[WCCF] AMD Radeon R9 390X Pictured

[exar333]

Nvidia has tons of products too. Are you seriously suggesting it's normal for nvidia to have a product lineup of
960
970
980
980ti
Titan x

? We all know they'd have refreshed the whole gpu stack but 0 competition to do so. Don't act like a 960ti 950 950ti and 940 aren't the norm....

People seriously are in these threads rewriting history right now.

What history is being re-written exactly based on what I posted? Please explain.

All I am saying is that AMD has so many products squeezed from the 270-285 that really don't differ all that much. NV opts for a simpler product stack. [Edit: And this looks to continue with the rumored 3xx SKUs]

What is difficult to understand?

I am not saying NV ALWAYS did this. But look at their recent success since Kepler was released. They have followed a consistent, simplified approach. This includes more top to bottom refreshes (6xx to 7xx then to 9xx) but each stack is relatively simple.

The number seem to correlate to some rather large gains too. Not saying this is causal in relationship, but for the average 'stupid joe' gamer, they just tell each other to grab the 970 or the 960. It's simple...

Look at the graph below

http://www.dsogaming.com/news/amdnvidia-market-share-graph-shows-nvidia-conquering-3-out-of-4-pc-gamers-own-an-nvidia-gpu/

Sorry - getting a bit off-topic. Will end this discussion here...

Edit: GPUs are becoming like our phones. New, shiny versions every year. Even thought the 'guts' may be often recycled to some degree. It is what it is.

 

[jamesgalb]

As .vodka stated, the 290X PCB might be compatible with the new 390X GPU. A PCB is not a GPU. You keep assuming things to fit your theories. And it's starting to get old.

If AMD pulled off an instant PCB swap for their partners, then a $400 launch price seems very possible/likely for a 8GB card with a respun Hawaii 'XTX' chip.

 

[jamesgalb]

What history is being re-written exactly based on what I posted? Please explain.

NVidia runs more products/skus... you tried to write it where a partially released Maxwell lineup reflect nVidias normal product/sku count.

 

[exar333]

If AMD pulled off an instant PCB swap for their partners, then a $400 launch price seems very possible/likely for a 8GB card with a respun Hawaii 'XTX' chip.

A $400 price would support a ~20% speed-bump over the 290x as well, if you compared directly to the 970. The 290x and 970 are pretty close today, trading blows in different games. That is assuming price differences roughly reflect performance. (not always an accurate metric, but since this is all speculation)

Definitely plausible.

 

[exar333]

NVidia runs more products/skus... you tried to write it where a partially released Maxwell lineup reflect nVidias normal product/sku count.

I was focusing on the key market, ~$200 and up. The trash OEM SKUs really don't matter a whole lot. I don't disagree there are a lot of those, but they are just filler and re-hashed stuff for low-end stuff.

Show me where 8 is less than 5?

This isn't an emotional post (please don't turn it into that). Just linking to info. Not even judging the SKus in question. I don't use an iPhone, but I understand why they are successful and simple in what is offered. Even if that product barely changes year over year, at times.

 

[tential]

NVidia runs more products/skus... you tried to write it where a partially released Maxwell lineup reflect nVidias normal product/sku count.

You now see why I'm at the point where I'll just leave this forum and not post hardware. These discussions at this point are a very small step up from neogaf type discussions....

 

[Azix]

For once I like wccftech. Just for agreeing with my estimates anyway. A bigger die might be more impressive, but i care about cost.

he Fiji die itself should measure around 480-540mm^2 (reverse engineering it from the pictures).

Read more: http://wccftech.com/amd-fiji-xt-bas...y-x-camera-interposer-pcb-size/#ixzz3c1uOg9WR

 

[Hitman928]

I was focusing on the key market, ~$200 and up. The trash OEM SKUs really don't matter a whole lot. I don't disagree there are a lot of those, but they are just filler and re-hashed stuff for low-end stuff.

Show me where 8 is less than 5?

This isn't an emotional post (please don't turn it into that). Just linking to info. Not even judging the SKus in question. I don't use an iPhone, but I understand why they are successful and simple in what is offered. Even if that product barely changes year over year, at times.

From the previous page:

GTX 760
GTX 760 Ti
GTX 770
GTX 780
GTX 780 Ti
GTX Titan
GTX Titan Black
GTX Titan Z

8 = 8

Maxwell most likely hasn't been fully released yet so using it as a comparison is nonsensical.

 

[exar333]

You now see why I'm at the point where I'll just leave this forum and not post hardware. These discussions at this point are a very small step up from neogaf type discussions....

The issue seems to be more defensive, emotional responses. I am not sure what's up today, but there is a LOT of reactivity on here today in VC&G. Just an observation.

All we have to go by are leaks, rumors and some VERY limited official info. Let's all just take that with a grain of salt and make some educated guesses and civil discussions. At this point that's all we can do.

If that's too much for someone, I might suggest they don't post and keep our threads rid of all the toxicity.

Back on-topic.

 

[exar333]

From the previous page:

GTX 760
GTX 760 Ti
GTX 770
GTX 780
GTX 780 Ti
GTX Titan
GTX Titan Black
GTX Titan Z

8 = 8

Maxwell most likely hasn't been fully released yet so using it as a comparison is nonsensical.

For the last time, I am not comparing Maxwell against GCN 1.2 (or whatever).

NV's current products are TX, 980Ti, 980, 970, 960 for ~$200 and above. Yes you can buy a 780, but that is not a current item. You can still get ATI branded cards on Newegg.

28nm has been here for a LONG time. We got 2 Kepler releases and 1 Maxwell family release. I am sure we may see some more before it's all done...

 

[swilli89]

You guys going forth about SKU proliferation are hilarious. Who seriously cares about that metric? Just let it go!

I really hope that last WCCF article projecting FIJI to be at 480mm^2 is dead wrong. We really need this die to be close to 600^2 with those denser libraries..

 

[jamesgalb]

I was focusing on the key market, ~$200 and up. The trash OEM SKUs really don't matter a whole lot. I don't disagree there are a lot of those, but they are just filler and re-hashed stuff for low-end stuff.

Show me where 8 is less than 5?

This isn't an emotional post (please don't turn it into that). Just linking to info. Not even judging the SKus in question. I don't use an iPhone, but I understand why they are successful and simple in what is offered. Even if that product barely changes year over year, at times.

Last Gen

r9 270
r9 270x
r9 280
r9 285
r9 280x
r9 290
r9 290x
r9 295x2

GTX 750
GTX 750 Ti
GTX 760
GTX 760 Ti
GTX 770
GTX 780
GTX 780 Ti
GTX Titan Black
GTX Titan Z

Current Gen (announced/leaked)

Radeon Fiji X
Radeon Fiji
r9 390x
r9 390
r9 380
r9 370

GTX 960
GTX 960ti
GTX 970
GTX 980
GTX 980ti
GTX Titan X

Typically NVidia runs around 8 SKUs at this level your speaking of. the 900/Maxwell series is not a proper representation of a full NVidia lineup, and comparing it to a 'complete' lineup from last gen is trying to re-write the realities of history.

 

[tential]

From the previous page:

GTX 760
GTX 760 Ti
GTX 770
GTX 780
GTX 780 Ti
GTX Titan
GTX Titan Black
GTX Titan Z

8 = 8

Maxwell most likely hasn't been fully released yet so using it as a comparison is nonsensical.

When the 900 series adds more skus we'll praise nvidia for fleshing their lineup out.
900 series is great because there are few products and it's simple!
900 series is great because they cover every price point!

Nvidia never loses. From a company/investor standpoint it makes me smile so much from a business perspective. I'm not even sure who is better nvidia or apple.

 

[maddie]

For once I like wccftech. Just for agreeing with my estimates anyway. A bigger die might be more impressive, but i care about cost.

They are saying 510mm2 +/- 30mm2.

No way in my opinion. One possible error is what they used as the HBM module. If 5 x 7 instead of the actual 5.48 x 7.29. then you get a 14% error.

This gives us Fiji using their measurements at 581mm2 +/- 34mm2

 

[SPBHM]

Like NVidia?

GT 705
GT 710
GT 720
GT 730
GT 740
GTX 750
GTX 750 Ti
GTX 760
GTX 760 Ti
GTX 770
GTX 780
GTX 780 Ti
GTX Titan
GTX Titan Black
GTX Titan Z

hidden behind these model numbers they have Fermi, Kepler and Maxwell lol

 

[Wag]

definition
Die = GPU die


I mean one Fiji model because the Die will be the same, but the interposer size will be different if more than 4 stacks HBM.
Its speculation, I know, but we know that there can be more or less than 4 stacks of HBM memory. Confirmed by AMD.

I have trouble with accepting that AMD will make a flagship possibly as fast as 295X2, and only put 4GB HBM maximum. It makes no difference if they can optimize the 4GB so there is no appreciable performance shortfall. They must know what Nvidia marketing will do in that situation.

I also see AMD as having a huge window of opportunity in professional/scientific markets as Nvidia dropped DP performance in Maxwell. 4GB is a huge problem in those markets.

They must have higher than 4GB models.

4GB will kill them this round if only for the fact that 4k gaming is the way things are going. I'm running @ 4k right now with 3 OG Titans, I can't honestly see myself going back to 4GB cards.

Even if their new card is significantly faster than the 980Ti @ 4k with only 4GB I'm not sure I'd buy it, as it might not be future proof, and I'm not sure I'd be willing to wait until Aug (or later) to buy a card with more memory. I'm ready to buy now.

 

[Vesku]

"future proof" no such thing when it comes to high end GPUs especially when the next gen will be a node shrink.

 

[stuff_me_good]

In reality the 45% increase in shaders will only apply to the shader portion of the die which is about 60% of the total area. This leads to (45 x 0.6) = 127% increase in total area not 145%.
This actually means that more shaders can be fitted than you calculated.

Your other points are spot on.

Remember the 4870. Forget about the low price now, NO ONE expected 800 shaders. It was thought to be impossible for the die size.

Could we actually be looking Fiji chip = twice Hawaii in shaders?

Everyone is forgetting this slide from Synapse. Are those 80+ blocks shader clusters of 64 shaders each, giving 5120 + shaders in Fury.
>500 mm2 @28HPM, 1 GHZ, 80+ Blocks

You are right, didn't realize that the shader part is only 60% of the die. Also didn't remember the fact that high density libraries with GF process brings even more space for more transistors. So if the die is 600mm^2, 4096 shaders sound quite small amount for that size of die. Unless there are ton of new compute shit that is not utilized on games or something like that... don't know??


From the beginning when we had those 4096 shader + HBM leaks, it ddin't make any sense to me, as to why amd would have so much effort on bringing 200-300% more bandwidth when there is already enough in hawaii? Why make the chip only 45% faster and waste all that 200-300% of more bandwidth? It just doesn't make any sense, unless the chip has way more shaders than people are expecting on?



So my follow up question is: Since I'm not engineer and don't understand that much the technical shit about AA and such, but since AMD is bringing so much more bandwidth, is there possibility that it can be now utilize to make much more efficient AA? Kinda like what microsoft tried with the eDRAM on xbox360 and when people were saying that in theory they can bring free AA. But in the end the eDRAM was big enough or something like that.

 

[Leadbox]

4GB will kill them this round if only for the fact that 4k gaming is the way things are going. I'm running @ 4k right now with 3 OG Titans, I can't honestly see myself going back to 4GB cards.

Even if their new card is significantly faster than the 980Ti @ 4k with only 4GB I'm not sure I'd buy it, as it might not be future proof, and I'm not sure I'd be willing to wait until Aug (or later) to buy a card with more memory. I'm ready to buy now.

A lot of people bought the OG titan 6GB with those same future proof misconceptions

 

[Wag]

A lot of people bought the OG titan 6GB with those same future proof misconceptions

Maybe then, but now games are starting to use more than 4GB of VRAM, especially at 4k. I guess we'll see in a couple of weeks.

 

[Despoiler]

4GB will kill them this round if only for the fact that 4k gaming is the way things are going. I'm running @ 4k right now with 3 OG Titans, I can't honestly see myself going back to 4GB cards.

Even if their new card is significantly faster than the 980Ti @ 4k with only 4GB I'm not sure I'd buy it, as it might not be future proof, and I'm not sure I'd be willing to wait until Aug (or later) to buy a card with more memory. I'm ready to buy now.

It's premature to make a statement about memory requirements between two totally different memory technologies. The information we do have suggests that 4GB is more than enough for HBM.

http://arstechnica.co.uk/informatio...hbm-why-amds-high-bandwidth-memory-matters/2/

AMD's CTO, Joe Macri
If you actually look at frame buffers and how efficient they are and how efficient the drivers are at managing capacities across the resolutions, you'll find that there's a lot that can be done. We do not see 4GB as a limitation that would cause performance bottlenecks. We just need to do a better job managing the capacities. We were getting free capacity, because with [GDDR5] in order to get more bandwidth we needed to make the memory system wider, so the capacities were increasing. As engineers, we always focus on where the bottleneck is. If you're getting capacity, you don't put as much effort into better utilising that capacity. 4GB is more than sufficient. We've had to go do a little bit of investment in order to better utilise the frame buffer, but we're not really seeing a frame buffer capacity [problem]. You'll be blown away by how much [capacity] is wasted.

 

[maddie]

You are right, didn't realize that the shader part is only 60% of the die. Also didn't remember the fact that high density libraries with GF process brings even more space for more transistors. So if the die is 600mm^2, 4096 shaders sound quite small amount for that size of die. Unless there are ton of new compute shit that is not utilized on games or something like that... don't know??


From the beginning when we had those 4096 shader + HBM leaks, it ddin't make any sense to me, as to why amd would have so much effort on bringing 200-300% more bandwidth when there is already enough in hawaii? Why make the chip only 45% faster and waste all that 200-300% of more bandwidth? It just doesn't make any sense, unless the chip has way more shaders than people are expecting on?



So my follow up question is: Since I'm not engineer and don't understand that much the technical shit about AA and such, but since AMD is bringing so much more bandwidth, is there possibility that it can be now utilize to make much more efficient AA? Kinda like what microsoft tried with the eDRAM on xbox360 and when people were saying that in theory they can bring free AA. But in the end the eDRAM was big enough or something like that.

I am as lost here as you.

 

[maddie]

It's premature to make a statement about memory requirements between two totally different memory technologies. The information we do have suggests that 4GB is more than enough for HBM.

http://arstechnica.co.uk/informatio...hbm-why-amds-high-bandwidth-memory-matters/2/

This might be for gaming, but I'm fairly certain it won't apply to scientific/professional workloads.

If AMD is ignoring these HIGH value/margin markets, by having 4GB max with a new generation near monopoly in DP compute, then they are fools.

I don't believe they are fools.

 

[jamesgalb]

This might be for gaming, but I'm fairly certain it won't apply to scientific/professional workloads.

If AMD is ignoring these HIGH value/margin markets, by having 4GB max with a new generation near monopoly in DP compute, then they are fools.

I don't believe they are fools.

Don't worry, they will have a pretty damn powerful 8gb card at a fraction of the prices of nVidias 6gb and 12gb options.

 

[geoxile]

This might be for gaming, but I'm fairly certain it won't apply to scientific/professional workloads.

If AMD is ignoring these HIGH value/margin markets, by having 4GB max with a new generation near monopoly in DP compute, then they are fools.

I don't believe they are fools.

Then they'll be saving any high memory capacity chips strictly for the HPC market