Vega/Navi Rumors (Updated)

[Paratus]

Raja would have to take some Alka-Seltzer Extra Strength if he has to tell you how he really feels about Vega.

Look you need to step up your game here. No one is going to buy what you are selling the way you are doing it.

If you want to talk down AMD find some bad bad news supported by respectable sites.

If you want to talk up NV again find some positive news from respectable sites. NV is doing well right now and there's plenty of good news.

But making stuff up or trying to pass off opinion as fact isn't going to convince anyone here.

Just some friendly advice.

Now if I had to guess Raja is probably happy with performance but impatient to release Vega. But that's just my opinion.

 

[Paul98]

I really hope they are able to do what I have been waiting for. If so, vram size will be no issue at all, even at 4k with 2gb of vram you would have no issues. Just have to wait and see.

 

[Mockingbird]

Look you need to step up your game here. No one is going to buy what you are selling the way you are doing it.

If you want to talk down AMD find some bad bad news supported by respectable sites.

If you want to talk up NV again find some positive news from respectable sites. NV is doing well right now and there's plenty of good news.

But making stuff up or trying to pass off opinion as fact isn't going to convince anyone here.

Just some friendly advice.

Now if I had to guess Raja is probably happy with performance but impatient to release Vega. But that's just my opinion.

I just bought a Radeon RX 480.

I must be a huge fan of Nvidia. /s

 

[coercitiv]

I just bought a Radeon RX 480.

I must be a huge fan of Nvidia. /s

Your purchase decisions state nothing in regard to your perception of facts and truth in general.

It has been explained to you in plain language that in order to bring a piece of news into discussion you need a respectable source to go with it. It's either that or you take the role of the source yourself, in which case you need to establish credibility. You can't just waltz in the forums making unsupported claims and expect people to believe them, you will be greeted with the troll treatment instead, reputation shredding included.

The other way to discuss these things is to make personal predictions / speculations in which case one must clearly state it as such. It's perfectly fine to do that, as long as you understand that your prediction is not actual news, and should not be presented as such.

 

[cytg111]

NVM

 

[Krteq]

How the hell is this related to Vega GPU?

 

[ElFenix]

The level of discourse seems to be on a downward trend here. I suggest you all try to raise it a bit.

AT Moderator ElFenix

Raja would have to take some Alka-Seltzer Extra Strength if he has to tell you how he really feels about Vega.

and AMD invented time machine to predict the future and what the user will do in next frames... seriously ?
this topic becomes ridiculous

These are exactly the kind of posts I was referring to. Please stop.

AT Moderator ElFenix

 

[Mockingbird]

Let's say for the sake of the argument that AMD decides to release Vega without further enhancement at $499.99:

GeForce GTX 1080 performance + 10% on liquid cooling (no overclocking headroom)

Would you buy rather buy this or the GeForce GTX 1080 (assuming that NVIDIA doesn't change its prices)?

 

[tamz_msc]

Let's say for the sake of the argument that AMD decides to release Vega without further enhancement at $499.99:

GeForce GTX 1080 performance + 10% on liquid cooling (no overclocking headroom)

Would you buy rather buy this or the GeForce GTX 1080 (assuming that NVIDIA doesn't change its prices)?

If that is the case, you honestly believe that AMD have made no progress since the ES they demoed months ago?

 

[Red Hawk]

Let's say for the sake of the argument that AMD decides to release Vega without further enhancement at $499.99:

GeForce GTX 1080 performance + 10% on liquid cooling (no overclocking headroom)

Would you buy rather buy this or the GeForce GTX 1080 (assuming that NVIDIA doesn't change its prices)?

Sure would.

 

[swilli89]

If that is the case, you honestly believe that AMD have made no progress since the ES they demoed months ago?

And we have no idea even what that card was, its clockspeed, its drivers, even if it was a cut down or fully enabled chip.

 

[RoarTiger]

Equal performing products at same price? I would buy AMD. I believe they have better drivers at this point ( just a belief, since I currently run NVIDIA but NVIDIA experience hasnt been that impressive and I have had to rollback drivers many times). AMD GPU architecture does seem to age better too.

edit: Freesync is much cheaper for future monitor upgrade so another plus for AMD

 

[Crumpet]

Let's say for the sake of the argument that AMD decides to release Vega without further enhancement at $499.99:

GeForce GTX 1080 performance + 10% on liquid cooling (no overclocking headroom)

Would you buy rather buy this or the GeForce GTX 1080 (assuming that NVIDIA doesn't change its prices)?

In a heartbeat.

 

[Malogeek]

The Vega purchase for sure. By far much more forward thinking architecture in my opinion. Vega is going to have to be terrible perf/watt for me to consider 1080 but consider it I will. I'm planning this purchase to last me 3 years likely and I don't think a 1080 will do that for me.

 

[crisium]

I would not consider that a big enough jump over my Fury to upgrade at that price. I know that would make it similar to 980 to 1080 improvement, but I'm looking for more.

 

[Head1985]

Let's say for the sake of the argument that AMD decides to release Vega without further enhancement at $499.99:

GeForce GTX 1080 performance + 10% on liquid cooling (no overclocking headroom)

Would you buy rather buy this or the GeForce GTX 1080 (assuming that NVIDIA doesn't change its prices)?

Nope.
Only if its air cooled + have atleast decent oc like 10-15%.

 

[tamz_msc]

I think that was intended as a survey question. As to what that survey is about, who knows.

My bet is that the responses will be noted for future use.

 

[tential]

I would not consider that a big enough jump over my Fury to upgrade at that price. I know that would make it similar to 980 to 1080 improvement, but I'm looking for more.

I'd quit PC gaming and ask for a 1 year ban from here until amd comes back with a real high end gpu.

I can't game with that.

 

[SpaceBeer]

Nice rumour, just not gonna happen. With a launch at that time we would have leaks everywhere already. But no leaks mean, no mass production by partners, as from the point where mass production started (with Ref Designs just Packaging and Shipping) the partners can't keep the stuff secret anymore. So at least there will be no hard launch in the next two weeks. Best case paperlaunch but i still bet for computex.

RX 560 will be released in 2 weeks and there are no leaks. Only official information. Actually, all previous leaks stated only it is 14 CU chip, ie. RX 460 rebrand. And we have found out that's not true, 2 days ago, when AMD presented RX 500 cards.

 

[Mopetar]

RX 560 will be released in 2 weeks and there are no leaks. Only official information. Actually, all previous leaks stated only it is 14 CU chip, ie. RX 460 rebrand. And we have found out that's not true, 2 days ago, when AMD presented RX 500 cards.

Yes, but how often do people really care about leaking the low-end parts anyway?

I suspect we'll start to see some leaks coming out over the weekend and into next week. Hell, no even Apple can really keep their products completely secret anymore, so it's pretty unlikely Vega drops without any leaks.

 

[itsmydamnation]

Let's say for the sake of the argument that AMD decides to release Vega without further enhancement at $499.99:

GeForce GTX 1080 performance + 10% on liquid cooling (no overclocking headroom)

Would you buy rather buy this or the GeForce GTX 1080 (assuming that NVIDIA doesn't change its prices)?

Explain mi25 sub 300watt passively cooled 12.5tflop card? Your doing a piss poor job of evaluating known data points.

AMD has a large disadvantage in memory bandwidth consumption which will be removed in vega with the rop cache being attached to the L2. If you look at memory OC results a 480 gets around 10% perf increase from a 12% memory OC. (2000mhz to 2250mhz). Factor this in and right now video card performance is largely alligned to TFLOPS, NV still has an advantage but it not as big as it looks when comparing stock 1060 to 480 for example.

The perfect devices to compare to your ill throughout position is the 470D to the 1050ti. 2TF with 112gb/s mem bandwidth to 3.3 tf 224gb/s mem bandwidth. the 1050ti has 60% of the TFLOP and the 470D has an extra 40gb/s of memory bandwdith ensuring that it isn't bandwidth starved.

So does the 470D only score 10% above the 1050ti, no no it doesn't try around 35-40% faster.
So if you remove memory disadvantage it looks like NV have about a 15% performance per flop advantage over AMD.


So then how does a 12.5 Tflop card only just beat a 8.2 fltop card? Or are you saying that a gpu with:
more flexible higher "IPC" ALU execution ( better utilization)
less memory consumption (rop cache) which also lower power consumption,
lower power memory ( hbm vs gddr5) allowing more power budget for clocks
improved frontend/geometry engine

has gone backwards per flop to polaris?

Even if we assume a case of 15% perf per flop disadvantage for Vega but removal of memory bottleneck that bang on a 1080ti.

 

[xpea]

CPUs have been doing branch prediction and cache predictions for decades.

Yes precisely. Cache is a well know technique in CPUs but in GPUs the main memory consumption of a game is not shaders nor vertices but high res textures, so it's not applicable because the data is too big to be cache efficient...

There's no particular reason a GPU must load every texture into VRAM just in case. In fact from a statistical stand point most frames look almost the same from refresh to refresh.

Wrong. You need the textures in VRAM for smooth framerate and as I said they are the biggest assets of a AAA game.

Based on AMDs memory work my gut feel is that even with a smaller amount of VRAM Vega will not choke at any resolution and settings a Titan Xpp doesn't.

While I respect your "gut feel", my rational engineering background doesn't believe it

 

[xpea]

So then how does a 12.5 Tflop card only just beat a 8.2 fltop card? Or are you saying that a gpu with:
more flexible higher "IPC" ALU execution ( better utilization)
less memory consumption (rop cache) which also lower power consumption,
lower power memory ( hbm vs gddr5) allowing more power budget for clocks
improved frontend/geometry engine

I don't want to discard Vega improvements but recent history in AMD marketing material vs reality must keep us prudent. Anyone ?

Moreover and nobody reacted to this critical information, I was shocked when during first Vega demo, AMD it self said that the engineering sample were running on Fury drivers. Wow Really ? or Vega is not the totally new arch they want us to believe (otherwise it won't work on old gen drivers) or they were lying and the sample was using an alpha Vega driver and they said some kind of bullshit for us to believe that tje final performance will be much higher.
In any case, it doesn't smell good...

 

[Malogeek]

It's been shown many times that Vega is an evolution of GCN architecture, not a completely new one. It just has some large additions compared to previous GCN revisions.

 

[itsmydamnation]

Moreover and nobody reacted to this critical information, I was shocked when during first Vega demo, AMD it self said that the engineering sample were running on Fury drivers. Wow Really ? or Vega is not the totally new arch they want us to believe (otherwise it won't work on old gen drivers) or they were lying and the sample was using an alpha Vega driver and they said some kind of bullshit for us to believe that tje final performance will be much higher.
In any case, it doesn't smell good...

Thats because you not understand how GCN works compared to other uarches. GCN is designed to be hazard free in terms of register file usage, no complex banking or software scheduling. So long as they have kept this logic in Vega ( all the available patients look like they do) and the command processor shares the same minimum function level then you already have most of the starting point covered.

Then consider that vega has same number of ALU's and prob TEX and Rops with the Vega units having more capabilities the reason for using fury driver as a basis makes sense.

The funny thing about polaris is a lot of worst cases improved quite a lot, but sustained peak performance didn't. there are plenty of games ( gamesworks games as a good example) that saw good improvements but they tended to be badly performing games on GCN.

I expected some of the work in some form from the newer GCN based patients to show up in Polaris, but they didn't. Given the bigger focus put on the ALU improvements given in slideware so far i think we will see some/all of that in Vega.