Navi 5700 die size, and implications.

Snowdog

[H]F Junkie
Joined
Apr 22, 2006
Messages
11,262
Die size estimates have been floating around ranging from about 255mm2 to 275mm2:

https://www.anandtech.com/show/1441...ducts-rx-5700-series-in-july-25-improved-perf
Thanks to our own Andrei Frumusanu for doing the leg work, we now have a die size estimate based on these and some unpublished photos. Our working guess right now is 275mm2, give or take a few percent. Though that's going to be less precise than what a proper micro-caliper measurement will turn up, so this is very much still just an estimate.


https://wccftech.com/amd-navi-rx-57...gs-rtx-2070-performance-at-half-the-die-size/
Earlier today we’ve managed to confirm that the company’s upcoming Navi GPU, which was unveiled for the first time earlier this week and is set to power the company’s upcoming RX 5700 series graphics cards, is in fact only mid-range Navi with a die area of roughly 255mm².

The Wccftech story seems to imply that close to half the die size of 2070 will be substantially cheaper, but really it will probably much closer than that. 7nm is a much more expensive process, so things are closer than they appear. Cost per transistor is better for 7nm, but not by a huge amount. The important thing, is how many transistor have we to work with.

Using Radeon 7 on TSMC 7nm process and 13.2Billion transistors and 331mm2 die size in conjunction with the above estimates.

13.2/331x255 = ~10.2 Billion Transistors
13.2/331x275 = ~11.0 Billion Transistors

Compare to RTX 2070 die at 10.8 Billion. (1660ti at 6.6 Billion).

Either way transistor count is very close to 2070, so die cost should be a little less than 2070, but not substantially so.

Bottom line:
AMD is in a much better position than they were with Vega 64 needing substantially more transistors (12.5 Billion vs 7.2 billion ) than GTX 1080, along with more expensive HBM memory to match performance.

But it is more parity, than big advantage that Wccftech implies.
 
Die size estimates have been floating around ranging from about 255mm2 to 275mm2:

https://www.anandtech.com/show/1441...ducts-rx-5700-series-in-july-25-improved-perf



https://wccftech.com/amd-navi-rx-57...gs-rtx-2070-performance-at-half-the-die-size/


The Wccftech story seems to imply that close to half the die size of 2070 will be substantially cheaper, but really it will probably much closer than that. 7nm is a much more expensive process, so things are closer than they appear. Cost per transistor is better for 7nm, but not by a huge amount. The important thing, is how many transistor have we to work with.

Using Radeon 7 on TSMC 7nm process and 13.2Billion transistors and 331mm2 die size in conjunction with the above estimates.

13.2/331x255 = ~10.2 Billion Transistors
13.2/331x275 = ~11.0 Billion Transistors

Compare to RTX 2070 die at 10.8 Billion. (1660ti at 6.6 Billion).

Either way transistor count is very close to 2070, so die cost should be a little less than 2070, but not substantially so.

Bottom line:
AMD is in a much better position than they were with Vega 64 needing substantially more transistors (12.5 Billion vs 7.2 billion ) than GTX 1080, along with more expensive HBM memory to match performance.

But it is more parity, than big advantage that Wccftech implies.

Just not using HBM alone will cut the costs quite a bit, but I dont think either company is racing to a price war yet.
 
Let's see... half the die size and same performance or more of 2x the size 2070... I'd say rdna is going to be awesome.
 
Let's see... half the die size and same performance or more of 2x the size 2070... I'd say rdna is going to be awesome.

Die size really can't be compared on radically different processes these days. Each process shrink drives up the cost so much, that it isn't like a decade ago, where each shrink is a big economic win.

These days process shrinks only result is small drops in cost/transistor and these designs with similar transistor counts will have similar production costs. It's the relative cost/unit that reflects cost/transistor and those are flat lining.

main-qimg-4bed7def49e8bb0a8e9a596c79a093e3-png.png


Also it while it looks like approximate parity in transistor counts/performance here, remember that NVidia is devoting a significant amount of that transistor budget to RT HW that doesn't factor into traditional Raster performance. So it's Raster design still appears more efficient. Though this is just running on the AMD claims and guesstimates right now.


Bottom Line: AMD, no longer has the massive handicap that it suffered with Vega, but it really hasn't leapfrogged NVidia either. It's just a lot closer to parity.
 
Let's see... half the die size and same performance or more of 2x the size 2070... I'd say rdna is going to be awesome.
Awesome in what way?
For GPU that comes year after it should be considerably faster for that same price and this ain't happening so there is nothing awesome to wait for
 
Awesome in what way?
For GPU that comes year after it should be considerably faster for that same price and this ain't happening so there is nothing awesome to wait for

Going by that logic the last two sets of nvidia releases were also crap. The 2080 costs as much as a 1080ti, and isnt much of a performance upgrade at all.
 
Awesome in what way?
For GPU that comes year after it should be considerably faster for that same price and this ain't happening so there is nothing awesome to wait for

Hey you only have consumers to blame. AMD competed on price with Polaris and people just bought cheaper nvidia cards lol. I am sure AMD has the internal data, they probably see no benefit to reducing price and sell their cards cheap if they sell close to same amount. Or if the overall revenue will be the same. They just end up looking like a cheap brand and people still pay more for nvidia.

I am sure they will price these competitively but I don't expect them to hand 2070 performance for 250. Adoredtv should have tasted the salt when the was mentioning those rumors. I was laughing out loud, oh yea AMD is going to sell you fresh 7nm die at half the price of Nvidia and same performance. That is definition of insanity.

https://www.cnet.com/news/samsung-and-amd-partner-up-for-mobile-gpus/

"Terms include Samsung paying AMD licensing fees and royalties for its Radeon technology, and AMD licensing custom graphics intellectual property based on RDNA architecture."
 
Last edited:
Going by that logic the last two sets of nvidia releases were also crap. The 2080 costs as much as a 1080ti, and isnt much of a performance upgrade at all.
Opinions on Turing were rather negative
Let's wait and see if Navi is any better than Pascal/Turing...
 
Hey you only have consumers to blame. AMD competed on price with Polaris and people just bought cheaper nvidia cards lol. I am sure AMD has the internal data, they probably see no benefit to reducing price and sell their cards cheap if they sell close to same amount. Or if the overall revenue will be the same. They just end up looking like a cheap brand and people still pay more for nvidia.
AMD cannot expect consumers to be interested if they do not exist in top end, especially for so long and with such big gap between fastest cards. Card beating 2080Ti is what AMD really needs and they openly said they are going only to compete with 2070 this time. I do not see how they will convince people to rush to the stores and rebuild their low market share, even if there are people who will buy AMD card no matter what. There are more such one-sided people on Nvidia side anyway...
 
AMD cannot expect consumers to be interested if they do not exist in top end, especially for so long and with such big gap between fastest cards. Card beating 2080Ti is what AMD really needs and they openly said they are going only to compete with 2070 this time. I do not see how they will convince people to rush to the stores and rebuild their low market share, even if there are people who will buy AMD card no matter what. There are more such one-sided people on Nvidia side anyway...

A non enthusiast does not care who has the fastest card as they cant afford it anyway. The sub 300 market only cares to get the fastest card they can for the money regardless of who made it, that is the bulk of the market. The high end we enthusiast care about is where they make most of their money and thus why Nvidia margins are good but volume is way down. Also AMD is competing with the 2080 and down which is 98% of the market so only 2% is left out due to not having a 2080ti competitor. So for market share it's all about the sub 300 market the high end is far too limited to actually effect market share much at all. Problem for both sides will remain the same, if you bought a Pascal or equivalent card and are not running 4K then there is little reason for you to want to upgrade anyway as your performance is more then good enough.
 
A non enthusiast does not care who has the fastest card as they cant afford it anyway. The sub 300 market only cares to get the fastest card they can for the money regardless of who made it, that is the bulk of the market. The high end we enthusiast care about is where they make most of their money and thus why Nvidia margins are good but volume is way down. Also AMD is competing with the 2080 and down which is 98% of the market so only 2% is left out due to not having a 2080ti competitor. So for market share it's all about the sub 300 market the high end is far too limited to actually effect market share much at all. Problem for both sides will remain the same, if you bought a Pascal or equivalent card and are not running 4K then there is little reason for you to want to upgrade anyway as your performance is more then good enough.

I would disagree. I recently had to buy a new washing machine, and so I looked at who owned the top end and then bought the highest model of that brand that I could afford. The logic/belief being that even the lower models would have benefited from the R&D put into the high end.

But maybe I did that because I'm a computer hardware enthusiast! :p
 
I would disagree. I recently had to buy a new washing machine, and so I looked at who owned the top end and then bought the highest model of that brand that I could afford. The logic/belief being that even the lower models would have benefited from the R&D put into the high end.

But maybe I did that because I'm a computer hardware enthusiast! :p

I did the same thing decades ago when I was into cassette tapes. The 'best' on the market tape deck at the time was the Nakamichi Dragon but 2000 USD for a tape deck was a rather hard swallow. I ended up getting the 'cheapest' of that lineup (which was still an insane 500 USD) as it was clear from listening that the quality was still there.
 
Yeah the Nakamichi Dragon was sweet, but i also dident get it ended up with a Akai GX-R99 and a ton ( almost literally ) metal TDK tapes.

Reading Samsung also jumping on the AMD band wagon, do make me think AMD have good things on offer.
 
AMD cannot expect consumers to be interested if they do not exist in top end, especially for so long and with such big gap between fastest cards. Card beating 2080Ti is what AMD really needs and they openly said they are going only to compete with 2070 this time. I do not see how they will convince people to rush to the stores and rebuild their low market share, even if there are people who will buy AMD card no matter what. There are more such one-sided people on Nvidia side anyway...
Actually if AMD can clearly show their $300 part is way faster/better than Nvidia $300 part and even AMD $399 part better than Nvidia $499 part - most folks will probably conclude Nvidia $1200 part is just another rip off from Nvidia. Someone looking at $200-$400 parts would also probably laugh at the 2080Ti price as well. I really think it is irrelevant who has the faster card. If AMD does good marketing, helps developers and gets the cards out so customers can easily obtain them they will do well. AMD suck pretty bad the whole Vega launch from marketing and getting the cards available to buy. Nvidia does a much better job in getting the cards available and in more places.
 
Actually if AMD can clearly show their $300 part is way faster/better than Nvidia $300 part and even AMD $399 part better than Nvidia $499 part - most folks will probably conclude Nvidia $1200 part is just another rip off from Nvidia. Someone looking at $200-$400 parts would also probably laugh at the 2080Ti price as well. I really think it is irrelevant who has the faster card. If AMD does good marketing, helps developers and gets the cards out so customers can easily obtain them they will do well. AMD suck pretty bad the whole Vega launch from marketing and getting the cards available to buy. Nvidia does a much better job in getting the cards available and in more places.

nVidia has definitively moved the pricing expectations for the utlra high-end part i.e. 2080 ti in consumers minds (they have been trying to move that goal post for years with the introduction of the Titan series). It's disturbing and sad, but people expect a ridiculously (objectively-considered) overpriced "ultra" part at this point. An AMD card that performs 10% better than nV's in the $300 performance segment is not going to change any hearts and minds on the top end.
 
I did the same thing decades ago when I was into cassette tapes. The 'best' on the market tape deck at the time was the Nakamichi Dragon but 2000 USD for a tape deck was a rather hard swallow. I ended up getting the 'cheapest' of that lineup (which was still an insane 500 USD) as it was clear from listening that the quality was still there.

Coincidentally, I was at a vintage audio swap last week and spotted a Nakamichi Dragon (the father-in-law and I are building an old ferograph model he designed years back at Thorn and were scouting for remaining parts). I was unaware they were so sought-after at the time, but he kept going on and on about "that was the one to have..." For me, it was an NAD single disc player that fit the profile of "they have the best, so buy the cheapest model").

However, I do think that while having the flagship part may well influence enthusiasts and it helps with branding, it's really not going to affect sales in mass. Most people are looking for best price/performance, especially in the mid-range -- and most people are looking in the mid-range.
 
Card beating 2080Ti is what AMD really needs

Until very recently AMD has sold every card they built.
What AMD really needs is a card that the market wants to buy, for mining, for machine learning or for gaming.
AMD built a card that is more or less competition to the RTX 2080.
Every one of them was sold hours after went on sale.

Tell me again why AMD business model is in risk if they can not show a 2080ti competitor?
 
Actually if AMD can clearly show their $300 part is way faster/better than Nvidia $300 part and even AMD $399 part better than Nvidia $499 part - most folks will probably conclude Nvidia $1200 part is just another rip off from Nvidia.

How does this conclusion follow logically? Especially when AMD has no part that can compete with the $1200 part, not even close?
 
nVidia has definitively moved the pricing expectations for the utlra high-end part i.e. 2080 ti in consumers minds (they have been trying to move that goal post for years with the introduction of the Titan series). It's disturbing and sad, but people expect a ridiculously (objectively-considered) overpriced "ultra" part at this point. An AMD card that performs 10% better than nV's in the $300 performance segment is not going to change any hearts and minds on the top end.
Actually AMD does not have to do anything special for the top end because those people will buy anything as long as it is faster.

How does this conclusion follow logically? Especially when AMD has no part that can compete with the $1200 part, not even close?
Target market is bigger then the $1200 one.
 
Actually AMD does not have to do anything special for the top end because those people will buy anything as long as it is faster.

Exactly. If AMD put out a power monster that was technically the fastest, with a price to match, it would fucking sell.

However, AMD knows Nvidia is simply bigger and would ultimately win any aggressive arms race. It would just cost them too much to try and tick that 'bestest' box. They realise they just need to be profitable all the time, and they do that by making and selling precise amounts of cards. They didn't have a mining related inventory problem like Nvidia did as far as I'm aware.

Nvidia just flood the market wildly as they command the top end. The scrappy 20 series launch reflects this.
 
Exactly. If AMD put out a power monster that was technically the fastest, with a price to match, it would fucking sell.

However, AMD knows Nvidia is simply bigger and would ultimately win any aggressive arms race. It would just cost them too much to try and tick that 'bestest' box. They realise they just need to be profitable all the time, and they do that by making and selling precise amounts of cards. They didn't have a mining related inventory problem like Nvidia did as far as I'm aware.

Nvidia just flood the market wildly as they command the top end. The scrappy 20 series launch reflects this.

They did but no where near as bad as Nvidia did.
 
Until very recently AMD has sold every card they built.
What AMD really needs is a card that the market wants to buy, for mining, for machine learning or for gaming.
AMD built a card that is more or less competition to the RTX 2080.
Every one of them was sold hours after went on sale.

Tell me again why AMD business model is in risk if they can not show a 2080ti competitor?
Selling out stock is not what generates a ton of profit for a company if you are selling not a lot of it and without much of a margin.
They need to increase both margins and sales but for that to work they need to have competitive products. If product is not competitive only die hard fans will buy it and everyone else will ignore it including people who currently own their product. And thing with changing brands is that once you do it it is harder to not get another product from the same brand. This is why with current AMD discrete GPU shares they should be actually striving to have something extra competitive, better than competition, not merely match it in performance and demand the same price for it despite not even matching it in other aspects.

Radeon VII is more of a peculiarity than actual 2080 competition and if anything its pathetic performance shows that another Fiji rehash based on 7 years old GCN design is dead end.
I would be severely shocked if it did not sell given its memory size and amount of AMD fans compared to amount of cards they put on the shelves. It is not bad card in any way and have some advantages like HDMI VRR support so it was destined to sell. If AMD flooded market with it then it would only acquire dust on the shop and storage space shelves ...
 
it would only acquire dust on the shop and storage space shelves

You mean like NVIDIA did with their "successful" 10xx series?:ROFLMAO:

AMD was at less than 0.01% of the server CPU market.
It will end the decade above 10%.
Consumer CPU market? Already ahead of Intel on sales.

GPU market has competition from $100 to $700.
Navi will just guarantee that this competition continues at that price range.
$1000 consumers GPU are a niche market.

I don't care to buy (only) AMD cards.
I do care to shop ate a price range where there is competition and choice.
 
Back
Top