Navi Rumors

Status
Not open for further replies.
Nvidia has the GTX1660, and has access to the same processes that AMD does...

Not even talking about that. 1660 is going to be cheaper 1060 + 10% or so. My point was on RTX series. Never even mentioned the standard Turing cards.
 
Not even talking about that. 1660 is going to be cheaper 1060 + 10% or so. My point was on RTX series. Never even mentioned the standard Turing cards.

If we're talking about the DXR-less Navi GPU, the RTX-less Turing GPUs are clearly in play.
 
Says the man who doesn't understand semiconductor physics.

Larger dies tend to have vastly better yields on older process nodes, which was why Nvidia has released groundbreaking huge chips like the 8800 GTX and the RTX 2080 Ti on ancient processes nodes.

AMD only jumped the gun on 7nm shrink because that was cheaper for them to redesign the architecture to be 30% more efficient (see the 1660 Ti). Unfortunately, they have to take the piss-poor yields with that.

Remember how 14nm had been used in like TWO generations of cell chips before both AMD and Nvidia committed? It's the same thing here, except AMD has been too disparate to increase Pro Performance. That is going to kill any margins.

You don't STAND STILL , but you don't jump into the pool full of hungry sharks either. There are several rumors stating a 2019 or 2020 7nm release from Nvidia.
 
If we're talking about the DXR-less Navi GPU, the RTX-less Turing GPUs are clearly in play.

Man I purely discussing the die size. I don't have a horse the race. If navi comes in below 300 and gives you close to vega performance. It's because AMD can afford to do it due to die size. That's all I am saying.
 
Talking about die sizes. Not just process. RTX series is expensive because they are larger even if 12nm is cheaper at this point. AMD will Be able to harvest more dies from 7nm even if it's more expensive.

RTX series is more expensive to make. It's pretty common knowledge at this point.

You have to use some common sense.

If NVIDIA can take Turing and die shrink it to 7nm to make it cheaper, NVIDIA would have already done it.

Clearly, at the moment, it's cheaper to have a bigger die on an older process than a smaller die on a newer process.
 
You have to use some common sense.

If NVIDIA can take Turing and die shrink it to 7nm to make it cheaper, NVIDIA would have already done it.

Clearly, at the moment, it's cheaper to have a bigger die on an older process than a smaller die on a newer process.


So nvidia should have mass produced 7nm when it wasn't ready last September? Yeah common sense. 12nm is mature process doesnt mean those dies are cheap. If nvidia could have made them on 7nm they would have. Problem is it wants ready for prime time. Why keep debating this RTX chips are big and expensive even every report has said that and it reflects in their pricing. When they go to 7nm it will be better but clearly it wasn't an option, in 2020 they will.
 
Last edited:
So nvidia should have mass produced 7nm when it wasn't ready last September? Yeah common sense. 12nm is mature process doesnt mean those dies are cheap. If nvidia could have made them on 7nm they would have. Problem is it wants ready for prime time. Why keep debating this RTX chips are big and expensive even every report has said that and it reflects in their pricing. When they go to 7nm it will be better but clearly it wasn't an option in 2020 they will.

Then how come Apple can do it?
 
Then how come Apple can do it?

mobile chips vs discrete GPUs? I thought you were better than this lol. Why are you so hell bent on debating this. now comparing mobile chips to discrete GPUs. Common!
 
And what are you proposing?

...for AMD to go into a price war with NVIDIA?
Possibly.

What I am saying is that they need something that is either faster than Nvidia, cheaper, better in some tangible way (efficiency, etc.), or all of those things.

Nvidia is just too dominant for AMD to release another card that is maybe equal to Nvidia's last gen with the same prices.
 
Point is, TSMC's 7nm was clearly available.

Point is its not the same process, 7nm has different tiers. You can't compare apple pushing mobile chips to discrete GPUs common now. One is high performance node the other is mobile chips. Much easier to make then high end GPUs.

Lets not debate this anymore. Never thought you would use apple pushing mobile chips as the same equivalent of Nvidia designing high-performance chips on 7nm.
 
Possibly.

What I am saying is that they need something that is either faster than Nvidia, cheaper, better in some tangible way (efficiency, etc.), or all of those things.

Nvidia is just too dominant for AMD to release another card that is maybe equal to Nvidia's last gen with the same prices.

You should read about price wars esp. the part about what happens to smaller, more marginal firms.
 
Possibly.

What I am saying is that they need something that is either faster than Nvidia, cheaper, better in some tangible way (efficiency, etc.), or all of those things.

Nvidia is just too dominant for AMD to release another card that is maybe equal to Nvidia's last gen with the same prices.

I think AMD isn't worried too much about GPU side. It will get better in time. As long as they keep executing on Zen 2 and keep building on their server share they will be plenty happy. Eventually those two are going to help the GPU side.
 
Point is its not the same process, 7nm has different tiers. You can't compare apple pushing mobile chips to discrete GPUs common now. One is high performance node the other is mobile chips. Much easier to make then high end GPUs.

Regardless, NVIDIA has access to the same process as AMD does.

If (and that's a big if) 7nm is such a huge advantage as you make it out to be...

Do you really think that NVIDIA would let AMD have that advantage all to itself?
 
Regardless, NVIDIA has access to the same process as AMD does.

If (and that's a big if) 7nm is such a huge advantage as you make it out to be...

Do you really think that NVIDIA would let AMD have that advantage all to itself?


AMD will have 7nm chips before them, that's a fact. Again my point is this. Turing is expensive because those big dies are expensive. Nvidia released it on 12nm because they could price it high. If they had the capacity on 7nm and it was already mature they would have gone for it. Because the cost savings and being able to harvest more dies would have been fine since they were going to price this shit sky high anyways.

They didn't do it because 7nm wasn't ready for primetime for high end GPUs last year in September and looks like TSMC was pretty booked. Reason Nvidia is leaning on Samsung for 7nm as the rumors suggest.

Anyways cheers. We can agree to disagree.
 
AMD will have 7nm chips before them, that's a fact. Again my point is this. Turing is expensive because those big dies are expensive. Nvidia released it on 12nm because they could price it high. If they had the capacity on 7nm and it was already mature they would have gone for it. Because the cost savings and being able to harvest more dies would have been fine since they were going to price this shit sky high anyways.

They didn't do it because 7nm wasn't ready for primetime for high end GPUs last year in September and looks like TSMC was pretty booked. Reason Nvidia is leaning on Samsung for 7nm as the rumors suggest.

Anyways cheers. We can agree to disagree.


Evidently that's not actually true now due to the maturity of the process. The yield for 12nm silicon is much higher than 7nm, so 7nm chips even with the smaller die size are actually more expensive than nvidia's larger chips. Nvidia is charging more money because they can get away with it, who's to stop them?
 
AMD will have 7nm chips before them, that's a fact. Again my point is this. Turing is expensive because those big dies are expensive. Nvidia released it on 12nm because they could price it high. If they had the capacity on 7nm and it was already mature they would have gone for it. Because the cost savings and being able to harvest more dies would have been fine since they were going to price this shit sky high anyways.

They didn't do it because 7nm wasn't ready for primetime for high end GPUs last year in September and looks like TSMC was pretty booked. Reason Nvidia is leaning on Samsung for 7nm as the rumors suggest.

Anyways cheers. We can agree to disagree.

I can tell you right now that yield from a brand new process is going to be worse than a matured process.
 
Evidently that's not actually true now due to the maturity of the process. The yield for 12nm silicon is much higher than 7nm, so 7nm chips even with the smaller die size are actually more expensive than nvidia's larger chips. Nvidia is charging more money because they can get away with it, who's to stop them?

That's not a proven fact though. Yields will be fine on 7nm. Please present data that 7nm yeids will be less. And chips are more expensive if Turing was to be shrunk to 7nm. So far only data and facts available are Turing is expensive because it has the largest dies ever made. May be they will be expensive but if it was true AMD would'nt make navi on 7nm. That is a sign of mature process.
 
So AMD is not launching 7nm zen parts mid year? If yields were so bad it wouldn't be happening.

That's a red herring

The 7nm die used in Zen 2 is tiny compared to, let's say, the 7nm die used in Radeon VII

Also, GlobalFoundries's 14nm is far worse than Intel's 14nm, so AMD would want to move to a better process to be more competitive
 
Last edited:
That's a red herring

The 7nm die used in Zen 2 is tiny compared to, let's say, the 7nm die used in Radeon VII

Lol weren't you the one who said Nvidia could do it if apple could do it. So apple isn't making tiny chips?

As far as radon 7 goes. Yeilds seem fine. They have been making the 7nm instinct card for months now. Seems overblown for yield issues. It's not a giant die despite the gcn inefficiency in games.
 
Lol weren't you the one who said Nvidia could do it if apple could do it. So apple isn't making tiny chips?

Yes, NVIDIA could.

Yield would be bad and price would be high

As far as radon 7 goes. Yeilds seem fine. They have been making the 7nm instinct card for months now. Seems overblown for yield issues. It's not a giant die.

AMD Radeon Instinct MI60/MI50 sell for thousands of dollars.

AMD is hardly making a profit with Radeon VII.
 
Last edited:

Yes, NVIDIA could.

Yield would be bad and price would be high



AMD Radeon Instinct MI60/MI50 sell for thousands of dollars.

AMD is hardly making a profit with Radeon VII.

Please prove yields are bad on 7nm. Let's not keep switching and throwing things at the wall. There is no indication yields are horrible.
 
Man I purely discussing the die size. I don't have a horse the race. If navi comes in below 300 and gives you close to vega performance. It's because AMD can afford to do it due to die size. That's all I am saying.

But that is the point- if AMD has pricing freedom due to node size, Nvidia has the same.

RX480/580 (Polaris) is 230mm2, GTX1060 (GP106) is 200mm2. Why would AMD have an advantage?

You do realize that they're usually less space and power effecient for gaming, right?
 
But that is the point- if AMD has pricing freedom due to node size, Nvidia has the same.

RX480/580 (Polaris) is 230mm2, GTX1060 (GP106) is 200mm2. Why would AMD have an advantage?

You do realize that they're usually less space and power effecient for gaming, right?

Never debated they are less efficient. But I am not going to throw navi under the bus already. Radeon 7 is just a shrink while navi seems to have bigger tweaks and designed around 7nm in mind. I am betting its lot more efficient then what we have seen recently from nvidia.
 
..navi seems to have bigger tweaks and designed around 7nm in mind. I am betting its lot more efficient then what we have seen recently from nvidia.

It looks like an upsized and shrunk Polaris; that's not inherently bad, but AMD has a really bad record of 'optimizations' getting more than a linear boost in performance. I'm not writing Navi off either; maybe AMD will break their decade-long trend, maybe not.
 
Please prove yields are bad on 7nm. Let's not keep switching and throwing things at the wall. There is no indication yields are horrible.

How do I do that?

Are you expecting some top secret information that's not publicly available?

Obviously, there's no way that that can be known for sure.

Like many things in life, you have to observe and make inference.
 
Last edited:
Please reference the topic before posting again.
 
It looks like an upsized and shrunk Polaris; that's not inherently bad, but AMD has a really bad record of 'optimizations' getting more than a linear boost in performance. I'm not writing Navi off either; maybe AMD will break their decade-long trend, maybe not.

I think navi is suppose to be alot more leaner than vega. More work seems to have gone in to it over the years and rumor about it doing Variable rate shading. Rumor about more engineers devoted to Navi with Sony. May be will finally make something that will be efficient at gaming not only when used for compute.
 
I think navi is suppose to be alot more leaner than vega. More work seems to have gone in to it over the years and rumor about it doing Variable rate shading. Rumor about more engineers devoted to Navi with Sony. May be will finally make something that will be efficient at gaming not only when used for compute.

Well, it is leaner; what we're speculating on is whether it's leaner than Polaris, and whether that will make Navi more competitive not just compared to what we have now, but also compared to what Nvidia has coming and hell what Intel has coming.
 
Well, it is leaner; what we're speculating on is whether it's leaner than Polaris, and whether that will make Navi more competitive not just compared to what we have now, but also compared to what Nvidia has coming and hell what Intel has coming.

I would be truly surprised if Intel has something to match the competition first go around. May be second time they might have something really impressive when Raja has had more time.
 
I would be truly surprised if Intel has something to match the competition first go around. May be second time they might have something really impressive when Raja has had more time.

I'm surprised that they haven't yet. Making effective iGPUs seems like it would be harder than discrete GPUs, where you can throw as much power, die size, and memory bandwidth as you'd like at the problem.
 
Status
Not open for further replies.
Back
Top