AMD vega 10 information leak

[Curufinwe_wins]

50 minutes ago, Sakkura said:

Again, if we go by the source here, the 12 TFLOPS is in double precision. The Titan XP has a grand total of 0.3 TFLOPS of double precision performance.

 

But the source doesn't seem particularly reliable when it comes to the TFLOPS values, so we should probably just not read much into it.

No. The source said 6TFLOPS of half precision  (FP8) which unless explicitly neutered is 1/2 of the FP16 (single precision) standard metric.

[Zeeee]

8 hours ago, huilun02 said:

AMD no pls my wallet...

easy dont buy amd hardware, buy titan xp if you want high end stupid performance. Guaranteed AMD wont even touch 1080ti performance level just like fury x couldnt beat the 980ti even though it was advertised by them as FAST AS A TITIAN X lmao gtfo amd youre drunk

[Citadelen]

1 hour ago, Coaxialgamer said:

-snip-

The Fury X could barely overclock at all. While Polaris has healthy headroom. Especially seeing as it's on a new process.

 

1 hour ago, Sakkura said:

-snip-

Nope! Check the article.

 

@ZeeeeUhh, it matched it at launch. It's faster now.

Vega 10 will probably be as fast as the 1080 Ti or be a tad slower.

I'm not getting excited until Vega 20 or Navi 10. 

[AresKrieger]

3 minutes ago, Citadelen said:

 

Nope! Check the article

I think the main issue here is the article is inconsistent, and thus likely BS, it does mention 12Tflops of double but also mentioned 24 Tflops of half, those two things are contradictory if performance is scaling properly

[IAEInferno]

Newbie question, why won't Nvidia or AMD just focus on a 2nm chip instead of slowly progressing from 28nm, 22nm, 16/14nm, 7nm?????/

 

Newbie here no haterino or else I will cry.

[Citadelen]

1 hour ago, Coaxialgamer said:

-snip-

Look to where I was talking about the multiple GPUs on a single interposer.

5 minutes ago, IAEInferno said:

-snip-

Because nodes that small are incredibly hard to both design and manufacture. AMD are skipping 10nm and are going for 7nm.

Also, past around 5nm quantum tunnelling becomes a massive issue, the electrons essentially teleport past the tine transistors to where ever they want.

[Kunter]

10 minutes ago, IAEInferno said:

Newbie question, why won't Nvidia or AMD just focus on a 2nm chip instead of slowly progressing from 28nm, 22nm, 16/14nm, 7nm?????/

 

Newbie here no haterino or else I will cry.

Because they would have been stuck on 28nm until at lest 2025. Its not that easy to just jump to a way lower node

[IAEInferno]

Just now, Citadelen said:

Look to where I was talking about the multiple GPUs on a single interposer.

Because nodes that small are incredibly hard to both design and manufacture. AMD are skipping 10nm and are going for 7nm.

Also, past around 5nm quantum tunnelling becomes a massive issue, the electrons essentially teleport past the tine transistors to where ever they want.

 

Okay thanks for the info, what happens if electrons teleport past the tine transistors to where ever they want? What is quantum tunnelling in simplified terms that a guy that has no experience with electronics would understand if possible :D??

[Citadelen]

Just now, IAEInferno said:

-snip-

This video explains well.

[The Benjamins]

4 minutes ago, IAEInferno said:

Okay thanks for the info, what happens if electrons teleport past the tine transistors to where ever they want? What is quantum tunnelling in simplified terms that a guy that has no experience with electronics would understand if possible :D??

best way to think of it in a CPU perspective is that it is shorting. like your light switch is off but the electrons quantum tunnel past the switch to make the connection closed.

[IAEInferno]

1 minute ago, Citadelen said:

This video explains well.

 

Okay so the smaller it gets then the higher chances of electrons passing or teleporting through the switch via quantum tunneling, so it's basically a physical limitation when they go for smaller transistors.

[The Benjamins]

1 minute ago, IAEInferno said:

Okay so the smaller it gets then the higher chances of electrons passing or teleporting through the switch via quantum tunneling, so it's basically a physical limitation when they go for smaller transistors.

yes, they are magic shorts.

[Citadelen]

1 minute ago, IAEInferno said:

-snip-

Yes, hence quantum computers.

[IAEInferno]

2 minutes ago, Citadelen said:

Yes, hence quantum computers.

Do you think that there will be consumer quantum computers in the future?

[Zeeee]

1 hour ago, Citadelen said:

The Fury X could barely overclock at all. While Polaris has healthy headroom. Especially seeing as it's on a new process.

 

Nope! Check the article.

 

@ZeeeeUhh, it matched it at launch. It's faster now.

Vega 10 will probably be as fast as the 1080 Ti or be a tad slower.

I'm not getting excited until Vega 20 or Navi 10. 

oh its faster?

 

http://www.hardwaresecrets.com/video-card-review-geforce-gtx-980-ti-vs-radeon-r9-fury-x/9/

http://www.hardwaresecrets.com/video-card-review-geforce-gtx-980-ti-vs-radeon-r9-fury-x/8/

 

literally took me a second to find those i can spend 5 minutes and utterly destroy what u just said but honestly everyone knows fury x can manhandled by the 980ti in every single game and in dx 12 WOW in 2 games it can match the 980ti great thx lov it lots

 

 

WOW REKT again 

Edited September 20, 2016 by Zeeee
MORE INFO

[goodtofufriday]

1 hour ago, Citadelen said:

The Fury X could barely overclock at all. While Polaris has healthy headroom. Especially seeing as it's on a new process.

 

Youre telling me. I think I have the least overclock-able system you can have. 

[WereCat]

19 minutes ago, Zeeee said:

oh its faster?

 

http://www.hardwaresecrets.com/video-card-review-geforce-gtx-980-ti-vs-radeon-r9-fury-x/9/

http://www.hardwaresecrets.com/video-card-review-geforce-gtx-980-ti-vs-radeon-r9-fury-x/8/

 

literally took me a second to find those i can spend 5 minutes and utterly destroy what u just said but honestly everyone knows fury x can manhandled by the 980ti in every single game and in dx 12 WOW in 2 games it can match the 980ti great thx lov it lots

 

 

WOW REKT again 

This video is over year old. Go check comparisons with new drivers.

I am not saying the FuryX is better but now they trade blows a lot more and Fury X perform better in more games over 980ti than it did before.

 

Yes, 980ti is definitely better at 1080p.

[patrickjp93]

7 hours ago, Sakkura said:

That's single precision though. This leak is claiming 12 TFLOPS of double precision performance.

It is not claiming double-precision performance. The article is a mess and says only 6 for half precision, which means it's 12 for single.

[AresKrieger]

1 hour ago, IAEInferno said:

Do you think that there will be consumer quantum computers in the future?

No, since the applications of quantum computers are limited in terms of consumer use, maybe it a couple decades it might be an idea considered but I expect something like light based computing to be a more practical solution by then.

[patrickjp93]

2 hours ago, Curufinwe_wins said:

No. The source said 6TFLOPS of half precision  (FP8) which unless explicitly neutered is 1/2 of the FP16 (single precision) standard metric.

Half precision is FP16. Single is FP32, and double is FP64. Vega does not have 12TFlops FP64. The article is just a mess.

[Citadelen]

2 hours ago, IAEInferno said:

-snip-

I personally couldn't tell you.

2 hours ago, AresKrieger said:

-snip-

 

 

Disregarding the article for a second AMD mentioned at a financial a while back that Vega is an architectural change compared to Polaris and an entire lineup of cards. Thus the RX 5xx series.

1 hour ago, Zeeee said:

-snip-

 

Grammarly had a field day with your post...  

Anyway, you quoted a site I've never heard of, out of date benchmarks and used quite that are quite literally broken on AMD. At launch, the 980 Ti was faster some of the time, and the Fury X was faster others, now, due to driver revisions the Fury X is a few percent faster. Also, every, EVERY single proper DX12 implementation has AMD having an advantage. For example in Deus Ex: Mankind Divide the Fury X is faster than a GTX 1080 at 1080p. 

[Notional]

4 hours ago, Sakkura said:

Again, if we go by the source here, the 12 TFLOPS is in double precision. The Titan XP has a grand total of 0.3 TFLOPS of double precision performance.

 

But the source doesn't seem particularly reliable when it comes to the TFLOPS values, so we should probably just not read much into it.

According to Videocardzzzzzz:

Quote

So what does GFX9 really mean? It’s an internal codename for architecture, for instance Hawaii was GFX7 and Polaris was GFX8.

Meaning Vega is a new gen of architecture. After all, Polaris was designed specifically for laptops, which is why it has certain problems as high/mid end desktop cards.

---

 

The interesting part if their claim of Navi being postponed to 2019, where 2018 will get a 7nm Vega 20. Sounds a lot like Tick-Tock kinda deal. It would depend on 7nm being available and if they can scale up 14nm Vega to 7nm (more stream processors).

[RexinOridle]

Will it play Crysis? Yes. But good luck getting one.

[BingoFishy]

1,460GHz? Dang, that's fast.

[Curufinwe_wins]

2 hours ago, patrickjp93 said:

Half precision is FP16. Single is FP32, and double is FP64. Vega does not have 12TFlops FP64. The article is just a mess.

you are right I got turned around