Vega Thread

[BenSkywalker]

I would note that AMD doesn't "have" to make NVs R&D budget, they have to hire well. Have you seen the movie "Moneyball"?

I'm a Boston based geek, quite familiar with Billy Bean before the book or the movie(both of which I have consumed)- the Sloan Sports Analytics is based right down the road a bit from me(I'm also a big sports fan in general). This is a bit of an issue for AMD though- even if executed nearly perfectly- the A's have not won a World Series with their approach, it hasn't happened. The second problem is that nVidia has already been using this technique for a long time now- with the budget of the Dodgers. 

I think AMD just needs to hold on, sell some Vegas, and that for a while Ryzen and consoles are going to be funding the graphics division.

The problem, which we are starting to see evidence of, which I already mentioned multiple times- is their designs are absurdly expensive to build. Let's look at the 1080Ti, if nVidia has a 100% gross margin on those parts along with a 20% margin for OEMs, 10% for distributors and 25% for retailers- nV's average margin is close to 60%, so assuming a 100% margin on their high end parts is likely close to accurate.

That puts the cost at the 1080Ti at around $210 and it is retailing for $700. We know that Vega is *MORE* expensive to produce then the 1080Ti- even the Vega 56. If we remove $210- every penny of margin- we end up with Vega 64's MSRP. In other words, selling Vegas at their given MSRP could very well lose them money with every unit sold. Previously I was trying to illustrate pie in the sky numbers for AMD if they wanted to hope to compete with nVidia, reality is *FAR* bleaker.

AMD made too much of an investment in graphics and graphics are too intertwined with CPUs these days for AMD to just walk away.

It's simply losing money, a lot of it, and the situation for the next year and a half is likely to get far worse(I say likely based on not knowing Navi's time frame). AMD spent considerable money doing the incredibly smart thing years back when they built the Adreno platform and combined it with their ARM design to make Snapdragon. They sold that to Qualcomm for $65 Million. Qualcomm makes that in profit on a weekly basis now, owed almost entirely to the design AMD sold them. This is not the way to cultivate talent. 

Last, there are no "Al Bundys" in microchip design. If a person can do that job at any level they are far above you and I intellectually.

So the reason I've been rarely on for the last few weeks is I've been dealing with some major changes to inventory procedures where I work. We have to account for billions of dollars worth of product, it's a pretty big deal. Some extremely intelligent in accounting came up with a new way that we are supposed to do it to make the process far more accurate and cost efficient. I can humiliate the average person when it comes to accounting, but these guys make me look like Forest Gump(I am not an accountant). This new system was tested in three different facilities, all of them came back with massive problems, so it was handed to us next. 

While the methodology and reasoning behind everything they wanted done was extremely well though out, the practical implementation had a few *glaring* flaws on an operational basis that these people wouldn't understand- nor did they so much as attempt to argue the points. With *very* minor tweaks that resulted in a staggering cost reduction(I was able to pull it off with *no* additional labor costs incurred) their process is now ready to roll out to every location.

What am I getting at?

Vega is a fucking terrible design. I could explain GPU architecture to my 12 year old and he'd be able to point out some of the absurd idiocy(ROPs being a shockingly obvious one). The guys designing the transistors could be the best in the world(given the clock restrictions and overall performance, that is laughable, but even if they were) but if the overall design is utter shit- which Vega clearly was from day one, it will not allow their designs to work. There are game breaking flaws in the design- obvious ones, at a practical level.

People have told me in the past how off base I've been when slamming an architecture to this extreme of a level, I did with the S3 2000, the Matrox Parhelia and the 3dfx VSA-100. I was told people a lot smarter then me thought these were great designs and how dare I imply they were company(or business segment) ending poor. I have not said that about AMD's other failures, I am about Vega.