R9 390X Achieves 8GB HBM with Dual Link Interposer not HBM2

[BOT Edward]

what do you think the titan X is...

12GB is prety "leading edge" imo, something AMD isnt doing is making the most powerful GPU in the world

Most powerful for what purpose though?

"most powerful" is a very vague description, for instance a cpu with more performance per core will do better with java than a 18 core cpu :lol:

[Thought]

i dont see why not there are 8GB 290x and people will get the 390x for the better double precision performance lol 

Because of the designs for HBM already and the low yields, they can't produce cards with 16 GB VRAM capacities.

[Prastupok]

This needs to get locked...

 

While interesting, it's not a gamechanger. I enjoy seeing the progression of technology as much as the next guy, but I'd like to see a next gen core vs next gen memory.

 

Why do you people insist on arguing though? I mean jesus...

[Thought]

This needs to get locked...

 

While interesting, it's not a gamechanger. I enjoy seeing the progression of technology as much as the next guy, but I'd like to see a next gen core vs next gen memory.

 

Why do you people insist on arguing though? I mean jesus...

I agree. Too many people like to debate semantics over non-issues.

[BiG StroOnZ]

 

This needs to get locked...

 

While interesting, it's not a gamechanger. I enjoy seeing the progression of technology as much as the next guy, but I'd like to see a next gen core vs next gen memory.

 

Why do you people insist on arguing though? I mean jesus...

 

I don't think it needs to get locked it just needs to be cleaned up and directed back on topic.

[Opcode]

go take this argument somewhere else where you can predict the future, like maybe the next dimension or something

It's called speculation, taking AMD's own benchmarks the R9 390X is roughly 12% faster than the TITAN X on DirectX 11. On DirectX 12 that gap will lengthen.

[Enderman]

Most powerful for what purpose though?

"most powerful" is a very vague description, for instance a cpu with more performance per core will do better with java than a 18 core cpu :lol:

for everything

the titan X is better than the 290x in everything (except maybe GPU mining, but thats not what graphics cards are for)

[BobbyG2]

for everything

the titan X is better than the 290x in everything (except maybe GPU mining, but thats not what graphics cards are for)

Double precision?

[Enderman]

Double precision?

?

what about it?

neither of these cards have double precision

[Meladath]

?

what about it?

neither of these cards have double precision

erm... what?

 

The titan X has 1/32 FP64 performance since maxwell cut it all to save power. Hawaii has 1/8th and even the R9 290X LAUGHS at the titan X in FP64 performance.

Titan X has a maximum theoretical double precision performance of 192 GFLOPS, while the R9 290X is 702. The R9 390X will ofc be higher, guessing at 4096 cores and still 1/8th FP64 performance, it should have 1024 GFLOPS. Exactly 1TFLOP of theoretical maximum FP64 performance. Just over 5 times higher than the titan X.

[Enderman]

erm... what?

 

The titan X has 1/32 FP64 performance since maxwell cut it all to save power. Hawaii has 1/8th and even the R9 290X LAUGHS at the titan X in FP64 performance.

Titan X has a maximum theoretical double precision performance of 192 GFLOPS, while the R9 290X is 702. The R9 390X will ofc be higher, guessing at 4096 cores and still 1/8th FP64 performance, it should have 1024 GFLOPS. Exactly 1TFLOP of theoretical maximum FP64 performance. Just over 5 times higher than the titan X.

the only GTX GPUs that supported double precision were the original titan and the titan black

they removed the double precision from the titan X

[Meladath]

the only GTX GPUs that supported double precision were the original titan and the titan black

they removed the double precision from the titan X

I don't think you quite understand how double precision floating point numbers work....

Most GPU's support  double precision for gpu computing, the ones that don't are very low end. Double precision (FP64) almost always has a ratio of its performance corresponding to its single precision (FP32) performance.

 

The original titan and titan black simply had more FP64 performance ratio-wise than the other GTX cards. Traditionally AMD cards have higher FP64 performance (this is also kinda a part of the reason AMD cards have been using more power the last few generations) with tahiti being 1/4th and hawaii being 1/8th. Compared to kepler with 1/24th, the original titan and black beat it with 1/3 FP64 performance. The titan X on the other hand, does not have more FP64 performance than any other maxwell GPU (nvidia cut it to save power) and has 1/32th.

 

Although I find it unlikely AMD will stick with 1/8th FP64 performance for Fiji. Whilst it would make it a beast of a compute card, its one area they could easily save power like nvidia did with maxwell and AMD really needs to save power with the 390X if it has 4096 shaders compared to the 2816 of the 290X..

[mr moose]

This needs to get locked...

 

While interesting, it's not a gamechanger. I enjoy seeing the progression of technology as much as the next guy, but I'd like to see a next gen core vs next gen memory.

 

Why do you people insist on arguing though? I mean jesus...

 

People need to prop up their alter egos.  For some people the internet is the only place they can pretend to be important or good at something and and least have one or two other people pat them on the back.   Anyone who has to make an assumption or falls for all the marketing hype probably hasn't got much else in their life they are in control of. 

 

Personally I can't wait to see how these cards turn out. The writing is on the wall for AMD so they really can't afford for this not to be a raging success. 

[Bonn]

I smell burning memory... 

Good luck AMD, Good Luck.

[Enderman]

I don't think you quite understand how double precision floating point numbers work....

Most GPU's support  double precision for gpu computing, the ones that don't are very low end. Double precision (FP64) almost always has a ratio of its performance corresponding to its single precision (FP32) performance.

 

The original titan and titan black simply had more FP64 performance ratio-wise than the other GTX cards. Traditionally AMD cards have higher FP64 performance (this is also kinda a part of the reason AMD cards have been using more power the last few generations) with tahiti being 1/4th and hawaii being 1/8th. Compared to kepler with 1/24th, the original titan and black beat it with 1/3 FP64 performance. The titan X on the other hand, does not have more FP64 performance than any other maxwell GPU (nvidia cut it to save power) and has 1/32th.

 

Although I find it unlikely AMD will stick with 1/8th FP64 performance for Fiji. Whilst it would make it a beast of a compute card, its one area they could easily save power like nvidia did with maxwell and AMD really needs to save power with the 390X if it has 4096 shaders compared to the 2816 of the 290X..

http://www.anandtech.com/show/9059/the-nvidia-geforce-gtx-titan-x-review/2

[Meladath]

http://www.anandtech.com/show/9059/the-nvidia-geforce-gtx-titan-x-review/2

 

Linking an article you haven't read when you don't understand the topic doesn't really back up what you are saying btw.

That article says NVIDIA stripped out the FP64 support from maxwell completely from the die. That's also what I said.

That doesn't somehow mean that ONLY the original titan and titan black supported FP64.

[BOT Edward]

for everything

the titan X is better than the 290x in everything (except maybe GPU mining, but thats not what graphics cards are for)

 

 

The bold may be true but for some people that's why they buy them.

Something is only better than something else in specific comparisons mostly.

 

Like AMD make a better integrated graphics for x64 gaming (because to my knowledge, the only integrated gpus nVidia make are their Tegra processors? Although I may be completely wrong).

[PortentousLad]

  "2,5D" hahahhaha

 

anywho möre VRAM möre bedda!

[Zerreth]

This isn't news, PC Perspective reported on this MONTHS ago, I remember them talking about it in their weekly podccast.

http://www.pcper.com/news/Graphics-Cards/LinkedIn-Posts-Hint-Radeon-R9-380X-Features-Stacked-Memory

[Fetzie]

I don't think you quite understand how double precision floating point numbers work....

Most GPU's support  double precision for gpu computing, the ones that don't are very low end. Double precision (FP64) almost always has a ratio of its performance corresponding to its single precision (FP32) performance.

 

The original titan and titan black simply had more FP64 performance ratio-wise than the other GTX cards. Traditionally AMD cards have higher FP64 performance (this is also kinda a part of the reason AMD cards have been using more power the last few generations) with tahiti being 1/4th and hawaii being 1/8th. Compared to kepler with 1/24th, the original titan and black beat it with 1/3 FP64 performance. The titan X on the other hand, does not have more FP64 performance than any other maxwell GPU (nvidia cut it to save power) and has 1/32th.

 

Although I find it unlikely AMD will stick with 1/8th FP64 performance for Fiji. Whilst it would make it a beast of a compute card, its one area they could easily save power like nvidia did with maxwell and AMD really needs to save power with the 390X if it has 4096 shaders compared to the 2816 of the 290X..

 

Don't the 290(x) and Carrizo have 1/2 FP64?

[7850OC]

i have a 1gb card...

 

im sad

is so sad bro  :blush:

 

Im running a 256mb card, stop complaining

yea... at this time, you can just jump from a bridge man lol 

 

- - - - 

 

on topic

 

with a 4096 bit bus, i... really stop worry about bottlenecks (with any number of vram) 

 

- - - - 

 

i have a 256 bit bus (with my 7850 2GB) and i already feel like the king of the universe  :lol:

[Fetzie]

Do we have any data comparing how long it takes to read data from HBM compared to GDDR5?

[Glenwing]

Don't the 290(x) and Carrizo have 1/2 FP64?

 

It's artificially limited to 1/8 on Radeon cards.

[That Norwegian Guy]

are you seriously talking about a 2GB difference between 6GB and 8GB of vram as "leading edge" to "take back the market" ???

LOLOLOL

you've got to be kidding me

6GB is already enough for 6k at ultra

 

Nope, games such as Dying Light and Star Citizen are already capable of saturating a 6GB buffer as low as 4K resolution, given max settings and view distance, and in the case of Star Citizen the VRAM usage will only get worse.

 

Then there are games such as The Witcher 3 and The Division, which according to expected texture and shadow map resolutions look like they will use just as much if not more.

[Doobeedoo]

Well games can saturate 4GB of VRAM on 1080p today. Some being badly optimized and some taking advantage of it and pushing the limitations.