(Updated) AMD Navi GPU to Offer GTX 1080 Class Performance at ~$250 Report Claims

[Valentyn]

Just now, M.Yurizaki said:

I understand that sentiment too, but I'll just keep pulling my "Core i7 is an entry level processor" card out. Even Xeons didn't have way more cores than the consumer version until recently-ish

Very true; "back in my day" the i7 was the top end only 3-4 at most of them haha.

Good thing I don't need a tonne of CPU cores anymore, work has shifted to GPU stuff. Will hold out until Icelake and Zen 2 before upgrading there.

[Razor01]

23 minutes ago, Drak3 said:

Comparison wise, it was always 980 for a single 480, 980Ti/1080 for Crossfire. Outside of gaming and CUDA, the 480 is quite better than the 1060.

 

 

Not really even in mining where it does beat out a 1060, the 1070's same hash rates (actually a bit higher by 15% 30 mhs vs 34 mhs) and much better power consumption levels.  We are looking at 90 to 100 watts for the rx480 and the 1070; 80 to 90 watts.  That is because of the Tflops but the 1070 has the same level of flops.  So......nV is more efficient there too.

 

If the 1060's came out before or at the same time as the rx480, the reviews would not have given any props to the rx480.  All those gold awards or recommended awards would not even be there.

[Razor01]

39 minutes ago, Valentyn said:

Never said it was locked, it's just the smaller cut down one.

 

The GTX 580 also had those components; though; just like the Kepler Titan. The Titan was just faster at FP64; but overall Fermi had decent FP 64 compute also. More so than most Kepler parts outside of the Titan line.

 

 

FP 64 performance

GTX 580 1/8 

GTX 680 1/24 

780Ti 1/24

Titan 1/3 

Titan X 1/32  

Titan Xp 1/32 

Titan V 1/2

 

NVIDIA introduced a new segment under a new name; but it already existed as the x80 in the past. Silly buggers like myself just craving top performance kept buying them anyway; fueling it all.

the 1/3 and 1/2 Titans are full speed they are equivalent to the Tesla line of cards.

 

GTX 580 for gaming had laser cut DP units, they were full 100 (or what ever they were called at the time) dies but were cut down. The 1/24, 1/32 ones were the same dies used in the tesla line but they didn't have many DP units at all, DP units take up too much die space and for those nodes just weren't made.

 

So far the only "full" chips we have seen has been the first Titan and recently Titan V which has a cut down memory bank.

[Valentyn]

10 minutes ago, Razor01 said:

the 1/3 and 1/2 Titans are full speed they are equivalent to the Tesla line of cards.

 

GTX 580 for gaming had laser cut DP units, they were full 100 (or what ever they were called at the time) dies but were cut down. The 1/24, 1/32 ones were the same dies used in the tesla line but they didn't have many DP units at all, DP units take up too much die space and for those nodes just weren't made.

 

So far the only "full" chips we have seen has been the first Titan and recently Titan V which has a cut down memory bank.

Even the new Titan V doesn't have all the ROPs available in GV100; the Quadro GV100 has more ( 96 vs 128 ); and even a higher boost clock out of the box ( 1445 vs 1627 )

So we might even get a new Titan Vx or something later.

[Razor01]

4 minutes ago, Valentyn said:

Even the new Titan V doesn't have all the ROPs available in GV100; the Quadro GV100 has more ( 96 vs 128 ); and even a higher boost clock out of the box ( 1445 vs 1627 )

So we might even get a new Titan Vx or something later.

 

ROPs are connected to the memory bank

[Valentyn]

6 minutes ago, Razor01 said:

 

ROPs are connected to the memory bank

Yup; and it sucks. So like I said we can easily expect a bigger Titan down the road to get all the goodies available. ( Also a better shroud, the current one is meh compared to Maxwell Titan X, and the GV100 Quadro )

 

I got lucky at least, and my card boosts well enough ( 1747Mhz ); I need some modern games; because the GPU load is always 50% or less.

Maybe the new Metro game with RTX in will melt it.

 

[Razor01]

it depends they cut the memory bank because there probably was an error in I/O part of the silicon.  I don't consider the Titan V the same card as the GTX Titan tier.  I don't even think we will see another GTX Titan this time around at least not with a GV100, maybe with a gx102.

[Valentyn]

5 minutes ago, Razor01 said:

it depends they cut the memory bank because there probably was an error in I/O part of the silicon.  I don't consider the Titan V the same card as the GTX Titan tier.  I don't even think we will see another GTX Titan this time around at least not with a GV100, maybe with a gx102.

 

Me might if Tensor cores for Gameworks becomes big. The GV100 has 672, vs Titan V 640 also. ( 5376 vs 5120 CUDA cores )

I doubt they'll strip those out for a GV102 Variant since gameworks will be using those.

Otherwise they might just keep the Titan V, and leave it until the next generation when they're on 7nm

[Razor01]

11 minutes ago, Valentyn said:

 

Me might if Tensor cores for Gameworks becomes big. The GV100 has 672, vs Titan V 640 also. ( 5376 vs 5120 CUDA cores )

I doubt they'll strip those out for a GV102 Variant since gameworks will be using those.

Otherwise they might just keep the Titan V, and leave it until the next generation when they're on 7nm

I don't know how the tensor cores are set up, they might be able do the same thing as cutting down the amounts of DP units for the gx102 parts.  Yeah Titan V has  full extra SM blocked off too

[minibois]

Sounds like a similar claim they had with the RX 580. It was said to have GTX 980 performance for.. Also 250 USD MSRP!

[Kamina]

I hope the hype is real but I also heard Navi will be shit.

[Space Reptile]

On 13.4.2018 at 1:15 PM, GoldenLag said:

some that has looked at Nvidia in the past consider the 1070/1080 to be mid end and the 1080ti and Titan to be high end. i myself consider the 1080 to be mid end. the Vega 64/liquid is AMD`s high end. In performance you might consider it to be and high end card, but in reality it is just a mid-end card. Then again that wasnt the point i was making. also i knew someone was going to comment on me calling it mid-end. 

so you consider the GTX 1060 and RX X80 series as low end? 

[Drak3]

On 4/14/2018 at 11:59 AM, Minibois said:

Sounds like a similar claim they had with the RX 580. It was said to have GTX 980 performance for.. Also 250 USD MSRP!

Well, outside of the overpricing on launch and the overpricing due to outrageous demand, it was true.

[johnukguy]

Given that it's WCCF Tech, I won't hold my breath waiting to see if this is true. If it is, it would literally be a game changer, but I cannot see Radeon, which is the division responsible and who are utterly incompetent, changing anytime soon in order to help gamers out of the goodness of their hearts. Then again, I really really hope that I'm wrong on this one.

[rawrdaysgoby]

NAVI+RYZEN 2 moblie CHIP HYPE BOTH ON 7NM Will blow away Intel and NVDA combo doesn't matter what anyone says Intel still hasn't announce their 10nm chips yet Nvda doesn't seem to be moving towards 10nm yet too.  2019 HYPE

[RotoCoreOne]

I guess I can understand the hype but... This is 2-3 years after polaris and pascal. Isn't it to be expected that performance for the midrange to be at those levels?

[Drak3]

3 hours ago, johnukguy said:

Given that it's WCCF Tech, I won't hold my breath waiting to see if this is true. If it is, it would literally be a game changer, but I cannot see Radeon, which is the division responsible and who are utterly incompetent, changing anytime soon in order to help gamers out of the goodness of their hearts. Then again, I really really hope that I'm wrong on this one.

We already know that Navi is scalable, meaning able to be an effect MCM design. If yields are high, and subsequentially per die costs are low, it's possible.

 

Then again, with the RX480 and GTX 1060, seeing that type of tier drop isn't unreasonable.

[Humbug]

56 minutes ago, RotoCoreOne said:

I guess I can understand the hype but... This is 2-3 years after polaris and pascal. Isn't it to be expected that performance for the midrange to be at those levels?

Yes it is expected... Which is what makes this story believable. Normal technogical progress in silicon. Expect Nvidia to do the same with their next gen mid-range part in 2019 and go head to head with Navi.

 

We just have to hope that the market will be better by then and miners won't buy up all the Navi parts.

 

 

[GoldenLag]

4 hours ago, Space Reptile said:

so you consider the GTX 1060 and RX X80 series as low end? 

Well the X80 has been AMDs high end (highest end card) since polaris, though they aim it at a same pricerange as the GTX 1060. But before polaris it was mid-end. Exactly where to place the 1060 im not entirely shure but its not mid-end due to 1070/ti and 1080 so shure Low end. 

 

Despute this, this discussion should end allready. Does it matter what you call a GPU? I belive my original point was that 1080 performance should at least historically be offered in the xx60 naming and pricerange at the launch of Navi, but i could be misstaken

[leadeater]

2 hours ago, GoldenLag said:

I belive my original point was that 1080 performance should at least historically be offered in the xx60 naming and pricerange at the launch of Navi, but i could be misstaken

This is the part that wasn't understood very well.  All things going well generation over generation actual performance should slide down two models i.e. 1x80 > 2x60 or 1x50 > 2x30 etc.

[laminutederire]

47 minutes ago, leadeater said:

This is the part that wasn't understood very well.  All things going well generation over generation actual performance should slide down two models i.e. 1x80 > 2x60 or 1x50 > 2x30 etc.

That's a non argument though. If history had a say in performance, AMD would be ballsy good in the CPU market and GPU market solely because there was a time they had big improvements one after the other.

 

As it is right now, I'm not sure Nvidia can do a lot better than they offer now except keeping the same cards and moving them down and use the bigger dies they weren't using so far. That's why it seems they try to push pathtracers for gaming, because their tensor cores allow for better perf for a new task they designed themselves (so it's not a real progress in a way since it only really improves on a new task which didn't exist before).

[leadeater]

1 hour ago, laminutederire said:

As it is right now, I'm not sure Nvidia can do a lot better than they offer now except keeping the same cards and moving them down and use the bigger dies they weren't using so far. That's why it seems they try to push pathtracers for gaming, because their tensor cores allow for better perf for a new task they designed themselves (so it's not a real progress in a way since it only really improves on a new task which didn't exist before).

We already know a fair amount for what is coming performance increase wise, just need to look at Titan V. That's the biggest Volta die and unless the gaming architecture derived from that is significantly different i.e. no FP64 cores and that extra space used for FP32 cores performance is not going to be a large generational leap like people are hoping.

 

I was assuming Tensor cores were not going to be in the gaming cards but the recent raytracing announcements have confirmed that they will indeed have some amount of Tensor cores.

 

Volta went from 16nm to 12nm, increased transistor count by 38% (GP100 vs GV100) and only increased the CUDA cores by 33%, unsurprisingly in game performance increased roughly by 28% on the most friendly titles.

 

Unless what ever is coming has a way different SM core design with much more transistors used for FP32 CUDA cores everyone's in for a big disappointment, for a node shrink this will be the smallest performance increase yet for gamers.

[Valentyn]

16 minutes ago, leadeater said:

We already know a fair amount for what is coming performance increase wise, just need to look at Titan V. That's the biggest Volta die and unless the gaming architecture derived from that that is significantly different i.e. no FP64 cores and that extra space used for FP32 cores performance is not going to be a large generational leap like people are hoping.

 

I was assuming Tensor cores were not going to be in the gaming cards but the recent raytracing announcements have confirmed that they will indeed have some amount of Tensor cores.

 

Volta went from 16nm to 12nm, increased transistor count by 38% (GP100 vs GV100) and only increased the CUDA cores by 33%, unsurprisingly in game performance increased roughly by 28% on the most friendly titles.

 

On the latest gaming reviews, that's moved up to 30-40% with the latest drivers.
https://www.hardocp.com/article/2018/03/20/nvidia_titan_v_video_card_gaming_review

 

Quote

 With performance 40% faster than GeForce GTX 1080 Ti FE AMD has nothing even close to that level of performance in a GPU today. Even if you shrink Vega to 12nm, and boost clocks, it will still never make it to near the performance that NVIDIA TITAN V with Volta can achieve in games.

Even so you're right, Big Volta is an excellent indication of what to expect on the up coming smaller GPUs. Especially if they're aren't cutting out the Tensor Cores.

[Ryujin2003]

16 minutes ago, leadeater said:

I was assuming Tensor cores were not going to be in the gaming cards but the recent raytracing announcements have confirmed that they will indeed have some amount of Tensor cores.

I can only imagine what that's going to mean for the new entry price of PC gaming on green side.

[leadeater]

16 minutes ago, Valentyn said:

On the latest gaming reviews, that's moved up to 30-40% with the latest drivers.
https://www.hardocp.com/article/2018/03/20/nvidia_titan_v_video_card_gaming_review

Even with the updated drivers though there are plenty of games that gain very little to 15% at best, maybe Nvidia has just hit their scaling limit with CUDA cores like AMD did with stream processors ages ago (current game engine wise).

 

I do actually agree with what Nvidia is doing, everyone else is saying go general compute and scale like hell to brute force the performance but I just don't see that as sustainable. Fixed function extremely fast mixed with general purpose seems to be a much better idea to me. Be smarter and find ways to cut out unnecessary computation before it gets done.