down until a year or a year and half ago......
 
Yes but even before the cutbacks, R&D was lacking compared to their competitors and due to that had some tough decisions to make in regards to personnel and vision going forward in regards to developer relations and other investments. Its always been an uphill battle for AMD since aquireing ATI to invest as much in as much as their competitors. The fact they have been able to make some great GPUs(5870, 7970) since then is remarkable given their situation.
 
Last edited:
I can't see why you would start a piece with the lines:
I think we will not see real, head to head competition with Nvidia until Navi, because I believe that is going to be the first real graphics chip that is 100% innovated by Raja Koduri. And AMD is holding down the fort with Polaris and Vega as best they can until then.

Promising that 2018 will be a good year for AMD is not that smart. AMD always moved their GPU hardware in a price segment that made sense but why are people so focused on something that will beat Nvidia ?
You can buy any card that you like for a price you like and if that suits your gaming why worry about anything else?
AMD showed with Polaris that it does want to get away from the high end and if the roadmap is true then Vega will get HBM2 which they won't use for "budget" cards.
 
Raja is more a team lead these days than an actual design engineer isn't he? Vega and Polaris were designed by a bunch of guys in China according to his tweets and I bet the same will hold true in the future.

No matter how good he may be, AMD doesn't have the talent or money to simply catch nvidia with one new architecture. These are just fantasies concocted by AMD fanboys that we've all seen over the years. Unfortunately it creates a lot of unwarranted hype that AMD never lives up to and then that vicious little cycle starts again with "just wait until.." for AMD products.

In fact I've suggested this in the past but maybe AMDs official slogan should be "Just wait..", it would keep everyone amused.
 
Only way for AMD to catch nV is if nV really F's ups. Or all console developers use AMD code only lol.
 
I can't see why you would start a piece with the lines:


Promising that 2018 will be a good year for AMD is not that smart. AMD always moved their GPU hardware in a price segment that made sense but why are people so focused on something that will beat Nvidia ?
You can buy any card that you like for a price you like and if that suits your gaming why worry about anything else?
AMD showed with Polaris that it does want to get away from the high end and if the roadmap is true then Vega will get HBM2 which they won't use for "budget" cards.

For me its about competition at the highend and price. As long as Nvidia holds the crown in that respect, prices will continue to be too high.
 
Only way for AMD to catch nV is if nV really F's ups. Or all console developers use AMD code only lol.

Or the unlikely scenario where someone comes along and saves their ass with a huge investment. Its possible. AMD/ATI have some interesting IP/patents. If someone wanted to seriously launch a competitive front against Intell or Nvidia, I think they would have enough Ip to be competitive.
 
True, well if Zen can get AMD out of the first round of bond pay backs, investors and potential buy outs become very attractive so yeah I wouldn't rule that out either!
 
For me its about competition at the highend and price. As long as Nvidia holds the crown in that respect, prices will continue to be too high.
So basically what AMD "does" is "force" you to buy a cheaper card :)
Raja is more a team lead these days than an actual design engineer isn't he? Vega and Polaris were designed by a bunch of guys in China according to his tweets and I bet the same will hold true in the future.
No matter how good he may be, AMD doesn't have the talent or money to simply catch nvidia with one new architecture. These are just fantasies concocted by AMD fanboys that we've all seen over the years. Unfortunately it creates a lot of unwarranted hype that AMD never lives up to and then that vicious little cycle starts again with "just wait until.." for AMD products.
In fact I've suggested this in the past but maybe AMDs official slogan should be "Just wait..", it would keep everyone amused.
Can you explain the correlation between a bunch of guys in China and what AMD does wrong somehow , I'm curious if you can point out the specific problem with guys in China or what they do keep doing wrong according to yourself?
Only way for AMD to catch nV is if nV really F's ups. Or all console developers use AMD code only lol.
That last part would not even make sense since console code does not have a change in hell working on the PC anyway , lol
 
Yeah Zen needs to be a win for sure. A win meaning they need to be close in perf to Intell at cheaper prices. Cheaper pricing cause I dont expect Zen to be faster but if it can get within 10%, that would be remarkable.
 
Last edited:
So basically what AMD "does" is "force" you to buy a cheaper card :)

Can you explain the correlation between a bunch of guys in China and what AMD does wrong somehow , I'm curious if you can point out the specific problem with guys in China or what they do keep doing wrong according to yourself?

That last part would not even make sense since console code does not have a change in hell working on the PC anyway , lol

No it forces me to buy the way more expensive card

I always buy the best at the time of purchase....if both compete at the high end, that means price will naturally go down. If only one of the 2 dominates the high end, prices will go up. So yeah, for my wallet, I would prefer both were within 5% of eachother
 
Last edited:
Sorry for speaking too loosely. But as the other guy said, pretty much close enough. And perhaps its wrong of me to assume, but I would think you would have a certain aptitude to even think to look for this site, let alone be an avid reader of it's forums.

It has nothing to do with aptitude it's facts. I had my Maxwell Titan X on a 1920x1200 monitor and there were a lot of games where it couldn't hold 60 fps with "all the bells and whistles". Games will only be more demanding in the next year especially with VR. Good value card.... tempting to get one just to mess around with but it's been over hyped.
 
Yeah Zen needs to be a win for sure. A win meaning they need to be close in perf to Intell at cheaper prices. Cheaper pricing cause I dont expect Zen to be faster but if it can get within 10%, that would be remarkable.

even if it doesn't get within 10% of the high end I just want to see AMD make money in the mainstream segment, a 5% to 10% increase on their current margins, (10% or higher being optimal) will be enough in the mainstream segment. The problem they are having right now, is selling too low because the price of their chips are dictated by performance parity which always will happen but if they can have smaller dies with the same performance they are getting now, (which of course they will most likely have higher performance with smaller dies), but ya get the idea.

its not about taking over the market with high end components its about increasing margins and ASP's in the volume segments.
 
even if it doesn't get within 10% of the high end I just want to see AMD make money in the mainstream segment, a 5% to 10% increase on their current margins, (10% or higher being optimal) will be enough in the mainstream segment. The problem they are having right now, is selling too low because the price of their chips are dictated by performance parity which always will happen but if they can have smaller dies with the same performance they are getting now, (which of course they will most likely have higher performance with smaller dies), but ya get the idea.

its not about taking over the market with high end components its about increasing margins and ASP's in the volume segments.

I 100% agree. I think people need to lower their expectations. They need a product that is competitive and it doesn't have to match intel. Heck I say if they are within 20% of intel I say its a win. They need to price it accordingly and have some margin. They need a product that say oh you know what amd finally has a chip that is not shit! Good thing is they have been shit for some time that they can't possibly be any shittier on the cpu side. lol. If they can get 30-40% IPC gain compared to their current gen product they can always shoot for more clock speed and further improvement. That is what intel has been doing. They need a base design that is significantly improved.
 
Well people might say isn't that what AMD is doing in the graphics card market at the mainstream level, it isn't, they are lowering margins for the rx480, Raja even stated this they aren't getting more margins with the rx480 comparative to the rest of AMD. This is a major problem for AMD, they need to increase margins and volume *regaining* marketshare at the same time. They both go hand in hand, if people think you have good products you can sell them for higher and you will still sell more of them unless they are ridiculously priced.
 
What every they did in the past prior to AMD's R&D cut backs, were already in the pipeline line, we don't know what the R&D cut backs have done to AMD's graphics division yet, as we haven't see any products with the cut back R&D as of yet, if I'm not mistaken, R&D wasn't cut down until a year or a year and half ago......

What?

AMDs R&D cuts began big time ~4 years ago.
Advanced Micro Devices Research and Development Expense (Quarterly) (AMD)

This is also why you see products today that is pretty much just old wine on new bottles. And for the RTG division its even worse, since money gets used on Zen. The changes between GCN 1 and GCN 4 is tiny. And so will any future changes as well. RTG is working with a skeleton crew and hardly any resources. Rule number 1, AMD is a CPU company first and always.

Its only getting worse in the future.
rd1.png
rd2.png
 
Last edited:
Fuck r&d just put 8000 shaders up in that bitch!!!

-Sarcasm over

The above chart makes sense. Starting going downhill after 290 series. That was still a decent series. Then rehash and then hbm. Nothing really major in last 3-4 years. Here is to hoping that may be Vega might actually be something different because we haven't really gotten anything substantial since 2012. Atleast give me some high clocks with it damnit!!!
 
What?

AMDs R&D cuts began big time ~4 years ago.
Advanced Micro Devices Research and Development Expense (Quarterly) (AMD)

This is also why you see products today that is pretty much just old wine on new bottles. And for the RTG division its even worse, since money gets used on Zen. The changes between GCN 1 and GCN 4 is tiny. And so will any future changes as well. RTG is working with a skeleton crew and hardly any resources. Rule number 1, AMD is a CPU company first and always.

Its only getting worse in the future.
rd1.png
rd2.png

Ah I remembered the last round of cuts.
 
I don't understand why every AMD thread has to end in doom and gloom. I don't care about R&D dollars, this is not the thread for it. Take the trolling elsewhere.
 
I don't understand why every AMD thread has to end in doom and gloom. I don't care about R&D dollars, this is not the thread for it. Take the trolling elsewhere.

R&D spending gives you a good hint of how much a company is investing in future products, which in turn gives you a hint about how competitive those future products may ultimately be.
 
I am not entirely sure how accurate that chart continues to be - I'd love to see what money is being invested in R&D now - not so much believing that it would be back to 2010 levels, but I do think we'd see a larger investment than has been in the past.

Also, I see no problem with the majority of cash being funneled to Zen, provided it succeeds. If Zen does end up a success and makes AMD flush with cash, that will also help the RTG group - a rising tide raises all ships.

And frankly, I don't know if the RTG group can survive at this point in time without Zen being a success for the company. Vega in that case would need to be a large enough surprise that it buys the company time, and even in that event, I think it would be reasonable to think that the RTG group gets sold off to someone else as a stop gap for the AMD bleeding. If that happens, I'll be selling my shares ASAP.

That being said, the things I have heard thus far about Zen have been good, and I learned from following Polaris from the time of the dawn of the async compute debate until now that the earlier rumors tend to be more true than the ones that come months within launch.

If that's the case, then if the early rumors are true, Zen will be at Haswell level performance with a high OC overhead. Given the minimal gains over Haswell by skylake and the diminishing returns expected with Kaby Lake, I think Zen has a legit shot to be a huge win for AMD, but I also think much of that is also going to depend on the chipset and how many advanced features it can support. Case in point - if Zen has Haswell performance, but the chipset is superior to intel's in what it can support (perhaps more PCIE lanes for example), then I could see many enthusiasts flocking to Zen given the price point, especially if the chipset can accommodate multiple PCIE 3.0 lanes at full 16x, vs the 16x8x4x4 or the 8x8x8 we usually see from intel.

Price and chipset, provided Zen is Haswell, will win the day for AMD, or lose it for them.
 
If that's the case, then if the early rumors are true, Zen will be at Haswell level performance with a high OC overhead. Given the minimal gains over Haswell by skylake and the diminishing returns expected with Kaby Lake, I think Zen has a legit shot to be a huge win for AMD, but I also think much of that is also going to depend on the chipset and how many advanced features it can support. Case in point - if Zen has Haswell performance, but the chipset is superior to intel's in what it can support (perhaps more PCIE lanes for example), then I could see many enthusiasts flocking to Zen given the price point, especially if the chipset can accommodate multiple PCIE 3.0 lanes at full 16x, vs the 16x8x4x4 or the 8x8x8 we usually see from intel.

Price and chipset, provided Zen is Haswell, will win the day for AMD, or lose it for them.

If AMD can match Haswell for single thread, and dump 8core/16thread to sweeten the deal, I feel it will be a success if priced correctly. Games coded for more threads will become more common as time goes on as we hit physical limitations on core speeds and diminishing returns on IPC improvements.

I guess I should also get back on topic:

Navi will be an interesting proposition if they make it a patchwork approach. Stitch your smaller parts together that have higher yields to make one big monster.

if you cram much lower cost parts together you can cobble them to be much bigger than the current 600mm max target die size even without the insane costs due to yields.

Easier to scale for different pricing points to boot.
 
Last edited:
If AMD can match Haswell for single thread, and dump 8core/16thread to sweeten the deal, I feel it will be a success if priced correctly. Games coded for more threads will become more common as time goes on as we hit physical limitations on core speeds and diminishing returns on IPC improvements.

I guess I should also get back on topic:

Navi will be an interesting proposition if they make it a patchwork approach. Stitch your smaller parts together that have higher yields to make one big monster.

if you cram much lower cost parts together you can cobble them to be much bigger than the current 600mm max target die size even without the insane costs due to yields.

Easier to scale for different pricing points to boot.


I don't think Navi will be that though. My expectation is that this will be an entirely new design.
 
R&D spending gives you a good hint of how much a company is investing in future products, which in turn gives you a hint about how competitive those future products may ultimately be.


Yes and no, having multiple design teams working on different gens your R&D goes up, nV has at least 2 on tegra and 2 on Geforce, then adding in other teams for neuronets and what not working on different generations and possible even back ups. Its all a common IP but different teams focused on different aspects gens and fall back plans. AMD doesn't do this as they are only focused on Desktop Graphics for the time being for one generation ahead and possible a back up plan. If something doesn't come out as planned, they aren't as nimble and quick to get a replacement, that is true, but the time it takes even to spin up a back up takes quite a bit of time, we have seen it any where from 1 year to 2 years on the GPU side.

So AMD essential has to make things work out of the gate otherwise for them it will take longer than nV to fix something or push ahead a next or implement a back up. Simply put "skeleton crew" is not the exact right word, but something like bare necessities would be better. They only have enough experienced engineers to work on one product at a time.
 
I don't think Navi will be that though. My expectation is that this will be an entirely new design.

I think it will be both. If they want multiple chips to talk to each other without the issues with crossfire, they'll have to redesign the architecture significantly.

Navi is pegged as "scalability" first on the road maps, with "nexgen memory" second. I think this is a big hint where they're going. This may end up being the Athlon dual-core push all over again.

AMDGPU.jpg
 
Back
Top