[WCCF] AMD Radeon R9 390X Pictured

[maddie]

From Hardware.fr:
http://www.hardware.fr/news/14241/computex-amd-fiji-ses-4-go-hbm-photo.html

Fiji is limited to 4GB. They didnt care for more with HBM1.



What has Tonga do with AMD's statements about the memory size?
Tonga uses a color compression from the memory chips to the GPU. It doesnt help to optimize the amount of data.



They have been selling cards with less or the same amount of memory since years. They sold a $1500 graphics card with 4GB last year and didnt bother to look after the memory utilization... It's marketing. The competition sells card with more memory.

google translate had this on the linked page.

"So far we had not dropped the possibility that AMD can embark 8 HBM modules from 4 to 8 GB by sending them through four channels of 2 Gb (or in 512-bit mode) instead of 8 channels of 1Gb (fashion 1024-bit). A not very complicated flexibility to integrate into an HBM module as one designed by SK Hynix. Just after AMD's conference, we were able to put the question directly to Joe Macri, AMD CTO and Chairman of the Subcommittee JEDEC DRAM. It explained that although in fact relatively simple to implement, this possibility was not foreseen in the HBM 1."

This sounds a lot like the dual link tech some spoke about.

Are you saying it will not exist as Macri says its simple to implement.

I have a hard time accepting at face value that a technique thats simple to implement was not foreseen and abandoned.

I feel AMD wants and needs a knockout punch at this point in time so are misdirecting until official release.

For those claiming so much lost sales, I have two questions.

Do you believe that the market for 980TI and TitanX class cards will be fully saturated with 2-3 weeks of sales? That this class of products will cease to sell until 14nm arrives?

 

[jackstar7]

Doesn't change a thing as it has to go through the vastly slower PCIe interface.

You're sure there are going to be no efforts to optimize the resources to take advantage of everything?

 

[sontin]

4GB is 4GB, or? A $1500 295x2 with 4GB is limited by the amount of memory. If they are believing in better optimizing of the memory why havent they started a long time ago?

HBM2 will increase the amount of memory at least 4x times over HBM1 or Fiji. Next year at this time people will talk what they can do with 16GB oder 32GB of memory and nobody will discuss how great 4GB for a high-end GPU is.

I feel AMD wants and needs a knockout punch at this point in time so are misdirecting until official release.

This doesnt make sense. They are talking about HBM1 and didnt care to put an end to discussions like the memory limitation? A simple "We will launch with 8GB" would be enough. That would be a "knockout punch". Saying nothing will only increase the suspicion.

 

[SimianR]

4GB is 4GB, or? A $1500 295x2 with 4GB is limited by the amount of memory. If they are believing in better optimizing of the memory why havent they started a long time ago?

HBM2 will increase the amount of memory at least 4x times over HBM1 or Fiji. Next year at this time people will talk what they can do with 16GB oder 32GB of memory and nobody will discuss how great 4GB for a high-end GPU is.

Right, and for all these people that buy $600+ GPU's to simply swap them out when the next best thing comes out, it won't really matter. You are right that 1-2 years from now there might be a situation where 4GB becomes a bottleneck at certain settings, but you'll have a tough time finding one now.

 

[swilli89]

From Hardware.fr:
http://www.hardware.fr/news/14241/computex-amd-fiji-ses-4-go-hbm-photo.html

Fiji is limited to 4GB. They didnt care for more with HBM1.



What has Tonga do with AMD's statements about the memory size?
Tonga uses a color compression from the memory chips to the GPU. It doesnt help to optimize the amount of data.



They have been selling cards with less or the same amount of memory since years. They sold a $1500 graphics card with 4GB last year and didnt bother to look after the memory utilization... It's marketing. The competition sells card with more memory.

Dude what? Radeons have always had more ram, pretty much every single generation.

 

[AnandThenMan]

Look at Nvidia's best sellers the last 5+ years then come back and say memory is the end all be all.

 

[DDH]

4GB is 4GB, or? A $1500 295x2 with 4GB is limited by the amount of memory. If they are believing in better optimizing of the memory why havent they started a long time ago?

HBM2 will increase the amount of memory at least 4x times over HBM1 or Fiji. Next year at this time people will talk what they can do with 16GB oder 32GB of memory and nobody will discuss how great 4GB for a high-end GPU is.

Because it was a case of just adding more GDDR5. It was easier than optimizing the use of the current mem. That has been explained previously



Maybe, maybe not. Maybe next year at this time we will talk about how 4GB is perfect for 4k with decent optimization, and how wasteful it was of GPU manufactures when they ignored it in favor of higher memory size and marketing.

Incidentally, if 4gb of HBM is more than sufficient for 4k gaming, would you be happy to pay a premium for 8gb or 16 or 32?

 

[Despoiler]

Right, and for all these people that buy $600+ GPU's to simply swap them out when the next best thing comes out, it won't really matter. You are right that 1-2 years from now there might be a situation where 4GB becomes a bottleneck at certain settings, but you'll have a tough time finding one now.

Not to mention the next generation is a node shrink + HBM2 for both companies. Everyone will buy new cards because of it.

 

[Noctifer616]

How will a 4GB limit impact their FirePro cards? They stated they wanted to increase their market share in the professional graphics marker this year and a big chip like Fiji would be great for that. Are they just going to make a bigger interposer and use more stacks?

 

[maddie]

4GB is 4GB, or? A $1500 295x2 with 4GB is limited by the amount of memory. If they are believing in better optimizing of the memory why havent they started a long time ago?

HBM2 will increase the amount of memory at least 4x times over HBM1 or Fiji. Next year at this time people will talk what they can do with 16GB oder 32GB of memory and nobody will discuss how great 4GB for a high-end GPU is.



This doesnt make sense. They are talking about HBM1 and didnt care to put an end to discussions like the memory limitation? A simple "We will launch with 8GB" would be enough. That would be a "knockout punch". Saying nothing will only increase the suspicion.

That is your opinion. Having invested however much $ into a big launch, maybe they want maximum effect instead of a dribble here and there.

Your views please on the two questions below.

1) Do you believe that the market for 980TI and TitanX class cards will be fully saturated with 2-3 weeks of sales?

2) That this class of products will cease to sell until 14nm arrives?

 

[maddie]

How will a 4GB limit impact their FirePro cards? They stated they wanted to increase their market share in the professional graphics marker this year and a big chip like Fiji would be great for that. Are they just going to make a bigger interposer and use more stacks?

Firepro card sales will suffer greatly having 4GB as the limit.

That is why 4GB limit makes little sense.

Yes, I believe a larger interposer + more stacks is the solution. Interposers are not expensive or difficult to make.

 

[flopper]

Dude what? Radeons have always had more ram, pretty much every single generation.

yea the 970 3.5gb hasnt enough ram anyone bought those should get a free titan x as a compensation same goes for the 4gb 980 also. its not enough..

look forward the new fury as its godlike new tech.

 

[tential]

Firepro card sales will suffer greatly having 4GB as the limit.

That is why 4GB limit makes little sense.

Yes, I believe a larger interposer + more stacks is the solution. Interposers are not expensive or difficult to make.

Maybe 8gb was only saved for those cards then to avoid issues? Maybe we get the 4gb model now and then 2-4 months from now we get the 8gb model? Amd would be taking a page out of the nvidia book in staggering your releases to maximize the amount of money your consumers spend.

As for why they won't leak/release info? It's amd.... They literally do the exact opposite of what a company is supposed to do marketing wise.

 

[Ajay]

I'm a bit confused here. I thought the Fiji module shown by Lisa Su and others showed 4 HMB 1 stacks. That's a stack of 4 dies times 4 stacks times 1/2 GB = 8 GB. What is all the confusion about?

 

[Silverforce11]

How will a 4GB limit impact their FirePro cards? They stated they wanted to increase their market share in the professional graphics marker this year and a big chip like Fiji would be great for that. Are they just going to make a bigger interposer and use more stacks?

If its 4GB now, Firepros will be out later when they can design more vram capacity into it. It's true that the low vram will gimp HPC hard considering the current lineup has the W9100 with 16GB!

They've been successful in clawing into NV's HPC marketshare so that area will definitely be their focus.

 

[swilli89]

I'm a bit confused here. I thought the Fiji module shown by Lisa Su and others showed 4 HMB 1 stacks. That's a stack of 4 dies times 4 stacks times 1/2 GB = 8 GB. What is all the confusion about?

Haha apparently you are! J/K I'm just poking fun.. Not sure where you are getting that equation from though but a 4 layer stack of HBM1 memory is 8 gigabits. That's 1 Gigabyte. Four stacks of that and we have 4 Gigabytes.

 

[Noctifer616]

I'm a bit confused here. I thought the Fiji module shown by Lisa Su and others showed 4 HMB 1 stacks. That's a stack of 4 dies times 4 stacks times 1/2 GB = 8 GB. What is all the confusion about?

1 stack has 4DRAM chips and each chip is 2Gb, so 8Gb total for a stack which is 1GB. That's what it's assumed to be.

 

[maddie]

@ SONTIN

Your views please on the two questions below.

1) Do you believe that the market for 980TI and TitanX class cards will be fully saturated with 2-3 weeks of sales?

2) That this class of products will cease to sell until 14nm arrives?

 

[takiat]

ok, so I have a friend working at AMD for over 10 years, and I spoke to him today. I asked him if I should buy the 980ti or wait for the Fury.
He told me all the talk around the office today was about Nvidia being scared of the 300 series with the 980ti being priced so low. This is because AMD is about to destroy the Nvidia lineup in performance. It is going to be much, much better than the usual bump in perf. He mentioned it will be one of those times where you know it's a day one purchase. So he told me to wait, and I would not only NOT be disappointed, but I would be unbelievably surprised.

Now, you might say he wants me to buy AMD but seriously, is my ONE GPU purchase really that important to AMD that he would steer me wrong just so AMD gets a sell and we end our friendship???? No way.

So, of course this is only his opinion, but he is a gamer like me for many years, so I am pretty sure his opinion lines up with my expectations for a "wow" moment.

Take it as always with a grain of salt but there is no way I am buying a 980ti until the 390/390x Fury comes out.

 

[Kenmitch]

So tell us again how you know this as a FACT?

It was in last night's propaganda update to Nvidias guide to world domination. Think it was page 2....Hehe....J/K 🙂

 

[Noctifer616]

snip

You really shouldn't post this as there is no way for us to confirm what you are saying is true.

 

[DarkKnightDude]

Take it as always with a grain of salt but there is no way I am buying a 980ti until the 390/390x Fury comes out.

Might as well wait, since there's no 980 Ti aftermarket versions yet.

 

[Kenmitch]

You really shouldn't post this as there is no way for us to confirm what you are saying is true.

Just another rumor. Anybody that is actually willing to buy AMD is in a holding pattern anyways. Some are pretending to be just to save face.

 

[.vodka]

Well for Fiji itself we know it could easily blow nV's high end out of the water with that die size. But the rest of the lineup? That sounds like good news! Taken with a grain of salt, of course.

And I'd agree the logical path of action is to wait until AMD releases their new lineup in three weeks, why buy now if you've waited this long.


I know if I were still gaming on my 6850 and hadn't upgraded to a 290 last november after 4 years, I'd wait a little more.

 

[RampantAndroid]

ok, so I have a friend working at AMD for over 10 years, and I spoke to him today. I asked him if I should buy the 980ti or wait for the Fury.
He told me all the talk around the office today was about Nvidia being scared of the 300 series with the 980ti being priced so low. This is because AMD is about to destroy the Nvidia lineup in performance. It is going to be much, much better than the usual bump in perf. He mentioned it will be one of those times where you know it's a day one purchase. So he told me to wait, and I would not only NOT be disappointed, but I would be unbelievably surprised.

Now, you might say he wants me to buy AMD but seriously, is my ONE GPU purchase really that important to AMD that he would steer me wrong just so AMD gets a sell and we end our friendship???? No way.

So, of course this is only his opinion, but he is a gamer like me for many years, so I am pretty sure his opinion lines up with my expectations for a "wow" moment.

Take it as always with a grain of salt but there is no way I am buying a 980ti until the 390/390x Fury comes out.

Leaking info like this is probably something he doesn't appreciate? He's NDAed, and just talking to you is likely an offense he could be fired for.

If I were in your shoes with inside info like that, I'd pass on posting it - for the sake of my friend.