AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

But ist realay strange. Why they dont promote this information to Devs at a Devs meeting locked with an NDA that Nvidia dont get to much Information? Why the puplished all these feature for final customer on all Review sides over the world?
Thats realy bad Marketing.
 
Digidi said:
But ist realay strange. Why they dont promote this information to Devs at a Devs meeting locked with an NDA that Nvidia dont get to much Information? Why the puplished all these feature for final customer on all Review sides over the world?
Thats realy bad Marketing.
Click to expand...

Were you drunk while writing this post? It reads like you were :D
 
English is not my native language and i'm fighting with Auto correction and i'm realy tiered after 14 houres of work.
 
Digidi said:
But ist realay strange. Why they dont promote this information to Devs at a Devs meeting locked with an NDA that Nvidia dont get to much Information? Why the puplished all these feature for final customer on all Review sides over the world?
Thats realy bad Marketing.
Click to expand...
Confidential disclosures under NDA don't generate interest in a product or assuage investors that RTG is generating results.
RTG's outward stance has a higher proportion of bark to bite, and has for some time, in my opinion.

Nvidia has a high likelihood of picking up on leaks or inferring what AMD needs to do anyway, so involving more than a small number of parties just to hide information from Nvidia doesn't seem useful. As evidence, Nvidia seems to have cast some accurate shade on Vega in its quarterly call prior to Vega's launch, where the official position was they expected Vega would not change the competitive landscape. That could be done with some combination of the above, NDA or not.

As far as developers, primitive shaders may be a toy for AMD's development staff, but not for most external ones. The rasterizer might have some items that developers could target to avoid various potential glass jaws, but it's supposed to generally work transparently.
 
CarstenS said:
Is it? I mean in the sense that Blender/ProRender does give you that number or did you calculate this from running at given clock speed? My Vega56 runs at 1430 MHz for XMR mining with 100% load and under 90 Watt GPU-only power draw and a 95 Watt delta before and after starting mining (it's not at idle clocks without mining due to "mining settings").
Click to expand...

Theoretical of course!
The comment simply referred to the ability to clock the 64 NCUs continuously at 1624MHz in compute scenarios.
No undervolt or any advanced setting whatsoever, just the power saver profile.
 
ah, I thought/hoped maybe ProRender tells you current TFLOPS or such which would be quite interesting to see whether it fluctuates with different project.
 
I think remember that Vega 20 with 32GB of HBM was expected to be more a professional product... and for half of 2018 in 7nm...

Now that we know that AMD will launch new product using the 12nm´Finfet certainly before it.. that change a bit some little thing.

AMD-VEGA-20-specifications-1000x377.jpg
 
Picao84 said:
Why are you making conclusions for new chips based on a slide about the Total Available Market? :-|
Click to expand...
Because there has to be a new product to increase their TAM and it has to be a high end product based off the slide. Vega 10 seems to be insufficient to fill that gap and Vega 20 rumors were pointing to Vega 10 levels of FP32.
 
itsmydamnation said:
Because there has to be a new product to increase their TAM and it has to be a high end product based off the slide. Vega 10 seems to be insufficient to fill that gap and Vega 20 rumors were pointing to Vega 10 levels of FP32.
Click to expand...

"Total Addressable market (TAM), or Total Available Market, is the total market demand for a product or service, calculated in annual revenue or unit sales if 100% of available market is achieved."

Or it just reflects the fact that in 2018 they have a high end part available all year, unlike 2017 where the line up ended with Polaris.
Or it reflects that the price will go down leading to a higher potential market.
That slide can be read in many ways.

Not saying that another chip isn't coming. It probably is, but just seemed weird to conclude anything from such a generic indicator.
 
Picao84 said:
"Total Addressable market (TAM), or Total Available Market, is the total market demand for a product or service, calculated in annual revenue or unit sales if 100% of available market is achieved."

Or it just reflects the fact that in 2018 they have a high end part available all year, unlike 2017 where the line up ended with Polaris.
Or it reflects that the price will go down leading to a higher potential market.
That slide can be read in many ways.

Not saying that another chip isn't coming. It probably is, but just seemed weird to conclude anything from such a generic indicator.
Click to expand...
I didn't realise conclusions ended in question marks :oops:
 
itsmydamnation said:
I didn't realise conclusions ended in question marks :oops:
Click to expand...

Well, I made you a question as well. Your answer was this:

itsmydamnation said:
Because there has to be a new product to increase their TAM and it has to be a high end product based off the slide. Vega 10 seems to be insufficient to fill that gap and Vega 20 rumors were pointing to Vega 10 levels of FP32.
Click to expand...

You were adamant that there needed to be a new product to increase their TAM: "Has to be" is pretty much a conclusion. You have just shown that you did not know what TAM is or how it is calculated. I have just showed you that and how it affects your deduction. But somehow you backtrack, just so you can "win" an argument? A little humbleness does not hurt you know?
 
Vega 10 doesn't feature fast FP64; therefore, there's a substantial portion of the professional market it cannot address. Vega 20 won't have the same limitation, hence the larger TAM.
 
Picao84 said:
Well, I made you a question as well. Your answer was this:



You were adamant that there needed to be a new product to increase their TAM: "Has to be" is pretty much a conclusion. You have just shown that you did not know what TAM is or how it is calculated.
Click to expand...
Now look at the person using a post in the future to address a point that occurred in the past......
You are also miss-interpreting here because that was the logic i used to make the initial comment, but i also added uncertainty to the original statement because i was not expecting higher level performance in this space from the Vega generation.

I've also never seen TAM calculated as a figure relative to the amount of time the products were in that market and your quote doesn't say that either. To me that would be very dangerous because the target is moving. The only sane way is to calculate the period of time relative to a fixed point( current product stack, current state of innovation etc) . So i went and did some reading and everything i found looked Something like this or this ( look at pages 9 and 10) and it shows in no way is the TAM calculated with when your products entered the market within a given time period. Seems to me SAM or SOM would match what you are calling TAM.

Now if you want to say that the line next to the bars is SAM then i couldn't discount your hypothesis but you couldn't also discount mine and i wasn't making a definitive statement.

I have just showed you that and how it affects your deduction. But somehow you backtrack, just so you can "win" an argument? A little humbleness does not hurt you know?
Click to expand...
I wasn't looking to win anything, i'm not that desperate..

Alexko said:
Vega 10 doesn't feature fast FP64; therefore, there's a substantial portion of the professional market it cannot address. Vega 20 won't have the same limitation, hence the larger TAM.
Click to expand...
My interpretation is two slide later they define what the 3 GPU markets are, and the slide in question only lists two of them, GPU compute is Radeon Instinct. So what content creation needs FP64? (honest question).
 
itsmydamnation said:
I've also never seen TAM calculated as a figure relative to the amount of time the products were in that market and your quote doesn't say that either. To me that would be very dangerous because the target is moving. The only sane way is to calculate the period of time relative to a fixed point( current product stack, current state of innovation etc) . So i went and did some reading and everything i found looked Something like this or this ( look at pages 9 and 10) and it shows in no way is the TAM calculated with when your products entered the market within a given time period. Seems to me SAM or SOM would match what you are calling TAM.
Click to expand...

There is no definitive formula for calculating the TAM since it is always an approximate number. However, from the moment there is a comparison between timeframes, the number they reached might / should have taken into account time. Even the documents you quoted state "5 million per year". How would you calculate something per year without having a time variable involved? For most of 2017 they had no products to compete against GP104 / GP102 / GP100, therefore its automatically true that their TAM in the year going forward is higher than the previous.

In fact, to see that this is true, you just have to look at Polaris:
- Polaris 10 was launched in June 2016
- Polaris 11 was launched in August 2016

Both were launched in 2016 and their TAM increased by a good bit in 2017 (25% - 30%?). Were there other Polaris GPUs launched in 2017? Not really, there were rebrands but they did not open a new market. If your reasoning that TAM is dissociated with the date the product is on the market, the Polaris bar block would be stable between years (inflation and monetary factors surely do not move the TAM that much between consecutive years).

Regardless, this is going largely off topic and I never refuted that other GPUs are coming (that would be silly), just that that chart is not really a valid data point for anything specific.
 
You must log in or register to reply here.
Back
Top