Will AMD release any high performance GPU's in Q2 or Q3?

[Tiberria]

The only reason why the Titan XP (first one) is lower than the 1080 Ti on most benchmarks is because the Titan is locked to the reference card cooler/reference clocks only, while the 1080 Ti has AIB versions that add 5-10% more performance. 1080 Ti Founders vs Titan XP performance is virtually identical.

[Shinzai]

Are you being funny with me now?

http://www.tomshardware.com/reviews/...ti,4972-8.html

They are margin of error differences with the titan taking wins in some games. Once you clock them the same its the same card minus 1gb of vram.

You are killing yourself with your own argument now, which wasnt even relevant to my 1080/titan comment.

The only way you can be this misinformed is you saw a 1080ti being benchmarked against a maxwell titan.

Surely we wont be seeing shinzai post in here again.

Ummm, no. You don't even understand what's causing the closeness of the results. The reason the Titan XP closes in in certain games is when the memory bandwidth becomes an issue and the X(P) can exploit its superior bandwidth to close the gap, if not surpass the 1080ti. This is why at lower resolutions, the 1080ti tends to outstrip the X(P) completely, but the gap rapidly closes when reaching 4k.

https://arstechnica.co.uk/gadgets/20...080-ti-review/

Spoiler: 

Again, the changes in the architecture are more important than just a small overclock and can have more significant effects.

[Fascinate]

Ummm, no. You don't even understand what's causing the closeness of the results. The reason the Titan XP closes in in certain games is when the memory bandwidth becomes an issue and the X(P) can exploit its superior bandwidth to close the gap, if not surpass the 1080ti. This is why at lower resolutions, the 1080ti tends to outstrip the X(P) completely, but the gap rapidly closes when reaching 4k.

https://arstechnica.co.uk/gadgets/20...080-ti-review/

Spoiler: 

Again, the changes in the architecture are more important than just a small overclock and can have more significant effects.

You're incompetence and gall is on a level rarely seen on this forum.

So you dismiss tomshardware for a site more befitting your viewpoint, that is testing at 1080p resolutions? Architectural changes, what? Its the same GPU with a few tweaks made to clocks and bus structure.

Lets not forget the tiny detail that the 1080ti released 6 MONTHS after the titan did, rx vega barely has one month.

This is my last reply to you, have a good day.

[Thunderball]

Not a huge fan of this site but:
https://www.youtube.com/watch?v=0l6Uc5kBrxY

Its an overclocked fury x from 2 years ago

Their methodology is currently the best out there. So if you need clear results their tests GN is the way to go. Their comparison tables are also huge.

RX Vega is going to be > FE Vega in games, its just a matter of how much

5%-15% or 30%-35% ?


the leak is belieavable to me

Yes, clocks are going to be higher, but the card is going to use cut down silicon (less stream processors, less memory), even if slightly in the case of the top variant. Imo there is no way it's going to beat partner GTX 1080s.

- - - Updated - - -

Well there is a tiny possibility that AMD could have done a slightly different VRM setup or something else to enable higher clocks than the FE edition, in which case Vega would be faster.

It's just pure speculation though.

Look at the Vega FE VRM. You cant really go better than that. That VRM is very powerful even for a top partner card.

Titan X Pascal and GTX 1080 Ti are the same chip, with higher clocks on the 1080Ti and 1/12 of ROPs and memory bus disabled due to changed memory config. All GTX 1080 Ti even have a BGA mount for the 12th memory chip.

[Life-Binder]

theres no reason why the top rx vega would have less then 4096 SPs

[Amalaric]

I don't know... I would shake my head at AMD's stupidity if gaming card Vega comes with a 2048-bit bus instead of a 4096-bit.

[Life-Binder]

I don't know... I would shake my head at AMD's stupidity if gaming card Vega comes with a 2048-bit bus instead of a 4096-bit.

? FE already has 2048-bit

https://pro.radeon.com/en-us/product...ntier-edition/

Memory Data Rate 1.89Gbps
Memory Speed (Effective) 945MHz
Memory Size 16GB
Memory Type HBC
Memory Interface 2048-bit
Memory Bandwidth 483GB/s

- - - Updated - - -

ran 3Dmark 11 just now on my Gaming X 1080 (old version, not 11 Gbps and core settles @ 1900-1950, I didnt push for max OC)

GPU score was 31 477 - http://www.3dmark.com/3dm11/12259683


though it was old freeware version from their site, it didnt properly recognize my GPU or drivers and couldnt validate the result, so idk if that affected it

[Lathais]

Both the 1080ti and the Titan X Pascal are based on GP102. They are different variants however.

1080ti: GP102-350-K1-A1
Tian X Pascal: GP102-400-A1

They are built on the same GP102 but are different. I see no reason why AMD could not be doing the same thing. In fact, it stands to reason that they are because otherwise they could launch at the same time. So there is the possibility that it could be based on the same platform, but be different in other ways, the same way there are differences in the 1080ti and the Titan X Pascal, the P5000 and the 1080, and the P4000 and 1070. They are similar, but different and have different performance in games/benchmarks/professional applications.

[Thunderball]

Both the 1080ti and the Titan X Pascal are based on GP102. They are different variants however.

1080ti: GP102-350-K1-A1
Tian X Pascal: GP102-400-A1

They are built on the same GP102 but are different.

GP102-350-K1s are made out of GP102-400s. Nvidia probably expected lower yields out of GP102 (or higher sales of P6000). AMD could be doing the same thing but that means that their top consumer card is not going to beat GTX 1080 (they have to make is slower than Vega FE, even if slightly). Now compare Vega FE PCB and 1080 FE PCB: there is no way top Vega is going to be cheaper as it requires more expensive memory, beefier VRM and has a larger die (Vega FE has the same die size as GP102, GP104 is much smaller), even if AMD can get good yields for those chips.

[Life-Binder]

https://www.3dcenter.org/news/ipc-te...n-vega-10-chip

[Thunderball]

https://www.3dcenter.org/news/ipc-te...n-vega-10-chip

Why link this? That article is solely based on GamersNexus results with a mention of the test by people of 3DCenter forum. Let's link both.

http://www.gamersnexus.net/guides/29...ame-clocks-ipc
https://www.3dcenter.org/news/hardwa...es-3-juli-2017

Two explanations to these results: either Vega is just a rescale of Fiji with small improvements like small primitives discarder (which GN note as a reason to the difference when using high tesselation settings) and HBM2 (albeit with a cut memory bus, which definetely helped Fury X in 4K; or drivers are not able to take advantage of architecture improvement. Honestly, my guess would be a combination of both: architecture improvements are likely very small, but they should definetely be getting us something apart from increased clocks.

P.S. Also, people, please dont start about games not being optimized: developers shouldnt optimize their games for specific hardware, hardware manufacturers should instead optimize for APIs developers use.

[Amalaric]

? FE already has 2048-bit

https://pro.radeon.com/en-us/product...ntier-edition/



- - - Updated - - -

ran 3Dmark 11 just now on my Gaming X 1080 (old version, not 11 Gbps and core settles @ 1900-1950, I didnt push for max OC)

GPU score was 31 477 - http://www.3dmark.com/3dm11/12259683


though it was old freeware version from their site, it didnt properly recognize my GPU or drivers and couldnt validate the result, so idk if that affected it

It should have been 4096-bit.

[Lathais]

P.S. Also, people, please dont start about games not being optimized: developers shouldnt optimize their games for specific hardware, hardware manufacturers should instead optimize for APIs developers use.

That's just not the way it works though. Devs need to optimize for the hardware and the hardware manufacturer's do help by working with them to optimize the code and by optimizing drivers for specific games. Doesn't matter how you think it should work, it matters how it does work.

[Deleted]

P.S. Also, people, please dont start about games not being optimized: developers shouldnt optimize their games for specific hardware, hardware manufacturers should instead optimize for APIs developers use.

Doesn't really work that way. There are multiple way of using given api properly and so some hardware will work better with one way of using said api some will work better with other way of using it. For example nvidia and amd and DX11.

Also interesting point from the same guy on why vega doesn't really perform all that well but more like overclocked fury in the video and reply to first pinned comment. Though I am not holding my breath that it will be amazing once released

[VooDsXo]

Ummm, no. You don't even understand what's causing the closeness of the results. The reason the Titan XP closes in in certain games is when the memory bandwidth becomes an issue and the X(P) can exploit its superior bandwidth to close the gap, if not surpass the 1080ti. This is why at lower resolutions, the 1080ti tends to outstrip the X(P) completely, but the gap rapidly closes when reaching 4k.

https://arstechnica.co.uk/gadgets/20...080-ti-review/

Spoiler: 

Again, the changes in the architecture are more important than just a small overclock and can have more significant effects.

everything synthetic and otherwise is actually showing Titan XP ahead by a decent amount in terms of raw numbers, if its worth the $ increase thats subjective, but I'm only finding results like you'd be claiming with a Titan X.

[Thunderball]

That's just not the way it works though. Devs need to optimize for the hardware and the hardware manufacturer's do help by working with them to optimize the code and by optimizing drivers for specific games. Doesn't matter how you think it should work, it matters how it does work.

I know that it's how it works right now, that's why we have AMD APIs and Nvidia APIs. Dev optimize for whoever pays better.

[Shinzai]

everything synthetic and otherwise is actually showing Titan XP ahead by a decent amount in terms of raw numbers, if its worth the $ increase thats subjective, but I'm only finding results like you'd be claiming with a Titan X.

Yeah, I explained earlier the Titan X was originally referred to as the Titan XP in the press, later became the Titan X (Pascal) and the new one is specifically the Titan XP or Xp.

The newer Titan XP is significantly faster than the 1080ti, despite sharing the same clocks, which is why I've re-stated over and over that it's the changes they've made on the board and to how the bandwidth is manipulated that makes the real difference.

[Thunderball]

Yeah, I explained earlier the Titan X was originally referred to as the Titan XP in the press, later became the Titan X (Pascal) and the new one is specifically the Titan XP or Xp.

The newer Titan XP is significantly faster than the 1080ti, despite sharing the same clocks, which is why I've re-stated over and over that it's the changes they've made on the board and to how the bandwidth is manipulated that makes the real difference.

1080Ti uses the slighly cut-down version of Titan X chip, Titan XP uses the full spec GP102 chip, the same one used in P6000. P6000 has been outperforming Titan X in gaming, it simply has more processing power.

[Kuntantee]

It will be around 1080 performance is my guess ya. Some benchmarks said it was 1070 but there is no way that is true, AMD would get laughed outta the market.

That's how they've been doing for ages and they survived.

[Amalaric]

That's how they've been doing for ages and they survived.

If AMD doesn't want to best Intel / nVidia they can just fuck off for all I care.