** The AMD Navi Thread **

[sparkymark75]

It is coming nearly 1 year later, has less featres, and offers was price-performance than existing AMD cards. Let alone the fact that it seems Nvidia will steamroll over Navi with faster and cheaper super cards.

Features (such as ray-tracing) which are pretty useless at this tier of graphics cards. AMD announced a couple of features in their Navi cards that aren't in nVidia cards (that I'm aware of). AMD GPU tech being in consoles means it could be easier for their value add features to receive widespread adoption with developers.

 

[henson0115]

It's not cheaper directly, it's down to the amount of usable "stuff" from a wafer.

Take a sheet of A4 paper. That paper costs £10 a sheet. Chop it into 4 pieces and each piece costs you £2.50. Now, take another A4 sheet that costs £20 and chop it into 10 pieces, each piece costs £2. Then if you sell these pieces you create for £10 each, you'll get back £40 total for the first sheet (£30 profit) but £100 total for the 2nd sheet (£80 profit).

So although the 2nd A4 sheet cost twice as much, the amount of pieces you get out of it are cheaper per unit and have a higher profit margin.

you are forgetting the recoup of RND etc.

 

[Sargatanas2511]

you are forgetting the recoup of RND etc.

Plus there's no guarantee a 7nm wafer costs the same to manufacture as a 12nm wafer. The amount of money TSMC put into it I would imagine per-wafer cost for 7nm being a considerable amount higher.

 

[henson0115]

Plus there's no guarantee a 7nm wafer costs the same to manufacture as a 12nm wafer. The amount of money TSMC put into it I would imagine per-wafer cost for 7nm being a considerable amount higher.

indeed seems most here have also forgotten that fact.

 

[LePhuronn]

That makes sense, cheers. Still a bit meh to just assume its cheaper to produce though as some people claim.

Well arguably it is cheaper to produce because the cost per unit of 7nm dies is lower than 14nm dies.

Let's assume every other component on a graphics card is the same price:

£10 for the PCB
£10 for the cooler
£10 for the RAM

So that's £30 per graphics card in bits, so £1,200 for 40 graphics cards

I want to make 40 cards on 14nm and 40 cards on 7nm

14nm wafer costs £10 each and I get 4 dies, so I need 10 wafers at a cost of £100
7nm wafer costs £20 and I get 10 dies, so I need 4 wafers at a cost of £80

40 14nm cards cost me £1,300
40 7nm cards cost me £1,280

Therefore, in this incredibly simplistic example, it is cheaper to produce the cards using the smaller die.

 

[D.P.]

The chip itself is made in Taiwan but I'm sure iirc, the cards are assembled in China and shipped from there, hence the tariff would apply?

It would not surprise me if the UK gets its products tariff free but the retailers add it anyway and hope nobody questions it, therefore pocketing an extra 25% or being able to offer big discounts whenever they want

Conspiracy theory right there... I've been watching too much American Politics

The tariffs are only on select products like steel, weapons, TVs. I do;t believe GPUs are covered, but either-way it makes no difference because nvidia GPus would have the same 25% tariffs applied if they existed.

 

[henson0115]

Well arguably it is cheaper to produce because the cost per unit of 7nm dies is lower than 14nm dies.

Let's assume every other component on a graphics card is the same price:

£10 for the PCB
£10 for the cooler
£10 for the RAM

So that's £30 per graphics card in bits, so £1,200 for 40 graphics cards

I want to make 40 cards on 14nm and 40 cards on 7nm

14nm wafer costs £10 each and I get 4 dies, so I need 10 wafers at a cost of £100
7nm wafer costs £20 and I get 10 dies, so I need 4 wafers at a cost of £80

40 14nm cards cost me £1,300
40 7nm cards cost me £1,280

Therefore, in this incredibly simplistic example, it is cheaper to produce the cards using the smaller die.

if that were true, why did global foundries annouce that they were abandoning the process a last year back due to costs involved? its not that simple i am afriad. as stated multiple times above.

 

[LePhuronn]

you are forgetting the recoup of RND etc.

if that were true, why did they almost abandon the process a few years back due to costs invovled? its not that simple i am afriad. as stated multiple times above.

Recouping R&D, manufacturing costs and whatnot are outside the scope of my incredibly simplistic example of how smaller dies can be cheaper despite larger wafer costs.

I never said actual GPU manufacture was this easy, it's purely a simplistic illustration to answer Chrisc's question.

 

[henson0115]

Recouping R&D, manufacturing costs and whatnot are outside the scope of my incredibly simplistic example of how smaller dies can be cheaper despite larger wafer costs.

I never said actual GPU manufacture was this easy, it's purely a simplistic illustration to answer Chrisc's question.

and thats how the spread of most false info starts

 

[pmc25]

I'll likely get a 5700XT 50th on day one if RAL (Radeon Anti-Lag) isn't coming to Vega, and sell my RVII.

Re: all the REEEEEE about pricing. What do you expect? Giveaways? If RAL is exclusive to Navi, then that's a huge killer app for them until NVIDIA respond. That's way more important than a couple more FPS, even if people don't yet realise.

Also, as I've repeatedly warned and people have repeatedly rubbished, there is no chance of low pricing because wafer supply is not infinite. If AMD priced Navi and Ryzen 3 significantly lower, then demand would vastly outstrip supply and it'd be a disaster. Until they have more wafers, there won't be any significant price reduction. Navi prices will probably fall quite a bit in Q4 when this happens, but not before.

I did say the 16 core wouldn't launch below $750 and if it did launch, wouldn't come in July. But no, people had convinced themselves that AMD wanted to give them a free lunch, and supply be damned.

Virtually none of the posts since the announcement have been even moderately realistic.

 

[LePhuronn]

if that were true, why did global foundries annouce that they were abandoning the process a last year back due to costs involved? its not that simple i am afriad. as stated multiple times above.

Global Foundries are ****, that's why. TSMC and Samsung have steamed along with 7nm and EUV. Why does 1 company abandoning their 7nm development invalidate my very simplistic illustration?

 

[D.P.]

We need to give AMD time to sort the GPU side out, now the CPU side is sorted and generating money. Remember the GPU R&D is miniscule in comparison to NVIDIA. When AMD get the cash flowing from the CPU side, it'll hopefully be put into GPU side and they can do more.

People on here forget that AMD's GPU R&D isn't even close to Nvidias, yet they are demanding AMD match or beat NVIDIA. Give them time. I remember that, VERY recently, people lamented AMD for rubbish CPU's. Now look...

Who cares? Amd isn;t a charity, as a conusmer you can only make purchases based on what they produce

 

[D.P.]

We need to give AMD time to sort the GPU side out, now the CPU side is sorted and generating money. Remember the GPU R&D is miniscule in comparison to NVIDIA. When AMD get the cash flowing from the CPU side, it'll hopefully be put into GPU side and they can do more.

People on here forget that AMD's GPU R&D isn't even close to Nvidias, yet they are demanding AMD match or beat NVIDIA. Give them time. I remember that, VERY recently, people lamented AMD for rubbish CPU's. Now look...

Who cares? Amd isn;t a charity, as a conusmer you can only make purchases based on what they produce

 

[D.P.]

Profit margin source?

His backside. Nvidia's margins are around 35% company wide, but that includes selling things like Volta GPU's at USD15,000 a pop, 8K Quadro cards, and ridiculous Titan cards. Margins on the 2070 are probably about 30%.

 

[D.P.]

www.google.co.uk

https://ycharts.com/companies/NVDA/gross_profit_margin

There's not much to read, just a chart covering the last ten years or so.

That is gross margin, whcih completely ignores costs such as R&D which is obviously huge.

The useful figure is net rpofit margin, whcih tells you the absolute profit Nvidia will make:
https://ycharts.com/companies/NVDA/profit_margin
Currently down to 17.75%.

 

[sparkymark75]

His backside. Nvidia's margins are around 35% company wide, but that includes selling things like Volta GPU's at USD15,000 a pop, 8K Quadro cards, and ridiculous Titan cards. Margins on the 2070 are probably about 30%.

And yet you've just done the exact same thing, pulled the numbers from your "backside". Only one's that will know for sure are nVidia.

 

[D.P.]

Plus there's no guarantee a 7nm wafer costs the same to manufacture as a 12nm wafer. The amount of money TSMC put into it I would imagine per-wafer cost for 7nm being a considerable amount higher.

It costs about 3x the amount of money tor produce a chip on 7nm vs 12/14/16nm for the same given size. The hope would be if you move to 7nm your chip gets smaller so yields increase.

Except with AMD the die size has remained the same, so the costs are now 3x higher. And that is only true if 7nm has the same nominal yields at 14nm, most sources say 7nm yields are worse for for equivalent chip size and process maturity.

 

[D.P.]

And yet you've just done the exact same thing, pulled the numbers from your "backside". Only one's that will know for sure are nVidia.

No, I linked to the number later on:
https://ycharts.com/companies/NVDA/profit_margin

Currently 17% but was 38% last year

 

[D.P.]

Well arguably it is cheaper to produce because the cost per unit of 7nm dies is lower than 14nm dies.

Let's assume every other component on a graphics card is the same price:

£10 for the PCB
£10 for the cooler
£10 for the RAM

So that's £30 per graphics card in bits, so £1,200 for 40 graphics cards

I want to make 40 cards on 14nm and 40 cards on 7nm

14nm wafer costs £10 each and I get 4 dies, so I need 10 wafers at a cost of £100
7nm wafer costs £20 and I get 10 dies, so I need 4 wafers at a cost of £80

40 14nm cards cost me £1,300
40 7nm cards cost me £1,280

Therefore, in this incredibly simplistic example, it is cheaper to produce the cards using the smaller die.

That is only true if die sizes decrease. Which they haven't, Navi looks to be 251mm^2, basically he same as Polaris.

 

[Chrisc]

Well arguably it is cheaper to produce because the cost per unit of 7nm dies is lower than 14nm dies.

Let's assume every other component on a graphics card is the same price:

£10 for the PCB
£10 for the cooler
£10 for the RAM

So that's £30 per graphics card in bits, so £1,200 for 40 graphics cards

I want to make 40 cards on 14nm and 40 cards on 7nm

14nm wafer costs £10 each and I get 4 dies, so I need 10 wafers at a cost of £100
7nm wafer costs £20 and I get 10 dies, so I need 4 wafers at a cost of £80

40 14nm cards cost me £1,300
40 7nm cards cost me £1,280

Therefore, in this incredibly simplistic example, it is cheaper to produce the cards using the smaller die.

That makes things clearer from a production point I guess.