AMD announces 32GB Fire Pro W9100

[Prysin]

14 minutes ago, patrickjp93 said:

HMC or 3DXPoint eventually.

3DXpoint is much more likely. It has a much broader appeal.

[patrickjp93]

20 minutes ago, AlexGoesHigh said:

Wait, isn't 3DXPoint flash for SSD when did Intel/Micron said it was volatile memory? Or am i missing something. 

It's non-volatile memory, but it can be used in DIMMs (Intel is already selling it). So clearly it could be used for GPU purposes too.

[Prysin]

1 hour ago, patrickjp93 said:

It's non-volatile memory, but it can be used in DIMMs (Intel is already selling it). So clearly it could be used for GPU purposes too.

intel is already selling 3DXpoint. Despite it not being in production yet?

It was like 2-3 weeks since some early pre-production test samples came out.

 

https://en.wikipedia.org/wiki/3D_XPoint

Quote

Production

Small quantities of 128 Gbit chips were made initially in 2015 at a wafer fab in Lehi, Utah, operated by IM Flash Technologies LLC, an Intel-Micron joint venture. They stack two 64 Gbit planes.^[2][17] In early 2016 Guy Blalcok, CEO of IM Flash stated that the mass production of the chips is still 12 to 18 months away.^[18]

The anticipated price per bit was expected to be higher than NAND and lower than DRAM, though dependant on the final product.^[19] In early 2016, IM Flash announced that the first generation of solid state drives will achieve 95000 IOPS throughput with 9 microsecond latency.^[18]

Intel announced the Optane brand for storage products based on the technology in mid 2015.^[20]

 

[Godlygamer23]

On 4/15/2016 at 10:28 AM, ShadowCaptain said:

Eventually they will put a CPU on it and the GPU will become a PC

wait, this is circular  I guess what we want is faster RAM that is accessible by both teh CPU and GPU, so that a GPU no longer needs VRAM but can use like a 200GB super fast RAM that is shared

I would still prefer a GPU having bandwidth all to itself without some other component sapping bandwidth from it. With faster memory will come faster GPUs to utilize more of it. If you have other components that are becoming faster and faster, you might get a bandwidth bottleneck that will hold back any components connected to that single pool. 

[Prysin]

6 minutes ago, Godlygamer23 said:

I would still prefer a GPU having bandwidth all to itself without some other component sapping bandwidth from it. With faster memory will come faster GPUs to utilize more of it. If you have other components that are becoming faster and faster, you might get a bandwidth bottleneck that will hold back any components connected to that single pool. 

with how things are looking up with 3DXpoint coming within a year or two for system RAM, the issue will be chipset bandwidth.

Soon, the choice of motherboard will make a huge difference as to whether or not you can run parts efficiently.

[Dabombinable]

4 minutes ago, Prysin said:

with how things are looking up with 3DXpoint coming within a year or two for system RAM, the issue will be chipset bandwidth.

Soon, the choice of motherboard will make a huge difference as to whether or not you can run parts efficiently.

Just like the days of the VIA chipsets.

[patrickjp93]

5 hours ago, Prysin said:

intel is already selling 3DXpoint. Despite it not being in production yet?

It was like 2-3 weeks since some early pre-production test samples came out.

 

https://en.wikipedia.org/wiki/3D_XPoint

 

They had demo units out at GTC and SC 15. Keep up.

http://www.legitreviews.com/intel-shows-off-512gb-optane-drive-with-3d-xpoint-memory-that-fits-in-ddr4-slot_176826

[Prysin]

5 hours ago, patrickjp93 said:

They had demo units out at GTC and SC 15. Keep up.

http://www.legitreviews.com/intel-shows-off-512gb-optane-drive-with-3d-xpoint-memory-that-fits-in-ddr4-slot_176826

i know they had DEMO units, but not PRODUCTION units. Which is what i said.

If you are going to argue with me, read my fucking post.

And the post i saw about 3DXpoint demo units was a repost. That is my own fault for not double checking the date.

[Notional]

11 hours ago, Godlygamer23 said:

I would still prefer a GPU having bandwidth all to itself without some other component sapping bandwidth from it. With faster memory will come faster GPUs to utilize more of it. If you have other components that are becoming faster and faster, you might get a bandwidth bottleneck that will hold back any components connected to that single pool. 

Well it worked out pretty well on the PS4. Right now there is quite a lot of redundancy when it comes to gaming, with assets being on both SSD/HDD, system RAM and video RAM. Once we have non volatile memory, we could replace all 3/4 with just one module. Until then, combining system and video memory might not be all that bad, but it would definitely hurt customizability when building a system.

[DXMember]

Meh.. still a Hawaii with more RAM

[Prysin]

5 minutes ago, DXMember said:

Meh.. still a Hawaii with more RAM

still more RAM then Nvidia offers. Still more bandwidth then Nvidia offers, Still cheaper then what Nvidia offers and still as powerful or more powerful GPU then what Nvidia offer.

I see no issues here. Other then AMD not having CUDA support out just yet.

[DXMember]

Just now, Prysin said:

still more RAM then Nvidia offers. Still more bandwidth then Nvidia offers, Still cheaper then what Nvidia offers and still as powerful or more powerful GPU then what Nvidia offer.

I see no issues here. Other then AMD not having CUDA support out just yet.

Yeah but that was also true before their released/announced W9100

nVidia hasn't had a compute a card in like 3 or 4 years...

I agree it's cool and all, but it's just not exciting

[Prysin]

6 minutes ago, DXMember said:

Yeah but that was also true before their released/announced W9100

nVidia hasn't had a compute a card in like 3 or 4 years...

I agree it's cool and all, but it's just not exciting

Nvidia recently released a 24GB M6000. This is simply an answer to that.

[DXMember]

25 minutes ago, Prysin said:

Nvidia recently released a 24GB M6000. This is simply an answer to that.

M6000 is not a compute card - M6000 features just 0.2 TFlops of FP64 compute performance
1/32 DP performance is laughable

[Notional]

7 minutes ago, DXMember said:

M6000 is not a compute card - M6000 features just 0.2 TFlops of FP64 compute performance
1/32 DP performance is laughable

Still a compute card for FP32.

 

What else should it be used for?

[Prysin]

26 minutes ago, DXMember said:

M6000 is not a compute card - M6000 features just 0.2 TFlops of FP64 compute performance
1/32 DP performance is laughable

The M6000 is for machine learning and AI. Stuff like that uses mixed precision (FP 16 and FP 32) rather then double precision.

[patrickjp93]

6 hours ago, Prysin said:

i know they had DEMO units, but not PRODUCTION units. Which is what i said.

If you are going to argue with me, read my fucking post.

And the post i saw about 3DXpoint demo units was a repost. That is my own fault for not double checking the date.

They're already selling NVDIMMs. Optane SSDs come later.

[patrickjp93]

4 hours ago, Prysin said:

still more RAM then Nvidia offers. Still more bandwidth then Nvidia offers, Still cheaper then what Nvidia offers and still as powerful or more powerful GPU then what Nvidia offer.

I see no issues here. Other then AMD not having CUDA support out just yet.

More RAM, but less bandwidth. The bus is larger, but Nvidia's memory clockspeed makes up for it. And actually it's just as expensive. Linpack benchmarks disagree.

[Sauron]

On 15/4/2016 at 4:15 PM, Fourmi Kill3r said:

Who will buy this knowing that we can't even use 12GB on a Titan X (gamer talking)

This isn't targeted at video games...

[TonyTheNain]

2 hours ago, Sauron said:

This isn't targeted at video games...

I understand now.. after all the quotes...

[Sauron]

53 minutes ago, Fourmi Kill3r said:

I understand now.. after all the quotes...

it might be a good idea to edit your original comment then ^^ have it say something like "-edit- I get it now"