AMD Polaris 10 "Ellesmere" as Fast as GTX 980 Ti: Rumor

[RejZoR]

In a way I hope it's true. Maybe this will be AMD's repetition of HD5000 series. Those made monumental leaps in performance for very affordable price.

 

[Parn]

Wow, $300 for 980Ti level of performance. If this is true I'll be switching to the red camp unless GP104 is even faster and comparatively priced.

 

[NeDix!]

4th generation GCN is not going to have large changes except for the shrinking itself, so I'm wondering how ~2.560 of these cores can outperform 4.096 cores from Fiji while still operating at a lower frequency. It would be nice, but I'm not convinced.

Sorry i didnt how to hide the image on a spoiler thing 3)rz

And one thing.... people kinda not remember the relation performance between 780ti and 970
Still my bet is on .... something between 980 and fury/980ti
And for the price 300-400USD

 

[Dethroy]

Can't wait to see what NVIDIA's and AMD's new gpu architectures have to offer. But I will probably wait for GP100 and Vega until I make my next purchase.

 

[Prima.Vera]

In a way I hope it's true. Maybe this will be AMD's repetition of HD5000 series. Those made monumental leaps in performance for very affordable price.

One can only dream about it...me included my friend, me included...

 

[N3M3515]

One can only dream about it...me included my friend, me included...

If you go back a bit, you'll realize it's not a dream, those leaps in performance were always present when shrinking the node and lets not forget that this jump is huge 28 to 14, because they skiped 20nm.

 

[rruff]

If you go back a bit, you'll realize it's not a dream, those leaps in performance were always present when shrinking the node and lets not forget that this jump is huge 28 to 14, because they skiped 20nm.

May be a dumb question, but why are we not getting leaps in performance with Intel's node shrinks?

 

[cadaveca]

May be a dumb question, but why are we not getting leaps in performance with Intel's node shrinks?

because they've already optimized CPU portion as much as they can, so they use the added space to increase cache performance and iGPU.

GPUs just add in more processing blocks, maybe chance cache hierarchy, layout, display controller, memory control. Very different is the purposes they are used for.

 

[Brusfantomet]

May be a dumb question, but why are we not getting leaps in performance with Intel's node shrinks?

Short and inadequate answer: Amdahl's law.

Long answer: The task a CPU and a GPU are fundamentally different, but the core of it related to your question is parallelism. The GPU tasks are almost perfectly parallelizable while the CPU tasks normal users are doing are not. That is why Fiji has over 4000 cores while S 1151 tops out at 4 cores. Adding extra cores for the CPU does not help, so Intel takes the space saved in each node shrink and ads GPU and other stuff.

 

[SliceT]

If they are going for a sub 300 USD card, you'd be lucky if they get 390X performance!
Then nvidia will have its own mainstream card around 300USD which ultimately if history is our friend, will beats AMD in 75% of the benchmarks, which will have AMD drop 30 bucks off their original price to remain "competitive" but nvidia will still sell shit loads of cards because in the end, gamers want the best frame rate.

That's my 2 cents ...

 

[TheinsanegamerN]

Alright, I only did it the second time becaus3 he got choked up on it the first time. I'll play nice. What does Sony have to lose from putting that much power in a console? They still have the PS4 for casual gamers. Since the same games can be scaled up or down, they would not be dividing the market that much. (unlike the GENESIS, 3DO, Sega CD days)

They say the games can be scaled easily, but it still throws a different set of hardware into the mix. Different capabilities, a different GPU arch, better CPU speed, ece. They can say all they want that games will "scale" and be playable on both platforms, but devs have a very poor history of actually doing this.

 

[HD64G]

If you go back a bit, you'll realize it's not a dream, those leaps in performance were always present when shrinking the node and lets not forget that this jump is huge 28 to 14, because they skiped 20nm.

Exactly! It's twice as big the leap in performance for same architectures already. So, it can be used either to make an almost 2X GPU in performance in same die size and same power consumption, or a same level performing GPU for half the size and consumption. Polaris 10 is the 2nd option one. Vega 10 might be the 1st.

 

[xLegendary]

http://arstechnica.co.uk/gadgets/2016/04/amd-focusing-on-vr-mid-range-polaris/

Roy Taylor talks about Polaris.

"If you look at the total install base of a Radeon 290, or a GTX 970 or above [the minimum specs required for VR], it's around 7.5 million units," explained Taylor. "But the issue is that if a publisher wants to sell a £40/$50 VR game, there's not a big enough market to justify that yet. We've got to prime the pumps, which means somebody has got to start writing cheques to big games publishers. Or we've got to increase the install TAM [total addressable market].

"The reason Polaris is a big deal," continued Taylor, "is because I believe we will be able to grow that TAM significantly. I don't think Nvidia is going to do anything to increase the TAM, because according to everything we've seen around Pascal, it's a high-end part. I don't know what the price is gonna be, but let's say it's as low as £500/$600 and as high as £800/$1000. That price range is not going to expand the TAM for VR. We're going on the record right now to say Polaris will expand the TAM. Full stop."

What he doesn't tell is that, nvidia will have a replacement at some point for 970, at for sure will have 980+ performance at lower prices.

 

[rruff]

Roy Taylor talks about Polaris.

His point doesn't make a lot of sense. Early adopters of VR will need to pony up $$$ for headsets, video cards, and games if they want a good experience. It's always like that with new tech. Cards have existed for many years that have adequate specs for VR, they just aren't real cheap.

The top end Polaris 10 might make it to 390/970 level, but just barely. That means competing 28nm cards will be even cheaper for the people wanting the cheapest minimum VR spec. And it's not like Nvidia is going *only* build high end cards with the new node. It's pretty certain they will cover the full range with it before AMD does.

 

[HD64G]

http://arstechnica.co.uk/gadgets/2016/04/amd-focusing-on-vr-mid-range-polaris/

Roy Taylor talks about Polaris.



What he doesn't tell is that, nvidia will have a replacement at some point for 970, at for sure will have 980+ performance at lower prices.

First bunch of GPUs from nVidia will be GP104 being available to buy in late summer. First bunch of AMD GPUs will be available in early summer and it's clear they will hit at price levels from the words Taylor let out. So, 290/970 level of performance for lot less money, power consumption and temps is the conclusion, which should make everyone happy.

 

[bug]

In a way I hope it's true. Maybe this will be AMD's repetition of HD5000 series. Those made monumental leaps in performance for very affordable price.

In the meantime, AMD made it official Polaris 10 is a mainstream chip and Polaris 11 is for notebooks. Huge letdown as this allows nvidia to milk the high-end for another 9-12 months.

 

[Dethroy]

In the meantime, AMD made it official Polaris 10 is a mainstream chip and Polaris 11 is for notebooks. Huge letdown as this allows nvidia to milk the high-end for another 9-12 months.

High-end is such a small % of the market anyway. That's why the performance segment is usually the first to hit the market. This way companies also get to milk the actual performance parts as high-end if the competition has nothing similar to offer (see NVIDIA).

 

[bug]

High-end is such a small % of the market anyway. That's why the performance segment is usually the first to hit the market. This way companies also get to milk the actual performance parts as high-end if the competition has nothing similar to offer (see NVIDIA).

So you're saying AMD not having high-end parts is an achievement then?

Edit: Having a mainstream card that's as fast as 980Ti would be an achievement, though.

 

[G33k2Fr34k]

So you're saying AMD not having high-end parts is an achievement then?

I think it's important to see what AMD is going for here. Polaris is around 2/3 the size of GP104. If Polaris XT manages to achieve 90% of say GTX1080's performance, then that's a win-win product for AMD. That was the case with the 4870 and GTX280, if you remember.

Polaris has 90% the FP resources of the Hawaii chip. I think Polaris has improved "shader-efficiency" over current GCN chips. Polaris should also have memory compression for frame buffer color data, which helped quite a bit with Tonga. Hawaii chips lack this feature.
In the worst case, Polaris XT will match Hawaii XT, which should place it in a very competitive position against Hawaii based cards.

 

[bug]

I think it's important to see what AMD is going for here. Polaris is around 2/3 the size of GP104. If Polaris XT manages to achieve 90% of say GTX1080's performance, then that's a win-win product for AMD. That was the case with the 4870 and GTX280, if you remember.

That's fanboy talk. We've been here before, here's how it plays out:
- Nvidia is first to the market with a flagship, milks customers for months
- AMD releases a flagship that's a tad faster than Nvidia's
- Nvidia cuts prices, so AMD gets no chance to make a profit

It's true that money's in the mid-range segment (and OEM), but let's not ignore the elephant in the room: this is the first time Nvidia will release a new flagship and AMD won't have an answer for about a year. And their argument about "VR TAM" is pure rubbish, that's what scares me the most - it means there's no other reason they did this except they couldn't build a proper flagship yet.

 

[john_]

this is the first time Nvidia will release a new flagship and AMD won't have an answer for about a year.

Em... nope. It happened before. There are cases as old as the first 8800GTX and more recently as GTX 980. Other than a dual gpu card, AMD had no answer against GTX 980 for months. And Vega is not a year away.

 

[rruff]

And Vega is not a year away.

AMD says it will not be introduced until 2017, so even a year from now would probably be lucky.

 

[G33k2Fr34k]

That's fanboy talk. We've been here before, here's how it plays out:
- Nvidia is first to the market with a flagship, milks customers for months
- AMD releases a flagship that's a tad faster than Nvidia's
- Nvidia cuts prices, so AMD gets no chance to make a profit

It's true that money's in the mid-range segment (and OEM), but let's not ignore the elephant in the room: this is the first time Nvidia will release a new flagship and AMD won't have an answer for about a year. And their argument about "VR TAM" is pure rubbish, that's what scares me the most - it means there's no other reason they did this except they couldn't build a proper flagship yet.

You're the one fanboy talking pal. A "flagship" card that costs +$500 and offer 10% more performance over $350 priced Polaris XT is not a win. That's the position Nvidia was in when they released their 200 series cards back in 2008, and they lost market share to AMD in that year.

As for Polaris 10 positioning, I don't think it's over-optimistic to expect Polaris10 XT to outperform Hawaii XT despite having 90% the SP count of Hawaii. Given the higher shader efficiency, new "geometry engine", pixel color compression, and potentially higher clock speeds of Polaris 10, I think it's highly unlikely for Polaris10 cards to not outperform their Hawaii counterparts.

The other point to be made is that it seems that AMD's cards perform quite a bit better than Nvidia's in DX12.
This trend will continue with the new cards, I think. This will help AMD in the benchmarks.

 

[Casecutter]

I think we need to focus on RTG has said, this next release is to target the best performance for a given resolution. I see them at such a price point $300-350 offering great 1440p. There will be a part above that for 4K and V-R.

I think RTG is distancing the traditional idea of some price/performance benchmark leader dictating you get the utmost experience. Honestly, I don't know how RTG can thwart the ingrained perception of if you don't lead in benchmarks you lose. RTG will have to find a way that says for 1440p (or whatever) just looking at max/average frame rate is no longer the "end-all-be-all". The technology and hardware that improved pixels, more color, offer a more immersive experience is what you should look for. As long as the frame-rate and frame-pacing is smooth and consistent, you rely more on the hardware to improve quality; and less on the settings in the game or other middle-ware to improve what your eyes perceive. The problem is that becomes more subjective than a B-M just run with all the setting Ultra-out. Instead of games holding on to the dark, brood'ish, Cinematic or Comic book appearance, many games will make use of the such new technologies and hardware to offer more real-life immersive appearance, as what V-R is going to offer and feel like.

 

[efikkan]

I think it's important to see what AMD is going for here. Polaris is around 2/3 the size of GP104. If Polaris XT manages to achieve 90% of say GTX1080's performance, then that's a win-win product for AMD.

That would be an enourmous achievement. GCN is currently way behind Maxwell, and Pascal is a major architectural overhaul while 4th generation GCN is not. Keep in mind that Fiji is just performing close to GM200, even though it has 50% more theoretical performance.

A 200mm² chip is not going to be hig-end, and even AMD says they're targeting mainstream. You'll have to wait at least until Vega before AMD is even trying, if ever again.

Polaris should also have memory compression for frame buffer color data, which helped quite a bit with Tonga. Hawaii chips lack this feature.

How would memory compression help with bad performance? AMD has plenty of memory bandwidth, that is really not the issue.