• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD VEGA confirmed for 2017 H1

Status
Not open for further replies.
N19h7m4r3 said:
I never once stated the Quadro cards used different cores?

The way I read that he's saying just the gaming cards have gimped compute, when in fact also the Quadro cards do.
The Tesla 100 is the first NVIDIA card since Kepler to feature full Compute performance and FP64, as even the Quadro's don't have it.

Where as the original thread of ours was about AMD's older cards from Hawaii and prior never had that issue; the top gaming cards had the same compute as top FirePro.

So I surmised that it could be the reason Vega is so darn big considering it's core count; which could mean that they went back to that design of huge chip ( 500-540mm2 ), and using it for top consumer as well.

Instead of like NVIDIA which has separated those lines now, DP and FP64 only available on G/P100 610mm2, and everything in the Consumer and Workstation line "gimped"
Click to expand...

All of AMD cards are also compute cards, the 7970, the 290X, the Fury-X the RX 480, so the reason Vega is so big on 14nm compared with Fury-X... or any other AMD card you may cite is not because Vega may also be a workstation card and [Insert name of AMD card here] was/is not.... is flawed. THEY ALL ARE! :p
 
humbug said:
All of AMD cards are also compute cards, the 7970, the 290X, the Fury-X the RX 480, so the reason Vega is so big on 14nm compared with Fury-X... or any other AMD card you may cite is not because Vega may also be a workstation card and [Insert name of AMD card here] was/is not.... is flawed. THEY ALL ARE! :p
Click to expand...

Fiji and Polaris have significantly reduced compute capabilities compared to Hawaii and prior cards.

Look at how great Hawaii does compared to everything else in DP. AMD is still great in single precision and OpenCL, but since Hawaii they've been cutting FP64 performance.

As Anandtech state, even the 7970 is beating the Fury X in those kinds of workloads.

452e42a662134108ac51f97553d0a6dc.png


4c1156bcecdb4768a87d3c7405cbab9f.png
 
N19h7m4r3 said:
Fiji and Polaris have significantly reduced compute capabilities compared to Hawaii and prior cards.

Look at how great Hawaii does compared to everything else in DP. AMD is still great in single precision and OpenCL, but since Hawaii they've been cutting FP64 performance.

As Anandtech state, even the 7970 is beating the Fury X in those kinds of workloads.

452e42a662134108ac51f97553d0a6dc.png


4c1156bcecdb4768a87d3c7405cbab9f.png
Click to expand...


They are still the same silicon tho, they have just had the compute performance arterially gimped.
 
Would it be a fair bet to say that Computex is looking like a paper launch and availability sometime in June? or could it be a H1 paper launch and H2 before consumers can get their hands on Vega.
 
humbug said:
They are still the same silicon tho, they have just had the compute performance arterially gimped.
Click to expand...

Aye, which is my point. Unlike NVIDIA AMD doesn't have the resources to have so many different cores to use in Consumer, Workstation, and Server.

With Fiji and Polaris though, we've not seen any professional cards that have all the compute functionality enabled.

Even AMD's RadeonPro's and Instinct cards based on those cores don't have the features.
Only Single and Half Precision, as the MI8 is based on the Nano, and MI6 is based on Polaris.

http://www.anandtech.com/show/10905/amd-announces-radeon-instinct-deep-learning-2017/2

So Top end Vega which is a darn big core should have all the goodies for enterprise hence, the size. Then they'll either just also sell it to consumers like with Hawaii and Prior, or they'll "cut" Double Precision and everything.
Although that'll still leave the die with some unused space, that's just big.
So either VEGA is a monster in gaming, or it's primarily a Deep Learning and Enterprise card that'll have section cut off and then sold for top and gaming as well.

What we do know though is that Instinct MI25 is based on Vega is apparently beating the 2016 Titan X rather well in current Deep Learning benchmarks.

117c0d0806a448d1b2b2c7c409e133d3.png
 
humbug said:
All of AMD cards are also compute cards, the 7970, the 290X, the Fury-X the RX 480, so the reason Vega is so big on 14nm compared with Fury-X... or any other AMD card you may cite is not because Vega may also be a workstation card and [Insert name of AMD card here] was/is not.... is flawed. THEY ALL ARE! :p
Click to expand...


AMD stripped the FP64 hardware right out of the Fiji and Polaris cards, form 1:2 down to 1:16.
 
humbug said:
They are still the same silicon tho, they have just had the compute performance arterially gimped.
Click to expand...


I don't know how many times I have had to correct you about this over the years , but no, AMD took out the hardware that enabled GP64 at the same performance as Hawaii. Those transistors are now gone There is no driver hack or anything else, the hardware has been removed. AMD has stated that publicly time and time again, which is why the Radeon pro Fiji still only offer 1:16 DP performance even for professional computer.


Both Nvidia and AMD have reduced FP64 in order to save transistors in the consumer card. That is a hard fact.
 
N19h7m4r3 said:
Aye, which is my point. Unlike NVIDIA AMD doesn't have the resources to have so many different cores to use in Consumer, Workstation, and Server.

With Fiji and Polaris though, we've not seen any professional cards that have all the compute functionality enabled.

Even AMD's RadeonPro's and Instinct cards based on those cores don't have the features.
Only Single and Half Precision, as the MI8 is based on the Nano, and MI6 is based on Polaris.

http://www.anandtech.com/show/10905/amd-announces-radeon-instinct-deep-learning-2017/2

So Top end Vega which is a darn big core should have all the goodies for enterprise hence, the size. Then they'll either just also sell it to consumers like with Hawaii and Prior, or they'll "cut" Double Precision and everything.
Although that'll still leave the die with some unused space, that's just big.
So either VEGA is a monster in gaming, or it's primarily a Deep Learning and Enterprise card that'll have section cut off and then sold for top and gaming as well.

What we do know though is that Instinct MI25 is based on Vega is apparently beating the 2016 Titan X rather well in current Deep Learning benchmarks.
Click to expand...


Ironically Deep learning is largely constrained by half-precision FP16 performance (and in some cases even Int8), FP64 would never be used for deep learning. that is why a big performance gain for Vega is enabling double speed half-precision FP performance. Double precision is important in many other applications of GP-GPU/HPC.

I have a feeling that in the future Nvidia will actually separate out a new ASIC for deep-learning that focuses on FP16, int8 and FP32 performance without the FP64 hardware. Strip away more of the graphics functionality, and concentrate on pure efficiency. Google's deep-learning accelerator and Intel's new FPGA offerings are a big threat to nvidia.
 
Could it be why they've released their 400s again, as it does seem odd, putting them back out again, when Vega is supposedly only a couple of weeks away.

Could be why they just upped the number to a 5, to tide them over, and as everyone says, these are the cards where the sales are, and the market is. :p
 
D.P. said:
I don't know how many times I have had to correct you about this over the years , but no, AMD took out the hardware that enabled GP64 at the same performance as Hawaii. Those transistors are now gone There is no driver hack or anything else, the hardware has been removed. AMD has stated that publicly time and time again, which is why the Radeon pro Fiji still only offer 1:16 DP performance even for professional computer.


Both Nvidia and AMD have reduced FP64 in order to save transistors in the consumer card. That is a hard fact.
Click to expand...

Because you're always wrong, have you ever thought about that?
They are still there, just disabled, as i said they are the same silicone, they don't have two separate production lines for gaming and workstation cards, the gaming cards just have some feature disabled or gimped.
 
Last edited:
Status
Not open for further replies.
Back
Top Bottom