AMD Readies Larger 7nm "Navi 12" Silicon to Power Radeon RX 5800 Series?

[ShurikN]

Or may be like the days of Tahiti with a 384 bit bus for 16GB of DDR6.

I was not aware you could run 16 gigs on 384 bit bus. For 16GB they can run either 256 or 512 bit. And the later seem unrealistic as it drives the price up. Thats for GDDR, HBM is a different story.

 

[Kapone33]

I was not aware you could run 16 gigs on 384 bit bus. For 16GB they can run either 256 or 512 bit. And the later seem unrealistic as it drives the price up. Thats for GDDR, HBM is a different story.

I was unaware that you could get 16 on a 256 bit. I think the price may be a non issue. I know if I could buy a card that is 30 to 40% faster than my Vega 64 for what I paid for my first Vega 64 I would have no issue. I do seriously wonder though how HBM2 would effect the performance of a Navi GPU.

 

[Vayra86]

as long as they go to wider memory bus and use 16gbps memory ic's it'll scale nicely

It doesn't scale linearly, and increasing bus widths... need I remind you of what that meant for AMD in the past? A dead end. The writing was already on the wall with the smaller Navi 10. And if they don't pump the clocks, performance will end up lacking.

That is why I'm more interested in actual performance than a bump in specs. History has a tendency to repeat. On paper the 2080ti and a potential Navi upgrade may seem equal, but we're still looking at a node difference to make it happen. Last time AMD used HBM to cover the gap (unsuccessfully) and now its 7nm; this will catch up on them fast.

 

[ShieldHead]

Fanboys are really funny.
Since when do you care for a 50w tdp difference when there is (maybe??) performance to back that up?
Let's all just buy 1030's then

 

[Chrispy_]

Historically, AMD's biggest dies have had codenames in multiples of 10; Polaris 10, 20, 30, Vega 10, 20 are all the largest dies of those particular architectures/generations.

Polaris 11, 12, and 21 were smaller dies, cut-down from 2304SP of the full Polaris chip to 1024 and 640SP

I don't think Navi 12 will be a larger chip - The halo market is vanishingly small and expensive to operate in. AMD are far more likely to be chasing the mass-market where 90% of sales are with a lower-TDP, cheaper-to-make chip that serves the $200 and lower market, as well as being laptop-friendly. I'm only guessing, but I think that *if* AMD make a larger Navi die, it'll be called Navi 20, or given an new codename altogether.

 

[iO]

2080ti performance seems unrealistic and potential market for $1000+ cards is pretty small, so 48CU chip targeting TU104 would make more sense.
Those 20% more shaders and a less crappy cooler could put it right into 2080 territory IF Navi scales well.

 

[Kapone33]

Historically, AMD's biggest dies have had codenames in multiples of 10; Polaris 10, 20, 30, Vega 10, 20 are all the largest dies of those particular architectures/generations.

Polaris 11, 12, and 21 were smaller dies, cut-down from 2304SP of the full Polaris chip to 1024 and 640SP

I don't think Navi 12 will be a larger chip - The halo market is vanishingly small and expensive to operate in. AMD are far more likely to be chasing the mass-market where 90% of sales are with a lower-TDP, cheaper-to-make chip that serves the $200 and lower market, as well as being laptop-friendly.

The current Navi and before that Polaris already cover the market you are talking about. I am sure that there will be cut down cards to replace Polaris once the current stock is gone, but if you don't think AMD will not do it's best to at least catch if not surpass Nvidia's best GPU's with a "halo" release you may not be right.

 

[ShurikN]

I was unaware that you could get 16 on a 256 bit. I think the price may be a non issue. I know if I could buy a card that is 30 to 40% faster than my Vega 64 for what I paid for my first Vega 64 I would have no issue. I do seriously wonder though how HBM2 would effect the performance of a Navi GPU.

Titan RTX has 24GB on 384bit, so 16GB on 256 should be possible. Someone correct me if I'm wrong.

 

[Vya Domus]

Titan RTX has 24GB on 384bit, so 16GB on 256 should be possible. Someone correct me if I'm wrong.

You are right. Each chip requires a 32 bit bus, 384/32 = 12 meaning something like the Titan RTX is using 2 GB memory chips. Therfore 16 GB would need 8 chips or a 8*32 = 256 bit interface.

Not that I believe a large Navi would use something like this, AMD would definitely go for 12 GB on 384 bit for the extra bandwidth.

 

[ArchStupid]

64CU Navi card can be sold for $999 for easy since it will be better then RTX 2080Ti, question is how much better should it be since in this price range the RTRT is more relevant since the 2080Ti can run games with RTX ON much better then the 2060/2070/2080.
AMD can use the current high pricing of the 2080Ti, we need to "thank" Nvidia for that :-(.

Holy delusional batman!

 

[Kapone33]

2080ti performance seems unrealistic and potential market for $1000+ cards is pretty small, so 48CU chip targeting TU104 would make more sense.
Those 20% more shaders and a less crappy cooler could put it right into 2080 territory IF Navi scales well.

Well every tech reviewer on Youtube has one for testing. The 2080TI holds the 9 and 10 position on Amazon's top 10. It even outsells the 2080 according to that

Titan RTX has 24GB on 384bit, so 16GB on 256 should be possible. Someone correct me if I'm wrong.

Interesting that would be using 2GB modules per every 32 bits. So they could do 24GB on 384. Wow when I mentioned Tahiti it came with 3GB of DDR5

 

[Fleurious]

Has AMD said which nVidia card this one will compete with for performance?

 

[HD64G]

Navi with 56 and 64 can have HBM2 since they will be easily sold for $700-800 and the cores are much smaller-cheaper than the one in the 2080Ti. So, from the consumption aspect I don't think there will be any problem. Only thing wondering is the scaling vs the Navi 10. If it is close to 90%, even with lower clocks to increase the efficiency they will get close to the 2080Ti with the full core model.

 

[SIGSEGV]

300w incoming tho, I suspect, wich scares me

meh...
if it turns out able to compete 2080 Ti Super Mega Beyond Your Imagination Huang Edition in terms of price/performance why not?
/funny

 

[ShurikN]

Not that I believe a large Navi would use something like this, AMD would definitely go for 12 GB on 384 bit for the extra bandwidth.

Yup, I was thinking about that as well. 16GB is an overkill, and 8GB is a bit on the low side for a high end GPU like that.

 

[FordGT90Concept]

60 CU/12 GiB/~15 billion transistors which puts it in between TU104 and TU102 makes sense to me.

Problem is, we don't know if Navi is actually scalable like that. Navi 12 might just be 48 CU/8 GiB which would target TU104.

 

[efikkan]

Why Manu, why are you selectively scared ?

He/she have a reason to worry.
And when it comes to power consumption, the main concern for heat, noise, throttling etc. is average consumption, not peak consumption.

RX 5700 XT already consumes ~225W for 40 CUs/2560 SPs, that's going to increase a lot if AMD wants larger dies, or they have to sacrifice a lot of clock speed.

 

[Xaled]

As @Manu_PT says, at what power cost? Nvidia could likely remove power limiters and have hotter, hungrier, noisier cards too. But it's not what the industry wants. Everything is tending to lower power or higher performance at same power.

All what the consumers care about right now is lowering the prices. Nobody would really care about hotter hungrier noisier cards as long as they are not way behind the limits and nobody would care about cards that have the same performance for the same price

60 CU/12 GiB/~15 billion transistors which puts it in between TU104 and TU102 makes sense to me.

This! with a price of TPU104 is all what AMD need to get back into competition. İf AMD succeeded the whole Nivida lineups prices will go down. Nvidia will pay the price for making million model-class just not to lower the prices

 

[Midland Dog]

300w incoming tho, I suspect, wich scares me

2nd card from amd to have a bigger tdp number than die size number scary?

 

[ShurikN]

All what the consumers care about right now is lowering the prices.

Amen to that

 

[nguyen]

My bet Navi 12 is a 300W TDP card priced at 650usd come equipped with a CLC to compete with the RTX 2080Super

 

[Steevo]

I'm going to guess it's a respin like the 1800 to 1900 was, 20+% more performance with better process understanding.

 

[FordGT90Concept]

That would be Navi 20, not Navi 12.

RX 480 = Polaris 10
RX 580 = Polaris 20
RX 590 = Polaris 30

Vega 64 = Vega 10
Radeon VII = Vega 20

 

[dicktracy]

Cool story bro

 

[efikkan]

All what the consumers care about right now is lowering the prices. Nobody would really care about hotter hungrier noisier cards as long as they are not way behind the limits and nobody would care about cards that have the same performance for the same price

When you have two comparable alternatives, like RTX 2060 Super vs. RX 5700 XT, the majority of buyers will go for the one with the overall best deal. RTX 2060 Super costs the same, performs practically the same (especially if we consider partner cards), is more power efficient and much quieter, and the list goes on. For AMD to obtain market share, they need to offset their disadvantages with some equally attractive advantages; like significantly lower prices, 10-15% higher performance, or some useful "must-have" feature, etc. Otherwise RTX 2060 Super vs. RX 5700 XT will end up like GTX 1060 vs. RX 480/580 again; 9 out of 10 will go for the Nvidia option, not because they are fanboys, not because AMD's offer is terrible either, but simply because there is an obvious better offer.