Zen 3 is rumored to be flaunting monumental IPC gains in early testing

sirmonkey1985

[H]ard|DCer of the Month - July 2010
Joined
Sep 13, 2008
Messages
21,752
Let's analyze this with a comparison.

Boeing : Market Cap 187B$.
SpaceX : Market Cap... Private. Funding 3 B$, but not for what I'll be explaining (but for some even more weird stuff).
Frankly speaking SpaceX is so little compared to Boeing that it's crappy to compare. It's just ridiculous. They have no history vs Boeing. You wouldn't bet a dime on it if you're financial and look at the financial records. If you were a financial guy you would just say that SpaceX is a hoax.
NASA launched a competition between the two for a crowded ship and both answered. Boeing is a half failure, not yet ready, not sure will ever be, even missed a launch, will cost double (90M $) per launch. SpaceX is ready is on third test to be validated, costs 55M$ per launch.
to add on to what you said as well, market cap is great on paper but there's a massive difference between a company that has a huge market cap and their hands in everything vs a company that's specialized in one or two things while still having a significantly lower cap.. for a company like Boeing or Intel you can't just magically pull money out of your ass and shift it to a single division, you have to take that money from some where else within the company. hell just moving a billion dollars from one division to another for either of those companies is damn near impossible while spaceX or AMD who specialize in basically 2 major products within their company could easily shift money back and forth without any major negative effects for either of them. i think people tend to forgot just how divided a lot of these giant companies are within their own "brand".
 

DuronBurgerMan

[H]ard|Gawd
Joined
Mar 13, 2017
Messages
1,354
AMD has a bigger long-term question: will TSMC continue to be able to deliver?

What's the plan for when they inevitably do not?

What's the plan when TSMC decides that someone else is more deserving of their fab capacity?

This is why Intel stayed in the fab business; it's biting them in the ass today, but it has delivered long-term and will likely continue to deliver long-term.

At the very least, TSMCs current process family has been fairly successful, and thankfully while Intel is experiencing their rare stumble, AMD has something actually worth producing!
AMD can still screw this up. However, IIRC they've invested time/effort assuring that designs can be migrated between fabs/similar processes quickly enough. Being fabless is calculated risk. It pays off if you're smaller - as AMD obviously is. They don't have to rely on internal fab teams to stay in sync with design teams. They can just pick the best off the shelf processes and use that. But it's also a risk in that if nobody can keep up with Intel, they have no control over the situation, and if other fabs decide to charge too high of a fabrication cost to AMD, AMD has no choice but to eat it. The first thing is mitigated by Intel's stumbles, and TSMC's excellent execution. The latter is mitigated by the fact that other fabs continue to give TSMC competition - for now - and so TSMC cannot afford to alienate AMD by raking them over the coals. All of that could change.

That being said, I think AMD's gamble paid off, they are going in the right direction, and Intel has made some serious (though non-fatal) mistakes.
 
Joined
Jan 14, 2012
Messages
748
It always amazes me when you comment that AMD has tied their hopes to TSMC and how that is a looming question of a possible issue, then in another thread you have no Issue with Nvidia doing the same thing and has for years.
Nvidia is moving 7nm production to Samsung and TSMC this go around. There's no real question, in my mind, if TSMC is overbooked on capacity or not. Nvidia hasn't fabbed at Samsung in a number of years (since 2016 with Pascal) and there was a shortage of cards near the end of the their life cycle even with two fabs producing them. We don't have a mining craze going on now, however, it's pretty apparent what's going on. TSMC is overloaded.
 

Gideon

2[H]4U
Joined
Apr 13, 2006
Messages
2,464
Nvidia is moving 7nm production to Samsung and TSMC this go around. There's no real question, in my mind, if TSMC is overbooked on capacity or not. Nvidia hasn't fabbed at Samsung in a number of years (since 2016 with Pascal) and there was a shortage of cards near the end of the their life cycle even with two fabs producing them. We don't have a mining craze going on now, however, it's pretty apparent what's going on. TSMC is overloaded.
Well Nvidia is late to the 7nm game, so I am sure TSMC has sold most production to other companies and Nvidia is a bit uncertain what TSMC has left is enough for them. At the same time it seems like Samsung has put more effort into their fab business lately.
 

IdiotInCharge

[H]ardForum Junkie
Joined
Jun 13, 2003
Messages
13,229
Well Nvidia is late to the 7nm game, so I am sure TSMC has sold most production to other companies and Nvidia is a bit uncertain what TSMC has left is enough for them. At the same time it seems like Samsung has put more effort into their fab business lately.
Do you have a link with solid proof that Nvidia doesn't have 7nm capacity reservation at TSMC?
 

Derangel

[H]ard as it Gets
Joined
Jan 31, 2008
Messages
18,467
They probably spent 500 million on that. If they dropped 10 billion into R&D 10nm would be just fine.

Intel will NOT allow AMD to lead for long. They are chipzilla forms reason. They have bank and investors galore.
Throwing more money at something doesn't guarantee a good result. They could have thrown $100B at 10nm and still been in the same situation.

Intel isn't allowing anything. They fucked up, big time, and are paying for it.
 

drescherjm

[H]ardForum Junkie
Joined
Nov 19, 2008
Messages
14,639
Throwing more money at something doesn't guarantee a good result. They could have thrown $100B at 10nm and still been in the same situation.
I think it was a combination of Intel being too aggressive (too small of feature sizes) at 10nm and their 14nm was already too good to beat. Intel couldn't introduce 10nm on the desktop when their 14nm clocks better.
 
Last edited:

Gideon

2[H]4U
Joined
Apr 13, 2006
Messages
2,464
So that's a no on proof that Nvidia doesn't have stock reservations then?

Thanks :ROFLMAO:
So where is your proof otherwise? The fact that Nvidia has stated they are sourcing from both TSMC and Samsung leads more proof to my article I listed, while you have listed nothing to support yours.
 

IdiotInCharge

[H]ardForum Junkie
Joined
Jun 13, 2003
Messages
13,229
So where is your proof otherwise? The fact that Nvidia has stated they are sourcing from both TSMC and Samsung leads more proof to my article I listed, while you have listed nothing to support yours.
Your link provides support for the argument that Nvidia does have capacity reservations at TSMC, and mentions nothing about whether TSMC has 'enough' capacity for Nvidia, and whether Nvidia will need to look elsewhere, which you were alluding to above.

Thanks for playing.
 

Spirit_Retro

Limp Gawd
Joined
Apr 1, 2010
Messages
310
AMD can still screw this up. However, IIRC they've invested time/effort assuring that designs can be migrated between fabs/similar processes quickly enough. Being fabless is calculated risk. It pays off if you're smaller - as AMD obviously is. They don't have to rely on internal fab teams to stay in sync with design teams. They can just pick the best off the shelf processes and use that. But it's also a risk in that if nobody can keep up with Intel, they have no control over the situation, and if other fabs decide to charge too high of a fabrication cost to AMD, AMD has no choice but to eat it. The first thing is mitigated by Intel's stumbles, and TSMC's excellent execution. The latter is mitigated by the fact that other fabs continue to give TSMC competition - for now - and so TSMC cannot afford to alienate AMD by raking them over the coals. All of that could change.

That being said, I think AMD's gamble paid off, they are going in the right direction, and Intel has made some serious (though non-fatal) mistakes.
I don't know. I'm still waiting for the gotcha in going fabless.

If they continue to grow they will have to build their own capacity at some point. There's no way to scale with production and talent outsourced. It eventually breaks down.
 

Gideon

2[H]4U
Joined
Apr 13, 2006
Messages
2,464
Your link provides support for the argument that Nvidia does have capacity reservations at TSMC, and mentions nothing about whether TSMC has 'enough' capacity for Nvidia, and whether Nvidia will need to look elsewhere, which you were alluding to above.

Thanks for playing.
So you have none supporting yours and somehow missed this in the article I linked,

Huawei’s HiSilicon, Qualcomm and SuperMicro are planning 25,000 WPM reservations while MediaTek is going for about 20,000. That leaves roughly 10% of TSMC’s capacity for their other customers, including Nvidia who’ll be looking at 7nm for their Ampere GPUs.
 
Joined
Jan 14, 2012
Messages
748
I think they will have to split their production using more than 1 fab.

I don't think they will get back into the FAB business for many years because of the startup cost.
The Fab business would have to either stabilize or hit a wall, where one type of fabrication was pretty much all that was left. IIRC AMD still owes GloFlo bucks and production. Though GloFlo's decision to not Fab below 12nm (when they scrapped their 7nm) allowed AMD to step around a lot of that. I don't see AMD jumping back into that until they are at parity with Intel. They would have to be making some major bank to be able to dump 18-19 Billion into Capital development like Intel does, annually.
 
Joined
Jan 14, 2012
Messages
748
So you have none supporting yours and somehow missed this in the article I linked,

Huawei’s HiSilicon, Qualcomm and SuperMicro are planning 25,000 WPM reservations while MediaTek is going for about 20,000. That leaves roughly 10% of TSMC’s capacity for their other customers, including Nvidia who’ll be looking at 7nm for their Ampere GPUs.
Thus proving the fact that Nvidia needs to outsource due to insufficient capacity at TSMC. At least, from my point of view. I don't think we will really know the extent of Nvidia's farming out production to Samsung until we get some inside AIB leaks about more GPUs coming from Samsung than TSMC. What I'm curious about is if one fab will be yielding superior silicon (Samsung or TSMC) over the other this generation.

I think both of you have some good points, however, Idiot is just letting you build his case for him.
 

IdiotInCharge

[H]ardForum Junkie
Joined
Jun 13, 2003
Messages
13,229
Huawei’s HiSilicon, Qualcomm and SuperMicro are planning 25,000 WPM reservations while MediaTek is going for about 20,000. That leaves roughly 10% of TSMC’s capacity for their other customers, including Nvidia who’ll be looking at 7nm for their Ampere GPUs.
And 10% is not enough for Nvidia's product lines? Again, source?

I think both of you have some good points, however, Idiot is just letting you build his case for him.
He's making sub-scientific wild-ass-guesses. I won't guess as to why, but it's fun to point out -- I'm making him make his own damn case.

Or not, whichever.
 

Gideon

2[H]4U
Joined
Apr 13, 2006
Messages
2,464
And 10% is not enough for Nvidia's product lines? Again, source?
While this is not 100% accurate if you go by profit on sales it gives you a good idea of quantity needed.

NVIDIA today reported revenue for the third quarter ended Oct. 27, 2019, of $3.01 billion compared with $3.18 billion a year earlier and $2.58 billion in the previous quarter. GAAP earnings per diluted share for the quarter were $1.45, compared with $1.97 a year ago and $0.90 in the previous quarter.

AMD reported earnings of 8 cents per share and revenue of $1.53 billion, compared to estimates of $1.52 billion, according to Refinitiv. AMD revenue was down 13% compared to the same quarter last year, although it is up 20% from the previous quarter.

So AMD is running half the revenue and while that wont mean they are selling half the silicon of Nvidia it will be higher then what AMD needs in production when factoring in the average selling price. They will never break it down enough for you to get a real number but you can get a reasonable guess out of it. AMD is buying 20% of TSMC manufacturing and Nvidia will need more then 10% to meet their needs without a doubt. Also that article did not confirm Nvidia had reservations just that their capacity left and is speculating it's for Nvidia.
 

IdiotInCharge

[H]ardForum Junkie
Joined
Jun 13, 2003
Messages
13,229
They will never break it down enough for you to get a real number but you can get a reasonable guess out of it.
This is what you started guessing with:

and Nvidia is a bit uncertain what TSMC has left is enough for them
I don't see it as very reasonable, especially using revenue. AMD has been selling their products for pennies relative to their competition for quite some time, hence, it does make some sense that Nvidia would pull significantly more revenue out of significantly less volume from TSMC.


Now, it's clear that with Intel's fab struggles and Zen being competitive to what Intel can currently produce, AMD is going to demand more and more of TSMCs production -- but we don't really know where that's going to leave Nvidia, or anyone else.
 

Gideon

2[H]4U
Joined
Apr 13, 2006
Messages
2,464
This is what you started guessing with:



I don't see it as very reasonable, especially using revenue. AMD has been selling their products for pennies relative to their competition for quite some time, hence, it does make some sense that Nvidia would pull significantly more revenue out of significantly less volume from TSMC.


Now, it's clear that with Intel's fab struggles and Zen being competitive to what Intel can currently produce, AMD is going to demand more and more of TSMCs production -- but we don't really know where that's going to leave Nvidia, or anyone else.
Gross margin for AMD is 41% and Nvidia is 56% or a 15% difference. So for it to make a large difference Nvidia would have to be selling mostly high margin product and barely any low end product to live on that 10% left as TSMC. And if you go by the steam survey then yeah it doesnt support Nvidia is selling mostly higher end silicon https://store.steampowered.com/hwsurvey/videocard/

So your rebuttal to me has been just a nuh-uh, will see what happens when the cards get here as people have to want them for it to matter.
 

IdiotInCharge

[H]ardForum Junkie
Joined
Jun 13, 2003
Messages
13,229
Gross margin for AMD is 41% and Nvidia is 56% or a 15% difference. So for it to make a large difference Nvidia would have to be selling mostly high margin product and barely any low end product to live on that 10% left as TSMC. And if you go by the steam survey then yeah it doesnt support Nvidia is selling mostly higher end silicon https://store.steampowered.com/hwsurvey/videocard/
You're also not looking at product lines. I didn't use numbers because specificity will get mired down in the details we don't, can't, and shouldn't have. 10% of a fab's production like TSMCs is still a lot of dies. Further, Nvidia parts tend to be smaller for the same level of performance on the same node for gaming (since the Steam survey was referenced), at least up to Turing with extra core types mixed in; but AMD will have to do much the same with RDNA2, so that trend is likely to continue.

Ergo, Nvidia is making more dies per wafer than AMD, just comparing GPU lines, and each percent of production goes further for Nvidia.

So the question that your guess alludes to as to whether that remaining 10% is enough for Nvidia still stands. We don't (and again, can't, and shouldn't) have hard numbers, but it does stand to reason that since AMD is trying to supply the market that Intel is now undersupplying, they're going to demand more production from TSMC, however, as a matter of percentages of volume, this doesn't really shine any light on how TSMCs other customers are affected. Particularly since we don't know what their current allocations are, nor what they are planning, nor the effect of capacity increases in 7nm production will result in the number of wafers that each percent of production actually represents.


Also, given that Nvidia essentially makes more money per product than AMD and that Nvidia dominates the GPU market, especially the high-end GPU market for consumers and businesses, I'd expect TSMC to have whatever capacity that Nvidia thinks they need dutifully set aside. After all, Nvidia has been a much more reliable customer.
 

DuronBurgerMan

[H]ard|Gawd
Joined
Mar 13, 2017
Messages
1,354
He's making sub-scientific wild-ass-guesses. I won't guess as to why, but it's fun to point out -- I'm making him make his own damn case.

Or not, whichever.
Wild-ass guesses are fine if non-specific claims.

For instance, "Nvidia may not have enough bleeding-edge fab capacity either. They run the same risks as AMD." <--- perfectly fine. Probably true, even.

Versus: "I am sure TSMC has sold most production to other companies and Nvidia is a bit uncertain what TSMC has left is enough for them." <--- you are sure about TSMC production sales figures? Well then, supply them! If not, why do you think this?
 

DooKey

[H]ardForum Junkie
Joined
Apr 25, 2001
Messages
8,286
This is what you started guessing with:



I don't see it as very reasonable, especially using revenue. AMD has been selling their products for pennies relative to their competition for quite some time, hence, it does make some sense that Nvidia would pull significantly more revenue out of significantly less volume from TSMC.


Now, it's clear that with Intel's fab struggles and Zen being competitive to what Intel can currently produce, AMD is going to demand more and more of TSMCs production -- but we don't really know where that's going to leave Nvidia, or anyone else.
We do know. Nvidia is a bigger company and will have more money to spend. Easy peasy.
 

Gideon

2[H]4U
Joined
Apr 13, 2006
Messages
2,464
We do know. Nvidia is a bigger company and will have more money to spend. Easy peasy.
Really doesnt work that way,those contracts have long since been signed. So for the short term that would hold no bearing but next round of contracts it might matter. TSMC did increase production capacity as well and we only have some insight on production and what people are buying was because AMD disclosed it, to take blame off TSMC for AMD cpu shortages. Plus if you drive your margins down stock holders will remove you as a CEO pretty quickly these days. Were all just relying on rumors and a bit of fact to make our own conclusions.
 

IdiotInCharge

[H]ardForum Junkie
Joined
Jun 13, 2003
Messages
13,229
Really doesnt work that way,those contracts have long since been signed. So for the short term that would hold no bearing but next round of contracts it might matter.
That really is what we're talking about, though. That AMD has reserved more wafer runs -- what we don't know is whether those runs have come at the expense of others as you 'guessed' above, potentially Nvidia, if the runs are coming from increased production, or some combination, and further, whether that has any impact on what AMDs competitors will do, namely Nvidia.

And to add to that uncertainty, an example: Apple was selling phones in the US with processors made by Samsung alongside Samsung phones in the US with processors made by Qualcomm at TSMC... drawing conclusions of any sort seems a bit silly in this business, as enthusiasts.
 

Gideon

2[H]4U
Joined
Apr 13, 2006
Messages
2,464
You're also not looking at product lines. I didn't use numbers because specificity will get mired down in the details we don't, can't, and shouldn't have. 10% of a fab's production like TSMCs is still a lot of dies. Further, Nvidia parts tend to be smaller for the same level of performance on the same node for gaming (since the Steam survey was referenced), at least up to Turing with extra core types mixed in; but AMD will have to do much the same with RDNA2, so that trend is likely to continue.

Ergo, Nvidia is making more dies per wafer than AMD, just comparing GPU lines, and each percent of production goes further for Nvidia.

So the question that your guess alludes to as to whether that remaining 10% is enough for Nvidia still stands. We don't (and again, can't, and shouldn't) have hard numbers, but it does stand to reason that since AMD is trying to supply the market that Intel is now undersupplying, they're going to demand more production from TSMC, however, as a matter of percentages of volume, this doesn't really shine any light on how TSMCs other customers are affected. Particularly since we don't know what their current allocations are, nor what they are planning, nor the effect of capacity increases in 7nm production will result in the number of wafers that each percent of production actually represents.


Also, given that Nvidia essentially makes more money per product than AMD and that Nvidia dominates the GPU market, especially the high-end GPU market for consumers and businesses, I'd expect TSMC to have whatever capacity that Nvidia thinks they need dutifully set aside. After all, Nvidia has been a much more reliable customer.
I will simply just reply, no company gives a damn about previous business. If you have product you need to sell right now then you get that contract done, you dont wait and hope someone else will buy your production at a later date just because they have been a reliable customer before. Lots of variables come into play, let alone the fact that AMD also gets supplied by Global Foundries for the memory controller, so all TSMC is supplying is the chiplets which are quite small. Will see later which one of us was correct.
 

Jandor

Gawd
Joined
Dec 30, 2018
Messages
520
I will simply just reply, no company gives a damn about previous business. If you have product you need to sell right now then you get that contract done, you dont wait and hope someone else will buy your production at a later date just because they have been a reliable customer before. Lots of variables come into play, let alone the fact that AMD also gets supplied by Global Foundries for the memory controller, so all TSMC is supplying is the chiplets which are quite small. Will see later which one of us was correct.
And Zen 3 will also use Glofo I/O chiplet on 14/12nm tech vs 7nm+ TSMC for multicore and GPU. SO Glofo still has plenty of work from AMD.
 

Ready4Dis

Gawd
Joined
Nov 4, 2015
Messages
730
I will simply just reply, no company gives a damn about previous business. If you have product you need to sell right now then you get that contract done, you dont wait and hope someone else will buy your production at a later date just because they have been a reliable customer before. Lots of variables come into play, let alone the fact that AMD also gets supplied by Global Foundries for the memory controller, so all TSMC is supplying is the chiplets which are quite small. Will see later which one of us was correct.
Lol, you're still arguing with him? Just admit it, the only thing that makes sense is the article you linked is completely wrong, 10% is more than Nvidia could ever need from TSMC and they are just using Samsung for fun... See, much more logical conclusion, now move along. :)
 
Top