[Possibly fake] AMD losing 100+ dollars on every vega sold

[JoostinOnline]

How reliable is Fudzilla?  I've never heard of them.

 

Edit: I think the answer is "not very reliable".  They have a "confirmation" that Intel is licensing AMD graphics.  That was a rumor from several months ago that was quickly squashed.

[Mira Yurizaki]

10 minutes ago, JoostinOnline said:

How reliable is Fudzilla?  I've never heard of them.

 

Edit: I think the answer is "not very reliable".  They have a "confirmation" that Intel is licensing AMD graphics.  That was a rumor from several months ago that was quickly squashed.

Well I'm sure their name comes from "FUD" which stands for "Fear, Uncertainty, Doubt"

[Drak3]

AMD would absolutely sell VEGA at a $100 loss per card, if the alternative of scrapping them is a $101 loss per card.

 

Someone, please feel free to mention everyone wondering why AMD would sell at a loss, or claiming that they wouldn't, and quote this. I'm just gonna be too lazy to, this go round.

[yian88]

Can someone enlighten me why HBM2 is necesary and how does Nvidia get away with GDDR5 and GDDR5x? i know since 900 Maxwell series they are using some compression algorithms to maintain bandwidth, i dont think my GTX 970 would be satisfied with 210GB/s without compression.

But still why isnt AMD doing the same? nvidia found 2 ways compression + GD5x double the data.

AMD is killing themselves with HBM since fury X why cant they come up with something better?

At this rate HBM3-4 may be ready and cost effective post 2020's, is radeon going to sink? they havent released anything good for so many years, since hd 7000 series they only had busted cards highpower and not enough perf.

[JediFragger]

11 hours ago, Drak3 said:

AMD would absolutely sell VEGA at a $100 loss per card, if the alternative of scrapping them is a $101 loss per card.

 

Driver support over time also has to be factored in imo, so would be a LOT more.

[JediFragger]

40 minutes ago, yian88 said:

Can someone enlighten me why HBM2 is necesary and how does Nvidia get away with GDDR5 and GDDR5x? i know since 900 Maxwell series they are using some compression algorithms to maintain bandwidth, i dont think my GTX 970 would be satisfied with 210GB/s without compression.

But still why isnt AMD doing the same? nvidia found 2 ways compression + GD5x double the data.

AMD is killing themselves with HBM since fury X why cant they come up with something better?

At this rate HBM3-4 may be ready and cost effective post 2020's, is radeon going to sink? they havent released anything good for so many years, since hd 7000 series they only had busted cards highpower and not enough perf.

AMD took a punt and failed, at least from a cost perspective. HBM is going to continue to kick them for a while yet...

[Drak3]

1 hour ago, JediFragger said:

Driver support over time also has to be factored in imo, so would be a LOT more.

AMD wouldn't stop creating drivers for GCN cards, so no, it wouldn't be alot more to continue creating drivers for all GCN based cards.

[Misanthrope]

On 8/30/2017 at 9:32 PM, AnonymousGuy said:

Yeah I'm calling bullshit on this.  Why would AMD even sell it if it's costing them money?  They would just scrap the project.

If they already made a contract to buy x amount of HBM2 for example it might be more expensive to scrap the entire thing than to just release it anyway and lose money on every card if you're losing less money than you otherwise would by bailing on established contracts.

 

I don't think this is very likely however I'm with you of that but it's not completely unfeasible to release a product at lost just hoping to recover some of the costs you had for production.

[ZestyJalapeno]

Calling fake news on this "Industry source" 

[JoostinOnline]

@Coaxialgamer Given how sketchy the source is, you should mark this as a rumor.

[ZestyJalapeno]

5 minutes ago, JoostinOnline said:

@Coaxialgamer Given how sketchy the source is, you should mark this as a rumor.

Raja Koduri, a guy who actually works at AMD already refuted this on twitter. More misinformation than rumour.

[JoostinOnline]

2 minutes ago, Dogeystyle said:

Raja Koduri, a guy who actually works at AMD already refuted this on twitter. More misinformation than rumour.

Link?

[ZestyJalapeno]

4 minutes ago, JoostinOnline said:

Link?

Never mind, he didn't come out and say anything that directly addressed it, but just made a rather vague statement about the pricing rumours not being credible, as a reply to himself.

https://twitter.com/GFXChipTweeter

[Mihle]

 

[ravenshrike]

17 hours ago, Drak3 said:

AMD would absolutely sell VEGA at a $100 loss per card, if the alternative of scrapping them is a $101 loss per card.

 

Someone, please feel free to mention everyone wondering why AMD would sell at a loss, or claiming that they wouldn't, and quote this. I'm just gonna be too lazy to, this go round.

That's actually a good point. It's entirely possible that all or most of the dies and HBM2 used are parts that didn't quite meet quality control for Instinct and SSG but they had to buy because of their contracts anyway.

[79wjd]

15 hours ago, yian88 said:

Can someone enlighten me why HBM2 is necesary and how does Nvidia get away with GDDR5 and GDDR5x? i know since 900 Maxwell series they are using some compression algorithms to maintain bandwidth, i dont think my GTX 970 would be satisfied with 210GB/s without compression.

But still why isnt AMD doing the same? nvidia found 2 ways compression + GD5x double the data.

AMD is killing themselves with HBM since fury X why cant they come up with something better?

At this rate HBM3-4 may be ready and cost effective post 2020's, is radeon going to sink? they havent released anything good for so many years, since hd 7000 series they only had busted cards highpower and not enough perf.

HBM2 isn't necessary (at least for gaming), and AMD does use compression, but AMD for whatever reason has decided to commit itself to using HBM2 across the board on all high end cards -- regardless of whether or not its needed and regardless of whether or not it makes any business sense to do so.

 

Let's all just say a prayer that they were smart enough not to use HBM2 on Polaris. 

[Red Hardware]

On 8/31/2017 at 0:07 AM, Coaxialgamer said:

http://fudzilla.com/news/graphics/44401-amd-is-losing-100-on-every-vega (source)

 

This is bad . Very bad . Worse than i imagined t, given pre-launch information .

The chip is large , power hungry , expensive to manufacture and performance is largely underwhelming ; competing against a gpu 65% it's size.

I'm going to get hate for saying this , but i dare say that this is worse than Fermi .

 

Edit:Also this

  Amd sells all their vega cards at the same price , regardless of retail price. Amd isn't making money on vega price gouging. Retailers and re-sellers are.

Hi

Let me make it clear , nothing can be as bad as Fermi , NOTHING

 

[Coaxialgamer]

38 minutes ago, Red Hardware said:

Hi

Let me make it clear , nothing can be as bad as Fermi , NOTHING

 

Let me check... 

-High power draw : check

-High heat output :check

-Large die :check

-Expensive to manufacture : check

- Poor yields : check

-Hits thermal limit on stock cooler:check 

-Late :check

-Not enough stock at launch :check

 

The only difference is that the gtx 480 was ACTUALLY FASTER than its competition ( the cypress hd 5870).

Vega barely matches the gtx 1080, which is a mid range chip. 

 

[Red Hardware]

3 hours ago, Coaxialgamer said:

Let me check... 

-High power draw : check

-High heat output :check

-Large die :check

-Expensive to manufacture : check

- Poor yields : check

-Hits thermal limit on stock cooler:check 

-Late :check

-Not enough stock at launch :check

 

The only difference is that the gtx 480 was ACTUALLY FASTER than its competition ( the cypress hd 5870).

Vega barely matches the gtx 1080, which is a mid range chip. 

 

Lol

I was joking , but that 1080 is not a mid range chip, mid range is 1060. 1080 is a high end chip but it's not top of the line

 

[ravenshrike]

3 hours ago, Coaxialgamer said:

- Poor yields : check

We have no idea what the yields on the die are though. For all we know they've got scads of the damn things sitting around waiting for HBM2.

[Coaxialgamer]

Just now, Red Hardware said:

Lol

I was joking , but that 1080 is not a mid range chip, mid range is 1060. 1080 is high end but it's not the top of the line

 

gp104is a mid range chip . Nvidia just sells it as high end . 

A few years back , the xx4 class of chips was actually used for gtx XX60 class cards .

The XX6 class of chips were used in even lower class hardware , think gts 450 and gt class cards.

A chip that would have been in a <300$ gpu a fezw years back is now sold >500$

[Coaxialgamer]

1 minute ago, ravenshrike said:

We have no idea what the yields on the die are though. For all we know they've got scads of the damn things sitting around waiting for HBM2.

it's pretty clear they are having difficulties . The chips is large .

Most vega cards are overvolted way too much out of the factory . That's a typical sign that AMD couldn't get enough good dies ,and had to increase voltage to achieve enough quantity .

 

[TheRandomness]

1 minute ago, Coaxialgamer said:

gp104is a mid range chip . Nvidia just sells it as high end . 

A few years back , the xx4 class of chips was actually used for gtx XX60 class cards .

The XX6 class of chips were used in even lower class hardware , think gts 450 and gt class cards.

A chip that would have been in a <300$ gpu a fezw years back is now sold >500$

Then again, you'll realise that there are no GP100 cards with more than 3584 shaders, whereas GP102 has gone to 3840 or whatever the Titan Xp has. 

[Coaxialgamer]

11 minutes ago, TheRandomness said:

Then again, you'll realise that there are no GP100 cards with more than 3584 shaders, whereas GP102 has gone to 3840 or whatever the Titan Xp has. 

that's just yields . Gp100 does have 3840 FP32 shaders . Nvidia could enable them if they had the motivation to do so .

Sure , the nomenclature has changed a bit , but it used to be 

XX6 gpu : entry level (~100-150$)

XX4 gpu : mid range ( 200-300$) 

XX0 ( now XX2 ) : high end ( 500$)

 

That was the case until kepler anyway.

 

The gtx 1080 is a small gpu 

 

[DildorTheDecent]

7 minutes ago, Coaxialgamer said:

 

aye, they can get kinda wee

image pls show: 

Spoiler