Rapid Packed Math fp16 to be used in FM Serra, Wolfenstein 2 and Far Cry 5

#5458220

Kohlendioxidus

2017-08-02 09:33

coool but where can I buy a GPU like this? When is the NDA lifted as I see no reviews yet?:3eyes:

#5458225

BUDA20

2017-08-02 09:49

how does this compare to NVIDIA Multi-Res Shading in Shadow Warrior 2? :question: I remember using the middle settings on a mid-end graphics card and the result was pretty good

#5458226

RooiKreef

2017-08-02 09:52

From what I have seen in the last few years, if Nvidia is not going to jump on the band wagon of fp16 and support it on a hardware level, then you will only see a small hand full of games with this feature and then it will disappear in the dust. Same like all the other unique features that AMD has created in the last couple of years. Features like AMD true audio, mantle and the list go on... The reality is no support from Nvidia means no future support in games...

#5458231

no_1_dave

2017-08-02 10:03

So AMD need to sacrifice image quality now to keep up with the competition... Then they're selling it as a "feature". Interesting

#5458232

arakisb

2017-08-02 10:06

From what I have seen in the last few years, if Nvidia is not going to jump on the band wagon of fp16 and support it on a hardware level, then you will only see a small hand full of games with this feature and then it will disappear in the dust. Same like all the other unique features that AMD has created in the last couple of years. Features like AMD true audio, mantle and the list go on... The reality is no support from Nvidia means no future support in games...

that sucks

#5458233

Exascale

2017-08-02 10:11

So AMD need to sacrifice image quality now to keep up with the competition... Then they're selling it as a "feature". Interesting

Where are the image quality comparisons youre looking at to make that conclusion?

#5458234

fantaskarsef

2017-08-02 10:13

I guess it will be interesting to see if that new Futuremark scene Serra will show a major gap between AMD cards and Nvidia cards.

#5458236

Exascale

2017-08-02 10:17

From what I have seen in the last few years, if Nvidia is not going to jump on the band wagon of fp16 and support it on a hardware level, then you will only see a small hand full of games with this feature and then it will disappear in the dust. Same like all the other unique features that AMD has created in the last couple of years. Features like AMD true audio, mantle and the list go on... The reality is no support from Nvidia means no future support in games...

The AMD of today is no longer the AMD of the last few years. Epyc, Threadripper and Ryzen are proof of that. The fact that Vega has built in Infinity Fabric, NVMe, 49bit virtual addressing and HBCC also means that AMD is getting serious about AI and HPC. If people adopt Vega and its successors like they are Ryzen then developers will have an incentive to support it. I mean, AMD did come out with that whole x86-64 thing too.

#5458243

no_1_dave

2017-08-02 10:56

Where are the image quality comparisons youre looking at to make that conclusion?

I read the article

#5458247

Ryu5uzaku

2017-08-02 11:06

From what I have seen in the last few years, if Nvidia is not going to jump on the band wagon of fp16 and support it on a hardware level, then you will only see a small hand full of games with this feature and then it will disappear in the dust. Same like all the other unique features that AMD has created in the last couple of years. Features like AMD true audio, mantle and the list go on... The reality is no support from Nvidia means no future support in games...

Considering consoles apart from Xbox One X support fp16 it should be fine. Nvidia sure is limiting their proper fp16 currently to their Tesla products but that could change.

#5458252

Angel_Rex

2017-08-02 11:22

Ram? Didn't they use c14 3200mhz ram to test? Or did I miss read?

#5458258

user1

2017-08-02 11:31

So AMD need to sacrifice image quality now to keep up with the competition... Then they're selling it as a "feature". Interesting

Alot of things do not require 32bit precision, infact doing them in 32bit precision is a waste of energy, and now you not only save power by using half (fp16) precision , you get double the performance. There is a reason the GP100 and GV100 nvidia chips have this same feature (double rate fp16)

#5458269

edilsonj

2017-08-02 12:07

So AMD need to sacrifice image quality now to keep up with the competition... Then they're selling it as a "feature". Interesting

I can't get the point here. :stewpid: We can use FP16 in minor features that not require FP32. It's so simple to think.

#5458283

chronek

2017-08-02 12:55

So AMD need to sacrifice image quality now to keep up with the competition... Then they're selling it as a "feature".

NVidia have FP16 faster that FP32 too, so did not get your point, cause NVidia will benefit it too, but most will benefit games. It is just new technology (not gameworks), it is like mipmaping was one day.

#5458287

-Tj-

2017-08-02 13:03

https://s2.postimg.org/63fxp1d6x/111.jpg https://www.overclock3d.net/news/gpu_displays/amd_rx_vega_-_what_is_rapid_packed_math/1

#5458291

no_1_dave

2017-08-02 13:10

Alot of things do not require 32bit precision, infact doing them in 32bit precision is a waste of energy, and now you not only save power by using half (fp16) precision , you get double the performance. There is a reason the GP100 and GV100 nvidia chips have this same feature (double rate fp16)

I can't get the point here. :stewpid: We can use FP16 in minor features that not require FP32. It's so simple to think.

NVidia have FP16 faster that FP32 too, so did not get your point, cause NVidia will benefit it too, but most will benefit games. It is just new technology (not gameworks), it is like mipmaping was one day.

Nvidia reduced image quality back in the day and it didn't go down well. AMD have tried it before in some of their drivers and had a lot of complaints. Time will tell if selling it as a feature works out any differently

#5458292

Denial

2017-08-02 13:10

NVidia have FP16 faster that FP32 too, so did not get your point, cause NVidia will benefit it too, but most will benefit games. It is just new technology (not gameworks), it is like mipmaping was one day.

The only Nvidia card that has full packed math is GP100. GP104/102/106 have 1/32 support for debugging purposes only. This might change a bit with Volta but if the speculation about Vega's power consumption being related to packed math is true, Nvidia may forgo it entirely on it's gaming cards.

#5458296

Ryu5uzaku

2017-08-02 13:14

NVidia have FP16 faster that FP32 too, so did not get your point, cause NVidia will benefit it too, but most will benefit games. It is just new technology (not gameworks), it is like mipmaping was one day.

They do but not in their consumer cards at the moment iirc.

#5458298

chronek

2017-08-02 13:20

They do but not in their consumer cards at the moment iirc.

So blame NVidia to use old technology for consumers (even if they have new one)

#5458301

Loophole35

2017-08-02 13:26

Those 25% numbers look good but it's only a few instances that this is used. I wonder what the improvements will be overall? If it's only a 5-10% improvement over using just sp32 for everything the power increases from having hp16 active then it is useless.