AMD Navi Product Reviews and Previews: (5500, 5600 XT, 5700, 5700 XT)

[Silent_Buddha]

Amd admits that Navi 10 yielded dies are only double the cost of 14nm

https://seekingalpha.com/article/42...esents-semicon-west-conference-slideshow?dr=1

So at least the good news is there Will be plenty of room to lower price in the future... Lol

Interesting slide on the cost per mm^2 in the slide. Likely why NV didn't use 7 nm, and why they may wait longer than most people expect to use 7 nm for the consumer market. The current NV is really focused on high margins and 7 nm would currently eat into that.

The jump in cost from 14 nm to 7 nm is quite significantly larger than any previous node transition it looks like.

I'm not sure if NAVI offers enough competition to convince them to sacrifice margin to go to a significantly more costly node.

It remains to be seen if NV feel threatened by RDNA 2 coming next year.

Regards,
SB

 

[chris1515]

Interesting slide on the cost per mm^2 in the slide. Likely why NV didn't use 7 nm, and why they may wait longer than most people expect to use 7 nm for the consumer market. The current NV is really focused on high margins and 7 nm would currently eat into that.

The jump in cost from 14 nm to 7 nm is quite significantly larger than any previous node transition it looks like.

I'm not sure if NAVI offers enough competition to convince them to sacrifice margin to go to a significantly more costly node.

It remains to be seen if NV feel threatened by RDNA 2 coming next year.

Regards,
SB

Nvidia will go 7nm but with Samsung

https://www.extremetech.com/gaming/294441-nvidia-will-team-up-with-samsung-for-7nm-gpu-technology

https://semiwiki.com/semiconductor-manufacturers/samsung-foundry/7926-samsung-vs-tsmc-7nm-update/

And from here it seems Samsung has the best wafer price of the market

Pricing. Samsung has the best wafer pricing the industry has ever seen. Being the largest memory manufacturer does have its advantages and wafer pricing is one of them.

To catch up with the latest on foundry process technology I talked to Scotten Jones, internationally recognized semiconductor expert and founder of IC Knowledge, a technology consulting company that models the economics of semiconductors. Scott has been writing for SemiWiki since 2014, his blogs are on the IC Knowledge landing page. Here are Scott’s latest thoughts on TSMC versus Samsung at 7nm

 

[PizzaKoma]

https://www.tomshw.de/2019/07/11/un...erplaytables-fuer-die-rx-5700-und-rx-5700-xt/
"Unrestrained: Radeon RX 5700 XT overclocked to over 2.2GHz. Break the limits with the new SoftPowerPlayTables for the RX 5700 and RX 5700 XT"

"The power and resulting power consumption for the entire graphics card and all of its rails is now also evident in the detail graphics. The relevant for the power supply spikes with 10 to 20 ms in length are about 300 to 320 watts, which is therefore also to be taken into account. Everything about it is not relevant, since it is way too short to trigger a shutdown."

AIB cards and 5800-5900 should be interesting to see, this overclock was done with watercooling. Sadly Igor is only presenting data from one game.

 

[Silent_Buddha]

Nvidia will go 7nm but with Samsung

https://www.extremetech.com/gaming/294441-nvidia-will-team-up-with-samsung-for-7nm-gpu-technology

https://semiwiki.com/semiconductor-manufacturers/samsung-foundry/7926-samsung-vs-tsmc-7nm-update/

And from here it seems Samsung has the best wafer price of the market

So, looking at that, sometime at the end of next year at the earliest maybe. Looks like Samsung are desperate to get customers for their 7 nm node since they are late compared to TSMC, so NV might be getting good pricing from them. The drawback, of course, is that it may be quite a while before NVs first 7 nm GPUs come out.

Regards,
SB

 

[Kaotik]

Interesting slide on the cost per mm^2 in the slide. Likely why NV didn't use 7 nm, and why they may wait longer than most people expect to use 7 nm for the consumer market. The current NV is really focused on high margins and 7 nm would currently eat into that.

The jump in cost from 14 nm to 7 nm is quite significantly larger than any previous node transition it looks like.

I'm not sure if NAVI offers enough competition to convince them to sacrifice margin to go to a significantly more costly node.

It remains to be seen if NV feel threatened by RDNA 2 coming next year.

Regards,
SB

Didn't TSMC say somewhere that cost per transistor is already lower than 16FF and yields would reach parity by the end of the year? That should put making the same chip about the same price on either process

 

[Kaotik]

Nvidia will go 7nm but with Samsung

https://www.extremetech.com/gaming/294441-nvidia-will-team-up-with-samsung-for-7nm-gpu-technology

https://semiwiki.com/semiconductor-manufacturers/samsung-foundry/7926-samsung-vs-tsmc-7nm-update/

And from here it seems Samsung has the best wafer price of the market

Yet another site making conclusions by taking things out of context.
NVIDIA did confirm they would use Samsung 7nm, true, but they never said they wouldn't use TSMC 7nm or 7nm+. They even clarified later that TSMC continues to be important partner to them.
NVIDIA has previously used both Samsung and TSMC within same gen, too.

Edit: also, NVIDIA 7nm production at Samsung is apparently scheduled to start 2020

 

[Alexko]

Yet another site making conclusions by taking things out of context.
NVIDIA did confirm they would use Samsung 7nm, true, but they never said they wouldn't use TSMC 7nm or 7nm+. They even clarified later that TSMC continues to be important partner to them.
NVIDIA has previously used both Samsung and TSMC within same gen, too.

Edit: also, NVIDIA 7nm production at Samsung is apparently scheduled to start 2020

To be fair, the bolded part would still be true if NVIDIA went exclusively with Samsung on 7nm, since TSMC is currently making all of their 12nm parts, and most (all?) of their 16/14nm chips as well.

 

[Bondrewd]

cost per transistor

That's cost per xtor, and 7nm dies tend to have many more of these in general.

yields would reach parity by the end of the year?

For something the size of Navi10, yes.
nV's flagships are usually way bigger.

 

[del42sa]

https://www.tomshw.de/2019/07/11/un...erplaytables-fuer-die-rx-5700-und-rx-5700-xt/

2,2Ghz.

Oh, sorry it´s already been posted

 

[Kaotik]

To be fair, the bolded part would still be true if NVIDIA went exclusively with Samsung on 7nm, since TSMC is currently making all of their 12nm parts, and most (all?) of their 16/14nm chips as well.

True, but my point was that one shouldn't draw conclusions from the "we're going to use Samsung 7nm" announcement regarding what NVIDIA does or doesn't do with TSMC

That's cost per xtor, and 7nm dies tend to have many more of these in general.

Of course, point being that at this time at least price shouldn't be the reason to stick to older process anymore, even if you changed nothing else you'd get lower consumption etc without affecting the price

For something the size of Navi10, yes.
nV's flagships are usually way bigger.

Well they said "big dies", didn't refer to any specific size, could be Navi10 size, could be bigger.

 

[Alexko]

True, but my point was that one shouldn't draw conclusions from the "we're going to use Samsung 7nm" announcement regarding what NVIDIA does or doesn't do with TSMC.

Agreed.

 

[xpea]

So, looking at that, sometime at the end of next year at the earliest maybe. Looks like Samsung are desperate to get customers for their 7 nm node since they are late compared to TSMC, so NV might be getting good pricing from them. The drawback, of course, is that it may be quite a while before NVs first 7 nm GPUs come out.

Regards,
SB

Samsung late on 7nm ? Yes and no because, unlike TSMC, the Korean giant decided to skip 7nm DUV and go directly to 7nm EUV. So if we compare 7nm EUV, they are less than 3 months apart on volume production, with the advantage of higher density for Samsung.

 

[Per Lindstrom]

Amd admits that Navi 10 yielded dies are only double the cost of 14nm

https://seekingalpha.com/article/42...esents-semicon-west-conference-slideshow?dr=1

So at least the good news is there Will be plenty of room to lower price in the future... Lol

The die area of Navi is 251 mm^2 and area of Vega is 484 mm^2, so performance and cost for each one is on par.

 

[wearthefoxhat]

 

[Malo]

For the Hardware Unboxed RIS video the tl;dw is that they're very impressed with the result when downscaling a 4k to around 1700-1800p and applying the sharpening filter. In many cases the result was equal to or better than DLSS in several titles with basically no perf loss. Especially when it can be applied to any game immediately rather than having integration by developers then weeks/months of training.

No discussion in video about affect on UI or other aspects of the game and obviously more in-depth analysis by several parties is needed.

 

[snarfbot]

Im pretty sure nvidia confirmed that their next gen cards will be built on tsmc's 7nm.

 

[Ike Turner]

For the Hardware Unboxed RIS video the tl;dw is that they're very impressed with the result when downscaling a 4k to around 1700-1800p and applying the sharpening filter. In many cases the result was equal to or better than DLSS in several titles with basically no perf loss. Especially when it can be applied to any game immediately rather than having integration by developers then weeks/months of training.

No discussion in video about affect on UI or other aspects of the game and obviously more in-depth analysis by several parties is needed.

AMD CAS:

Spoiler

This is going to be extremely useful in the console space..

 

[Deleted member 2197]

Radeon Image Sharpening: I wonder why not enable for non-Navi cards?

But RIS currently has three hooks:

1.) RIS works only with a Radeon RX 5700 XT and Radeon RX 5700; AMD wants to provide the special function with an incentive to buy the new graphics cards.
2.) The Radeon Image Sharpening in the first implementation is not compatible with all program interfaces, currently DirectX 9 only supports the low-level APIs DirectX 12 and Vulkan.
3.) RIS is not offered on every screen, apparently an Ultra HD display is required. In our post-testing with a Full HD LCD connected via HDMI, the option is not available:

https://www.pcgameshardware.de/Rade...pecials/Radeon-Image-Sharpening-Test-1293637/

 

[Ike Turner]

Radeon Image Sharpening: I wonder why not enable for non-Navi cards?

?? Because AMD want's to sell Navi cards? Simple as that. CAS (which is what RIS is) works on GCN and Nvidia GPUs when implemented in game (Rage 2 or any other game or engine who decides to add support for it) but the "free for all" driver option is currently a Navi only feature.

 

[Bondrewd]

Well they said "big dies", didn't refer to any specific size, could be Navi10 size, could be bigger.

https://twitter.com/x/status/1145020125046804480