Radeon RX Vega for consumers might still be out a couple of months

No your right the most important metric to judge a company by is not its value or profits. It's how the customer feels when they buy the product.

Oh wait no it's not.

 

The market value is not a proper appreciation of a company. It's a good indicator, and NVIDIA is doing GREAT. But they're also riding a very specific wave almost by accident.

They had the same CEO when they f*cked up all their deals with Microsoft, SONY and Samsung, and when they released Fermi and the GeForce FX. CEOs aren't the beginning and end of anything.

 

Fermi turned out alright. FX was very early in the company's history. Iirc M$ and Sony wanted x86 and Nvidia doesn't have a x86 license. Meh Samsung appears to be run by a man-child half the time with their add campaigns (not the resent ones).

Look me and you are not going to see eye to eye on this one.

 

You know nothing about technology and possibilities HBM opens. Luckily some people at AMD do understand that there is more than today. And understand their limited cash flow.

If AMD went GDDR5X route, there would never be single HBM chip prototype and stuff that will come within next 3 years with HBM goodness would never exist.

IIRC nVidia's sales pitch had no chance because of prior horrid experience those companies had with nVidia.
And nVidia's support did suck badly too apparently.

When client comes and asks for technology to do this and that, AMD delivers what client asks for. (Semi-Custom Division)
When client approaches same way nVidia, they will tell him: "This is what we have, this is what you get!"

 

Over that 5 years having taken the company from non-competitor to able to compete and sporting a solid 5 year plan with technology that is highly competitive and built for the future, spending resources setting up an ecosystem for consumers that makes the highly threaded CPUs and cheaper GPUs they currently have on market competitive as well?
That's not very incompetent... if your measure for doing a job badly is making that amount of progress, coming from absolutely nothing, I don't think you have much of any weight behind your criticism.

It's "makeshift", btw.

 

Sorry for emotionally challanging you, thanks for posting Lisa's answer.

 

All of this said by a guy who owns a 1070, at least you're not using an 7700k (Yet).
A company can't grow if no one buys their products, Nvidia fanboys made this possible. I for once, in 2013, opted to go red with my GPU and very happy about it.
Also, after bulldozer fiasco the only thing they could have done was start from the ground up and we now have Ryzen. Designing a CPU/GPU takes years, it's not like you COD fanboys are used to games every year: it's just a mod.

Tell that to Nvidia and all their cards that beat yours.
Nvidia used HBM wisely: on expensive products bought for specific purposes.
Do you think your card is better just because it has HBM? It would run the same with 8Gb of GDDR5. AMD should have pushed HBM further in the future, not now that they're so behind.

 

You realize that a every piece of HBM sold has AMD patents in it, right? And that they hold almost all the patents for the interposer. It's not as if this won't pay off.

 

AMD targeting the server/business segment with their products seems like the best decision they can make. Am I wrong in thinking this?

I think it's been discussed before but, they probably couldn't afford to split the design at this point.

 

AMD doesn't collect royalties on HBM.

https://www.kitguru.net/components/...ng-usage-of-hbm-and-do-not-collect-royalties/

 

What about the interposer?

 

Wow! You bought a $190 GPU from them 4 years ago. How cute. I bought 2 $300 GPU's from them (AMD) 6-7 years ago and a $550 GPU from them (AMD) 5 years ago. So because I choose the GPU that gave me the performance I desired (performance that AMD did not offer mind you) I cant have an opinion. So basically I should shut up according to you unless I have skin in the game? Get real. Hey why don't you go buy an Ryzen system and a 580 then you can lecture me about my buying habits.

The fact that the first thing you look at is what GPU I use shows who the fanboy is here.

 

Idk, you'd have to link me to the patent to see the specific implementation. Interposers themselves have been used prior to HBM. If it does exist the patent is related to HBM and an interposer and probably really specific language regarding it's novelty.

Like this for example:

https://www.google.com/patents/US20130346695

Is insanely specific on how it's integrated.

Should also point out that the context of that quote is in reference to on die packaging anyway. So it would seem really weird to me that AMD's response to it would be specifically for HBM and not the way it's packaged, which would obviously include the interposer.

 

1) I agree, as Raja said: last year they didn't had an AI learning card, low they have.

2) I understand it's not easy, but it was a bad management decision going with HMB, delaying the card more than 12 months and leaving Nvidia to do as she pleased.

I didn't comment on his post because I had no info on this matter.
So my theory stands true: AMD ****ed up insisting in HBM.

It costed 205€ to be more exact. I'm sorry if currently I'm poor and can't afford to upgrade every year.

For reference: my first GPU ever was a 8mb ATI card (don't know the model). I still have my working Nvidia FX5200 on a P4, My FX6600 is gone, my 7600GS died, my 9800GTX was sold (sold the Intel E8400 too), my GTX460SE was sold and now I have this R9 270X. ATM I'm trying to sell this GPU to get a RX570. Am I really that of a fanboy having had 5 green cards and just 2 red?

PS: Yes, when I can afford it you bet I'm doing a Ryzen build.

 

We don't have access to all the information - it's easy to say in hindsight that AMD should have done something differently. Last year when Vega was announced the roadmap pegged it at early 2017.

https://www.extremetech.com/wp-content/uploads/2016/03/AMDGPU.jpg

Polaris falls where it launched, June/July 2016.. Vega falls right on the 2017 - which would indicate to me that either they said "**** it" and put Vega above 2017 because they knew it would launch sometime this year, or maybe they thought they'd have it out early 2017.

If the latter is the case, I think it all makes sense. Like if you're AMD and it's June 2016 and your launching Polaris to cover 90% of the market and you think you can get a Vega chip to cover the top 10% by Q1 2017 - why would you spin a Vega with a GDDR controller on it? And if HBCC can read GDDR, why would you even think about designing it? I think as that time approached, it became increasingly clear that they weren't going to hit that timeframe for whatever reason.

But the bring up on a chip that people would have liked to see in hindsight would have needed to start design mid last year. And if AMD thought they could have Vega out by Q1/Q2, why would they have designed that chip and what point would it have been too late?

 

Thanks for the interposer links. I believe that they really have HBM issues. NVIDIA using it means nothing, since the actual production required to cover those needs is miniscule compared to anything close to the mainstream market. I believe that they have ramp up issues, and that they might have expected NVIDIA to use it more, so the whole production line would be more streamlined by now.

 

We have that information. Fiji without HBM would be 10~15% weaker due to memory + IMC power consumption. And Fiji did save transistors left and right thanks to HBM. + Form Factor of Graphics cards.
If AMD went GDDR5X way, that top card would be just another Hawaii (Big, Hot, Power Hungry). And would not be that attractive.
Stuff AMD developed alongside HBM will one day change/design shape of All-In-One computers/notebooks.

Even today AMD can make APU with 8GB HBM2 on interposer with everything important integrated. Place that on PCB which just holds power delivery circuits and protection chips for USB/SATA/DP/...
They can deliver entire working computer in size of MXM3.0 card. It is just question of time. I expect it to be here by 2020 as they will not release product until they believe there will be good demand for it.

 

From a business perspective AMD did the right thing. But it´s hard not to feel "shafted" with this, like suddenly gamers are becoming secondary to AMD/Nvidia...
Not to mention AMD is giving the high end market to Nvidia in a silver plate.

That could be all true but i doesn´t change that in the end AMD has nothing on the high end GPU market. It would be better to have a slower part with GDDR5 than to have nothing at all.
Vega may well be the best GPU the universe has ever seen so far but that doesn´t matter if you can´t buy it...

 

But all that would come anyway..

The information I'm looking for is what are the lost sales for potentially 6 months of complete abandonment of the "10%" market, vs sales if they released a GDDR variant of Vega in Q1 this year and then, more importantly, when did AMD realize Vega wasn't going to be launched until Q2 potentially Q3? People argue the GDDR would be way more attractive than nothing. Which is what it's up against. But my point is that the argument is moot.

We don't know what AMD's roadmap was for Vega and without it questioning whether or not it was a good idea to build a GDDR card is irrelevant because if two years ago they thought they could build the HBM card by Q1 2017 - why would they do a GDDR variant in the first place? And basically the only information we do have, indicates that AMD thought they could get the card out by then.