[wccftech.com] Vega 20 GPU Estimated To Feature 20 TFLOPs of Compute - Page 3 - Overclock.net - An Overclocking Community

Forum Jump: 

[wccftech.com] Vega 20 GPU Estimated To Feature 20 TFLOPs of Compute

Reply
 
Thread Tools
post #21 of 151 (permalink) Old 07-14-2018, 06:59 PM
New to Overclock.net
guttheslayer is an unknown quantity at this point guttheslayer is an unknown quantity at this point
 
guttheslayer's Avatar
 
Join Date: Apr 2015
Posts: 3,525
Quote: Originally Posted by tajoh111 View Post
This is pure wccftech clickbait and the source is too.

The thing that AMD has given us so far for performance for vega 7nm is 2x performance per watt increase or 1.35x increase in performance. Not both according to wizzard who asked AMD. As a result, the later is likely the max and with some overclocking.

I would say the actual performance of vega 7nm is somewhere around 15 tflops to 16 tflops at stock settings considering real vega performance is around 1.5ghz or 12tflops when running at stock.

What wccftech forgets is that to get higher frequency, you need to add to the pipeline to accommodate the extra frequency. This is why the gtx 1080 has less cores than the gtx 980 ti but slightly more transistors even those the later has a 384bit memory controller.

They frequency jump is not free with the performance node increase and the pipeline has to be extended which enlarges each core and offsets some of the frequency jump. In addition, this is a professional card. What this means is large memory caches which don't shrink well with the process. Add in the double precision and potentially tensor cores, the 360mm2 die size and we are likely to see a modest increase in core count if any at all. This is very early 7nm on a complex node that was meant for low power applications like cell phone SOC. As a result, I suspect AMD had to be conservative with the design for now meaning the transistor density is not maxed out.
2x performance per watt and 1.35X increase in performance are not mutually exclusive. It can mean both such that with 1.35X performance increase, there is a 33% decrease in power consumption against Vega 64.


What does not tally is the die size vs. the fact given by AMD. The die size is not small as you can see from Lisa Su picture, and 360mm^2 estimation is not pull from the air (the finger size and the HBM module size can be used to infer). I am incline to believe the 2.8X more dense is accurate by GF. Which mean the chip that Lisa Su was holding contain at least twice as much transistor. (2.08X to be exact, so the extra 4% transistor is for the pipeline?)


Last edited by guttheslayer; 07-14-2018 at 07:03 PM.
guttheslayer is offline  
Sponsored Links
Advertisement
 
post #22 of 151 (permalink) Old 07-14-2018, 09:09 PM
New to Overclock.net
tajoh111 is an unknown quantity at this point tajoh111 is an unknown quantity at this point tajoh111 is an unknown quantity at this point
 
tajoh111's Avatar
 
Join Date: Aug 2013
Posts: 1,341
Quote: Originally Posted by guttheslayer View Post
2x performance per watt and 1.35X increase in performance are not mutually exclusive. It can mean both such that with 1.35X performance increase, there is a 33% decrease in power consumption against Vega 64.


What does not tally is the die size vs. the fact given by AMD. The die size is not small as you can see from Lisa Su picture, and 360mm^2 estimation is not pull from the air (the finger size and the HBM module size can be used to infer). I am incline to believe the 2.8X more dense is accurate by GF. Which mean the chip that Lisa Su was holding contain at least twice as much transistor. (2.08X to be exact, so the extra 4% transistor is for the pipeline?)

https://www.techpowerup.com/forums/t.../#post-3851821

In this case it is. You can't have both.

A 35% increase in performance while using 33% less power would mean 2x increase in performance per watt. E.g 1.35/0.66 = 2. So that would not meet what AMD has stated. As per wizzards post, it is either but not both.

As a result, 35% is likely what you get when you max out the performance while 2x performance per watt is what happens when you max out efficiency. Adding back the double precision, bumping up the clocks is going each into the efficiency.

And you have to take these figures with skepticism as AMD promised performance gains and performance per watt gains, have missed the mark severely after hawaii.

I think we will get something like a 1.2 to 1.25% increase in performance while consuming about 225watts(down from 300) which is about a 1.66x increase in performance per watt. A balance between performance and efficiency. The last 10% of performance from AMD has been very painful in terms of efficiency for their videocards as of late, Fiji, Polaris and Vega all share this.

AMD won't be able to undervolt these cards very much to squeeze out the efficiency because these are professional products, potentially going into super computers. Reliability will have to trump efficiency and as a result, AMD is likely to give these cards a healthy dose of voltage.
tajoh111 is offline  
post #23 of 151 (permalink) Old 07-14-2018, 11:54 PM
New to Overclock.net
guttheslayer is an unknown quantity at this point guttheslayer is an unknown quantity at this point
 
guttheslayer's Avatar
 
Join Date: Apr 2015
Posts: 3,525
Quote: Originally Posted by tajoh111 View Post
https://www.techpowerup.com/forums/t.../#post-3851821

In this case it is. You can't have both.

A 35% increase in performance while using 33% less power would mean 2x increase in performance per watt. E.g 1.35/0.66 = 2. So that would not meet what AMD has stated. As per wizzards post, it is either but not both.

As a result, 35% is likely what you get when you max out the performance while 2x performance per watt is what happens when you max out efficiency. Adding back the double precision, bumping up the clocks is going each into the efficiency.

And you have to take these figures with skepticism as AMD promised performance gains and performance per watt gains, have missed the mark severely after hawaii.

I think we will get something like a 1.2 to 1.25% increase in performance while consuming about 225watts(down from 300) which is about a 1.66x increase in performance per watt. A balance between performance and efficiency. The last 10% of performance from AMD has been very painful in terms of efficiency for their videocards as of late, Fiji, Polaris and Vega all share this.

AMD won't be able to undervolt these cards very much to squeeze out the efficiency because these are professional products, potentially going into super computers. Reliability will have to trump efficiency and as a result, AMD is likely to give these cards a healthy dose of voltage.

You do not understand the term performance per watt. It is the performance given for the fixed amount of wattage. That 2X PPW is the best case senarios I agree, but I can assure you the 360mm^2 chip is not fake, and using 7nm mean there is so much transistors that is pack inside. You cant lie with Physics.

I wont find it out of any impossibility realm for Vega 20 to comes pack with 128 CUs. But the functional one could be lower like 112 or even 96 CUs (due to yield issue).

guttheslayer is offline  
Sponsored Links
Advertisement
 
post #24 of 151 (permalink) Old 07-15-2018, 04:05 AM
New to Overclock.net
Pro3ootector Level 1
 
Pro3ootector's Avatar
 
Join Date: Aug 2013
Location: Poland / Deutschland
Posts: 1,059
Isn't GCN limited to 4096sp?

35% increse in performance is easy achievable thanks to clock speed and wider memory bus.
The great thing about it is that since they made first 7nm GPU it means there is demand for it, and that means higher revenues = more R&D money.

And honestly who would thought thad AMD will be first with 7nm. Is there any other chip made in this tech? I mean some ARM low power chip or something??

Last edited by Pro3ootector; 07-15-2018 at 04:17 AM.
Pro3ootector is offline  
post #25 of 151 (permalink) Old 07-15-2018, 04:24 AM
New to Overclock.net
guttheslayer is an unknown quantity at this point guttheslayer is an unknown quantity at this point
 
guttheslayer's Avatar
 
Join Date: Apr 2015
Posts: 3,525
Quote: Originally Posted by Pro3ootector View Post
Isn't GCN limited to 4096sp?

35% increse in performance is easy achievable thanks to clock speed and wider memory bus.
The great thing about it is that since they made first 7nm GPU it means there is demand for it, and that means higher revenues = more R&D money.

And honestly who would thought thad AMD will be first with 7nm. Is there any other chip made in this tech? I mean some ARM low power chip or something??
As of now Titan V alone is 70% ahead of Vega 64, I dont think just 35% bump is doing AMD any flavour, especially if you consider the design cost of a 7nm chip is $271 million USD.


And I don't think AMD GCN is limited to 4096. They have 4 stack HBM now which means it could be double to 4096 x2 SPs.

guttheslayer is offline  
post #26 of 151 (permalink) Old 07-15-2018, 04:43 AM
New to Overclock.net
Pro3ootector Level 1
 
Pro3ootector's Avatar
 
Join Date: Aug 2013
Location: Poland / Deutschland
Posts: 1,059
Quote: Originally Posted by guttheslayer View Post
As of now Titan V alone is 70% ahead of Vega 64, I dont think just 35% bump is doing AMD any flavour, especially if you consider the design cost of a 7nm chip is $271 million USD.


And I don't think AMD GCN is limited to 4096. They have 4 stack HBM now which means it could be double to 4096 x2 SPs.
Depends on a workload. In workstation aplications it's comparable to Titan XP and Quadro P6000 ( FE that is ).
Pro3ootector is offline  
post #27 of 151 (permalink) Old 07-15-2018, 05:01 AM
New to Overclock.net
guttheslayer is an unknown quantity at this point guttheslayer is an unknown quantity at this point
 
guttheslayer's Avatar
 
Join Date: Apr 2015
Posts: 3,525
Quote: Originally Posted by Pro3ootector View Post
Depends on a workload. In workstation aplications it's comparable to Titan XP and Quadro P6000 ( FE that is ).
In term of compute power I agree they are not that far behind, but using a 7nm with extreme density advantage is a waste if AMD never take full advantage of it. Again design a 7nm chip is not cheap, and it is also only matter of time AMD will lose that advantage when NV jump ship to 7nm as well.

guttheslayer is offline  
post #28 of 151 (permalink) Old 07-15-2018, 05:17 AM
New to Overclock.net
Pro3ootector Level 1
 
Pro3ootector's Avatar
 
Join Date: Aug 2013
Location: Poland / Deutschland
Posts: 1,059
Quote: Originally Posted by guttheslayer View Post
In term of compute power I agree they are not that far behind, but using a 7nm with extreme density advantage is a waste if AMD never take full advantage of it. Again design a 7nm chip is not cheap, and it is also only matter of time AMD will lose that advantage when NV jump ship to 7nm as well.
Nvidia bigest advantage was higher clock and better efficency. AMD has send more engineering to power optimize server chips cause it was more important to them at the time, I can't find this article at the moment.
Will the same thing happen to Instinct/Pro chip? I doubt it.

edit: https://www.fudzilla.com/news/graphi...power-was-high

And thinking about it if Vega had very good pp/W, miners would simply grab every single card.

Last edited by Pro3ootector; 07-15-2018 at 05:21 AM.
Pro3ootector is offline  
post #29 of 151 (permalink) Old 07-15-2018, 06:16 AM
New to Overclock.net
tajoh111 is an unknown quantity at this point tajoh111 is an unknown quantity at this point tajoh111 is an unknown quantity at this point
 
tajoh111's Avatar
 
Join Date: Aug 2013
Posts: 1,341
Quote: Originally Posted by guttheslayer View Post
You do not understand the term performance per watt. It is the performance given for the fixed amount of wattage. That 2X PPW is the best case senarios I agree, but I can assure you the 360mm^2 chip is not fake, and using 7nm mean there is so much transistors that is pack inside. You cant lie with Physics.

I wont find it out of any impossibility realm for Vega 20 to comes pack with 128 CUs. But the functional one could be lower like 112 or even 96 CUs (due to yield issue).
I know what performance per watt is.

What you don't realize it is a scalable term.

2x the performance while using the same wattage is 2x performance per watt while similarly, having 33.33%extra performance while using 66.6% of the power is also 2x performance per watt. This is because mathematically they are the same.

Say a GPU gets 100 fps with first gen technology, using 100 watts. Lets say 2nd generation, they get 2x performance per watt. This could mean 200fps for 100watts or 133fps for 66.6 watts and they would be mathematically identical in terms of performance per watt.

Where the first gen of the technology was 100fps/100watts = 1fps/1watt

2nd gen both represent a 2x increase in performance per watt and are thus equivalent. When we normalize it, we see they produce the same values.

I.e 200fps/100 watts = 2fps/watt. 133.3/66.6 watt = 2fps/watt.

If you look at AMD end notes, AMD gets their inflated performance per watt figures by comparing it to their least efficient architectures. In polaris case, they compared it to hawaii. Polaris definitely does not have 2x the performance of hawaii but approaches 2x the performance per watt because it gets similar performance for half the wattage.

The thing is if AMD could get 2x the performance for the same wattage, they would have said 100% increase in performance and 2x increase in performance per watt. This is not the case.

Instead AMD is stating that their next product has 2x the performance per watt or 35% better performance. Not both at the same time.
tajoh111 is offline  
post #30 of 151 (permalink) Old 07-15-2018, 06:31 AM
New to Overclock.net
Pro3ootector Level 1
 
Pro3ootector's Avatar
 
Join Date: Aug 2013
Location: Poland / Deutschland
Posts: 1,059
Smile

Quote: Originally Posted by tajoh111 View Post
I know what performance per watt is.

What you don't realize it is a scalable term.

2x the performance while using the same wattage is 2x performance per watt while similarly, having 33.33%extra performance while using 66.6% of the power is also 2x performance per watt. This is because mathematically they are the same.

Say a GPU gets 100 fps with first gen technology, using 100 watts. Lets say 2nd generation, they get 2x performance per watt. This could mean 200fps for 100watts or 133fps for 66.6 watts and they would be mathematically identical in terms of performance per watt.

Where the first gen of the technology was 100fps/100watts = 1fps/1watt

2nd gen both represent a 2x increase in performance per watt and are thus equivalent. When we normalize it, we see they produce the same values.

I.e 200fps/100 watts = 2fps/watt. 133.3/66.6 watt = 2fps/watt.

If you look at AMD end notes, AMD gets their inflated performance per watt figures by comparing it to their least efficient architectures. In polaris case, they compared it to hawaii. Polaris definitely does not have 2x the performance of hawaii but approaches 2x the performance per watt because it gets similar performance for half the wattage.

The thing is if AMD could get 2x the performance for the same wattage, they would have said 100% increase in performance and 2x increase in performance per watt. This is not the case.

Instead AMD is stating that their next product has 2x the performance per watt or 35% better performance. Not both at the same time.
I have absolutely no idea why did You think i haven't figure that out. Thanks for the lecture anyways.
Pro3ootector is offline  
Reply

Quick Reply
Message:
Options

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off