Nvidia's 3000 Series RTX GPU [3090s with different memory capacity]

Maybe it's just me? but I don't get that same interpretation from briefly looking at the actual transcript (which seekingalpha has).

To me it seems more that it's a statement reflecting that the current supply constraints extend beyond just the chip silicon, which just pushing out more chips wouldn't solve supply.

Also if we want to extend back to the earnings call transcript, which I believe what the question in this case was in the context of, the silicon supply constraint issue was implied to be one involving cycle times/flexibility as opposed to any medium (or longer term) constraint in terms of output.

Could be yields too, these are the first(?) big chips made on the Samsung process after all. And if your yields are bad, you're gonna run out of wafers to meet your planned quotas too.

I believe there is speculation that GA102 has a new stepping that might have improved yields (relatively speaking, not necessarily that current yields are "catastrophic") which might also explain why they might be more conservative in pushing so much production early on. Especially if there is also component restrictions elsewhere as well.
 
Last edited:
Maybe it's just me? but I don't get that same interpretation from briefly looking at the actual transcript (which seekingalpha has).

To me it seems more that it's a statement reflecting that the current supply constraints extend beyond just the chip silicon, which just pushing out more chips wouldn't solve supply.

Also if we want to extend back to the earnings call transcript, which I believe what the question in this case was in the context of, the silicon supply constraint issue was implied to be one involving cycle times/flexibility as opposed to any medium (or longer term) constraint in terms of output.
Here's the link for lazy: https://seekingalpha.com/article/43...-credit-suisse-24th-annual-technology-virtual

Yes, extends beyond just the chips, which implies that even if everything else was normal, they wouldn't really have enough chips either, and vice versa, even if they had more chips they'd be running out of other components like the substrates etc.
tl;dr: there's not enough chips and there's not enough substrates and other components either
 
Yes, extends beyond just the chips, which implies that even if everything else was normal, they wouldn't really have enough chips either

Yup but it also doesn’t tell us anything about yields. With the current feeding frenzy who knows what level of supply would satisfy the market. I’m super skeptical that the supply chain will invest in long term capacity expansion because this covid thing is a short term effect. We just have to wait for demand to fall.
 
Yup but it also doesn’t tell us anything about yields. With the current feeding frenzy who knows what level of supply would satisfy the market. I’m super skeptical that the supply chain will invest in long term capacity expansion because this covid thing is a short term effect. We just have to wait for demand to fall.

At least based on my reading of the information it doesn't necessarily suggest that long term (or even medium term) silicon production itself at Samsung is necessarily the issue, nor does there seem to be an implication that Samsung itself is running at capacity/near capacity (different than the current TSMC situation). Short term supply is problematic due to cycle times. Short term supply injection is also "tricky" due to other component shortages (eg. spending more now just to have GA102 silicon itself sitting around wouldn't solve anything).

In terms of yields I believe there is some data that for instance GA104 ramp up (as in the rate of increase in shipments) is higher than GA102, at least in terms of end products. So there could be "sub optimal" yields with respect to GA102. The new stepping theory and therefore delayed ramp up somewhat supports.

It's worth noting that GA102 would also be only Nvidia's 2nd design at Samsung, and considerably larger than their previous one. I'd also speculate whether or not it would've been the "last" to possibly have been ported over (if Ampere was originally intended as 7nm). So it could be lack of/better understanding on Samsung's node on Nvidia's part as well to improve yields.
 
To understand one of the multiple issues, it also helps to look at availability in other parts of the market. FWIW, low-end dualcore APUs like the Athlon 200GE and its successor Athlon 3000G are hardly available in germany atm. Now, I have not researched of this is a more widespread issue, but it could hint at the supply issues with ABF. I've asked AMD about the Athlon-issue, wanting to know if they're maybe discontinued, but they did not confirm that either.

So, I picked the Athlon because it's neither in high demand from miners of gamers and OTOH production wise is far away from TSMCs or Samsungs sub-10-nm processes.
 
At least based on my reading of the information it doesn't necessarily suggest that long term (or even medium term) silicon production itself at Samsung is necessarily the issue, nor does there seem to be an implication that Samsung itself is running at capacity/near capacity (different than the current TSMC situation). Short term supply is problematic due to cycle times. Short term supply injection is also "tricky" due to other component shortages (eg. spending more now just to have GA102 silicon itself sitting around wouldn't solve anything).

In terms of yields I believe there is some data that for instance GA104 ramp up (as in the rate of increase in shipments) is higher than GA102, at least in terms of end products. So there could be "sub optimal" yields with respect to GA102. The new stepping theory and therefore delayed ramp up somewhat supports.

It's worth noting that GA102 would also be only Nvidia's 2nd design at Samsung, and considerably larger than their previous one. I'd also speculate whether or not it would've been the "last" to possibly have been ported over (if Ampere was originally intended as 7nm). So it could be lack of/better understanding on Samsung's node on Nvidia's part as well to improve yields.
Since you are using the GPU codenames, it would be their 3rd GPU design at Samsung, accounting for the Pascal chips GP107 (GTX1050/Ti) and GP108 (GTX1030/MX150). It would be their 4th fab partner over the years.

Of course, your general point still stands. IIRC, the little Pascal chips were made at Samsung, right as Nvidia and Samsung settled their lawsuits. I always wondered if that was part of the lawsuit settlement or if Nvidia and Samsung's legal teams evaluated each other's IP and saw some opportunities to work with each other.

Also IIRC, AMD has used Samsung (directly) for Polaris, towards the end of the first major mining boom (the one where I purchased a GTX 1050Ti for ~$120, since it was an unique product that came from a non-traditional Nvidia partner, and I questioned if it was good value. One year later, it was almost twice that on the market, if one could even get one). So Samsung isn't totally unawares of how GPU designers work, either.
 
I could see some "synergies" from NVIDIA + Samsung.
Factor in NVIDIA's look at ARM, their IP/Hardware and Samsungs mobile efforts, their manufacturing capabilities.

I could see Samsung phones using ARM CPU, NVIDIA GPU...produced by Samsung.
 
I could see some "synergies" from NVIDIA + Samsung.
Factor in NVIDIA's look at ARM, their IP/Hardware and Samsungs mobile efforts, their manufacturing capabilities.

I could see Samsung phones using ARM CPU, NVIDIA GPU...produced by Samsung.

Will samsung devices finally be able to compete with Apple sillicon by then?
 
Since you are using the GPU codenames, it would be their 3rd GPU design at Samsung, accounting for the Pascal chips GP107 (GTX1050/Ti) and GP108 (GTX1030/MX150). It would be their 4th fab partner over the years.

Of course, your general point still stands. IIRC, the little Pascal chips were made at Samsung, right as Nvidia and Samsung settled their lawsuits. I always wondered if that was part of the lawsuit settlement or if Nvidia and Samsung's legal teams evaluated each other's IP and saw some opportunities to work with each other.

Also IIRC, AMD has used Samsung (directly) for Polaris, towards the end of the first major mining boom (the one where I purchased a GTX 1050Ti for ~$120, since it was an unique product that came from a non-traditional Nvidia partner, and I questioned if it was good value. One year later, it was almost twice that on the market, if one could even get one). So Samsung isn't totally unawares of how GPU designers work, either.

nVidia used Samsung to have more supply. TSMCs 16nm was still in huge demand 2016. And the 14nm process was 10% more dense.
 
No clue, but then again I would never buy an Apple product, I have no clue what the target demography thinks so I will not speculate on it.

Hm, i have both apple and android products. I can appriciate both for different reasons. Regarding performance, they are close enough. Though there is a rather huge difference in video editing/processing times. But those things people rarely do on their phones (for now).

What apple should add to their iphones/pads is this DeX (samsung), in an instant you can transform your phone into a pc/laptop by wireless connection to a TV or any screen. Then processing power matters more i guess.
 
I could see some "synergies" from NVIDIA + Samsung.
Factor in NVIDIA's look at ARM, their IP/Hardware and Samsungs mobile efforts, their manufacturing capabilities.

I could see Samsung phones using ARM CPU, NVIDIA GPU...produced by Samsung.
You forgot that Samsung just licensed RDNA2 for it's SoC GPUs
 
I wonder if that's with SMT turned off on the Ryzen. Yes, the number of CPUs looks like it is, but is the overall load based on the "real CPUs" or the virtual ones?

Also, a lot of sharpening aliasing artifacts on some specular highlights (from 7:05 where V sees the car), also shadowing ugliness (the cop in the smoke). Not sure, if this is due to DLSS perf, DLSS in general oder CP2077 content. My download still is far from finished. :)

edit: Also suspiciously well within an 8 GByte budget at those settings.
 
I wonder if that's with SMT turned off on the Ryzen. Yes, the number of CPUs looks like it is, but is the overall load based on the "real CPUs" or the virtual ones?

Could just be that they configured the OSD to show CPUs and not threads. Even with SMT tuned off 50% of 16 threads is a very heavy load.
 
Yes it is. I hope to see some CPU scaling benchmarks... my hexacore already started to silently weep.
Screenshot-2020-12-10-180351.png

https://www.pcgameshardware.de/Cybe...k-2077-Benchmarks-GPU-CPU-Raytracing-1363331/
 
Back
Top