Vega/Navi Rumors (Updated)

[swilli89]

Who says that this site even knows the actual TDP and isn't just stating worst case scenario based on the power connectors? And finally.. can we please stop acting like people that buy $600-700 video cards give a crap about the difference between 250watts and 300watts even IF true?

Why can car enthusiasts have fun with maxing their hardware with turbo V8's that guzzle gas .. but some "enthusiasts" here cry over what amounts to a small lightbulb? Just reeks of disguised bias and agenda looking for something to poopoo.

If Vega beats 1080Ti by 20% yet uses 300 watts its just as damn efficient but most importantly its FASTER!

 

[Bacon1]

The chart is misleading. Vega's advantage is actually 100% lower than those numbers.

Here's a corrected version:

Vega may have a great advantage over the 580, we'll have to wait and see, but the raw numbers aren't really where that is to be found. It'll be under the hood.

That chart wasn't misleading, it was clear that 100% was the baseline while your chart makes it look like Vega is weaker in most instances.

 

[swilli89]

More Performance with a relatively small increase in Power consumption. Not the end of the world.

Exactly! Amazing how castrated nvidia fans have caused the discussion to become. Back when Fermi could literally fry an egg no one cared about efficiency then the MOMENT GTX 680 came out all of a sudden very specific people started to wail on efficiency instead of performance.

Don't flame half the board.

AT Moderator ElFenix

 

[Bacon1]

Who says that this site even knows the actual TDP and isn't just stating worst case scenario based on the power connectors? And finally.. can we please stop acting like people that buy $600-700 video cards give a crap about the difference between 250watts and 300watts even IF true?

Why can car enthusiasts have fun with maxing their hardware with turbo V8's that guzzle gas .. but some "enthusiasts" here cry over what amounts to a small lightbulb? Just reeks of disguised bias and agenda looking for something to poopoo.

If Vega beats 1080Ti by 20% yet uses 300 watts its just as damn efficient but most importantly its FASTER!

Yep I doubt most of those are final as it makes no sense for both a WC and Air version to have the exact same specs (minus TDP).

Don't worry though, in whatever spec AMD is worse in people will make sure to highlight it and the forums will be flooded with it. Price/perf, Wattage, Card Length, WC'd

 

[swilli89]

Yep I doubt most of those are final as it makes no sense for both a WC and Air version to have the exact same specs (minus TDP).

Don't worry though, in whatever spec AMD is worse in people will make sure to highlight it and the forums will be flooded with it. Price/perf, Wattage, Card Length, WC'd

No doubt about that. I will say I feel good about the fact AMD is charging these prices. People can give them flack about all kinds of stuff but one thing AMD always gets right is competitive price/performance. If they are starting Vega out at $1200 then that must mean something...

 

[beginner99]

I wonder what all those Ryzen users spending money on expensive ram with the hopes of clocking them even higher think they're doing? I'm tempted to think there's some performance to be had in that area? With quadchannel, you have the bandwidth on tap, without breaking a sweat. It's plug and play. AVX2 has proven very useful in handbrake and would probably get adopted more in content creation going forward. It's advantages can't be ignored for too long now.

Someone made that comparison of 580 vs Vega, pretty huge boost.

Wait what? Flops per core per Mhz is identical (in fact slightly lower) for Vega than 580? Is this bulldozer all over again? It will have higher IPC. Well this chart says it clearly does not have higher IPC, it has the same IPC but higher clocks.

Maybe I'm misunderstanding something here but that is not what i expected because AMD clearly claimed higher IPC. Does not look good. I hope I'm wrong.

 

[Hitman928]

Flops per core per mhz is always the same with modern architectures. It's pointless.

 

[Excessi0n]

Wait what? Flops per core per Mhz is identical (in fact slightly lower) for Vega than 580? Is this bulldozer all over again? It will have higher IPC. Well this chart says it clearly does not have higher IPC, it has the same IPC but higher clocks.

There's a lot more to real-world performance than the raw TFLOP count, at least as far as gaming is concerned.

 

[Bacon1]

Wait what? Flops per core per Mhz is identical (in fact slightly lower) for Vega than 580? Is this bulldozer all over again? It will have higher IPC. Well this chart says it clearly does not have higher IPC, it has the same IPC but higher clocks.

Maybe I'm misunderstanding something here but that is not what i expected because AMD clearly claimed higher IPC. Does not look good. I hope I'm wrong.

Thats how the math works.

TFLOPS = Cores * Speed * 2

So 4096 * 1600 * 2 = 13,107,200

Any "IPC" isn't calculated into that, which is why TFLOPS between different arch don't match much, especially in gaming.

 

[Stuka87]

That chart wasn't misleading, it was clear that 100% was the baseline while your chart makes it look like Vega is weaker in most instances.

Yes it was. For instance, for cores 178% advantage over 2304 is 6405. If something is 100% faster than something, its twice as fast. So 178% more means its 2.75 times more.

 

[Glo.]

Oh the doom and gloom.

The TBP numbers are based on maximum power draw available to those GPUs. Actual power draw numbers are more important.

 

[IllogicalGlory]

That chart wasn't misleading, it was clear that 100% was the baseline while your chart makes it look like Vega is weaker in most instances.

That's not what the chart said. The chart said "advantage over RX 580". If it said "percentage of RX 580", it will be correct. My chart is correct based on wording used, and it's much less misleading. There's no reason to use the latter wording except to make Vega look better than it is, in a very superficial way I might add.

Wait what? Flops per core per Mhz is identical (in fact slightly lower) for Vega than 580? Is this bulldozer all over again? It will have higher IPC. Well this chart says it clearly does not have higher IPC, it has the same IPC but higher clocks.

Maybe I'm misunderstanding something here but that is not what i expected because AMD clearly claimed higher IPC. Does not look good. I hope I'm wrong.

Every GPU since like Fermi has had the same FLOPS per shader per clock: 2. One multiply, one add. This applies to GCN, Kepler, Maxwell, Pascal, Vega and pretty much anything even slightly modern.

Oh the doom and gloom.

The TBP numbers are based on maximum power draw available to those GPUs. Actual power draw numbers are more important.

The bad news is that so far every 14nm AMD GPU consumes more power at load than its TDP. The RX 480 consumes about 160W (at launch at least) and the RX 470 consumes about 121W. The RX 460 has no reference version, so it's hard to tell, but all the aftermarket versions are quite power hungry. I miss the good old days when a 180W TDP R9 270X consumed 130W under load.

There's no reason this trend needs to hold, but it is a trend. AMD's TDP's have been very optimistic this gen.

 

[Veradun]

Some sites may get hands on AMD Vega Frontier Edition and release gaming results. Then, people will say wait for Vega Gaming edition (near end of July).

What a pityful use of resources

Gaming benchmarks without drivers, yay!

 

[T1beriu]

Someone made that comparison of 580 vs Vega, pretty huge boost.

Someone has no idea how percentage increases work.

The correct numbers are 78%, 19%, 112%, 89%.

 

[Glo.]

Has anybody spotted the elephant in the room?

Two key things: Cinebench usually has high Overhead on the CPU, because it uses... OpenGL, to test GPU performance.

RX 480 scores between 93 and 97 FPS in this benchmark.

P.S.

GLOBALFOUNDRIES 14LPP platform with 14nm 3D FinFET transistor technology provides best-in-class performance and power with significant cost advantages from 14nm area scaling. 14LPP technology can provide up to 55% higher device performance and 60% lower total power compared to 28nm technologies.

Just based on this, characteristics of the process, Vega should behave within specific core clock, just like Fury X - 250W GPU. 55% increase in the same thermal envelope means around 1.6 GHz core clocks.

 

[zinfamous]

Exactly! Amazing how castrated nvidia fans have caused the discussion to become. Back when Fermi could literally fry an egg no one cared about efficiency then the MOMENT GTX 680 came out all of a sudden very specific people started to wail on efficiency instead of performance.

people weren't making teeny boxes with small power supplies in those days. I don't generally care about the power issue, either--especially if I am expecting performance--and call it irrational, but there is a legitimate argument to be made for the aesthetics and efficiency and well, the necessity of low power/low heat when you want to build a small box that has no compromises on performance. I'm not saying that these unverified leaks are disastrous or that they tell the whole story, but I don't really like how it looks right now.

 

[Wall Street]

Pixel fill-rate seems to indicate that it's a Fury X @1.4GHz. Is the new geometry pipeline not working as intended?

Pixel fill rate primarily measures of the performance of the texturing units, not the geometry units.

 

[zinfamous]

No doubt about that. I will say I feel good about the fact AMD is charging these prices. People can give them flack about all kinds of stuff but one thing AMD always gets right is competitive price/performance. If they are starting Vega out at $1200 then that must mean something...

well, it would have to be significantly more powerful than Titan XP in order to move cards at the same price, simply due to branding. If it were just about price and performance, then it wouldn't matter: nVidia would never be able to sell another Titan XP at $1200 if AMD is selling a 10% better performer at the same price--but we all know that isn't true. People just buy nVidia, because that's what they know. Now, I do think this isn't always the case in the pro market (creative, graphic design), because I believe that Radeon has a longer history, and better name recognition with these clients, so I guess we will have to see how Frontier stacks up if that pricing is accurate.

and that being said, if RX is supposed to be faster )(via drivers, according to Raja), and cheaper....I do remain somewhat dubious over how that is going to happen if those leaked numbers up there are accurate. ....I guess this means that RX will be "unlocked" (does that actually apply to GPU?), and be able to push up to 1800mhz or something like that. ...but how are they going to sell it for less? lol

 

[piesquared]

So both these cards are 13 TF so that confirms 1.6 GHz clock. I guess one of the question left to be answered is if that is base or boost. And will the clock and boost clock be different on rendering workloads and compute workloads. I think the performance listed is for base clock as is usually the case for AMD, while nvidia advertises their performance at max boost. So that extra 75 watt board power that AMD is giving us on the liquid cooling edition probably provides healthy overhead to overclocking/boost. Especially considering the added cost. Thats my guess.

 

[nathanddrews]

NVIDIA: I have great performance and I'm efficient! Buy me!
Forums/Press/Reviews: Fantastic!

AMD: I use more power and give better overall performance, but not as efficiently! Buy me!
Forums/Press/Reviews: Performance is good, but TDP!

NVIDIA: I can beat your performance - watch me OC! Buy me!
Forums/Press/Reviews: You've got to expect TDP to increase with performance like this! Amazing!

AMD: ... but..?
Forums/Press/Reviews: Shame it can't OC as high due to all that TDP.

It's the same ol' shifting goalpost argument every time.


But but but this had nothing to do with the topic and is only here to flame.

AT Moderator ElFenix

 

[Krteq]

Pixel fill-rate seems to indicate that it's a Fury X @1.4GHz. Is the new geometry pipeline not working as intended?

Seems like pixel fillrate was derived from "base clock" (1400 MHz) and peak performance [TFLOPS] was derived from boost clock (1600 MHz)

 

[OatisCampbell]

No doubt about that. I will say I feel good about the fact AMD is charging these prices. People can give them flack about all kinds of stuff but one thing AMD always gets right is competitive price/performance. If they are starting Vega out at $1200 then that must mean something...

I hope you're right but:
A. Pre-release prices are often incorrect.
B.http://www.hardwarecanucks.com/foru...62166-amd-fx-9590-review-piledriver-5ghz.html
That $799 at launch AMD CPU was getting beat by the 2500K that launched 3 years earlier for $216
https://www.techpowerup.com/reviews/AMD/R9_Fury_X/31.html
The FuryX launched for the same price as 980Ti even though it had lower performance and 50% less VRAM

 

[tential]

No doubt about that. I will say I feel good about the fact AMD is charging these prices. People can give them flack about all kinds of stuff but one thing AMD always gets right is competitive price/performance. If they are starting Vega out at $1200 then that must mean something...

Fury X....

 

[nathanddrews]

https://www.techpowerup.com/reviews/AMD/R9_Fury_X/31.html
The FuryX launched for the same price as 980Ti even though it had lower performance and 50% less VRAM

33% less RAM. At launch, the 980Ti was overall faster in many games, but they were neck and neck in some, and the Fury beat handedly it in a few. Depending upon resolution, things also weren't so cut and dry. They also traded blows in compute/non-gaming tests. The 980Ti was ultimately the better card for $649, but not a crushing landslide. The real goldmine was the Fury non-X for $549 that gave near-X performance and could even be unlocked in some cases.

I have no doubt that AMD will price their top Vega at or near its most expensive performance competitor. I mean, why miss out on the margin while you can get it? Then you can drop prices when NVIDIA does.


Quit arguing 2 year old crap.

AT Moderator ElFenix

 

[tential]

Ugh, this tired argument again. Fury X was no where close to GTX 980Ti in performance.

That's without OCing where FuryX had barely any headroom and the 980Ti was an overclockers dream.

This was a crushing landslide.
This isn't a debate this is a fact.

Fury X was terrible perf per dollar, otherwise ya'll would have gotten it, however, even used at GTX 1070 prices, ya'll wouldn't touch it. So many of you love AMD cards for Price/Perf, so don't pretend like Fury X was great Price/Perf, otherwise many of you would have it, instead of STILL having R9 290s.


Quit arguing 2 year old crap.

AT Moderator ElFenix