AMD Confirms Vega is Launching this Quarter

[Vayra86]

I'll leave that here: TPU - AMD Radeon Vega GPU Architecture

And wtf is https://www.techpowerup.com/gpudb/2871/radeon-rx-vega?

Carefully read between the lines.

"The GPUs will still employ GCN compute units, but will augment them with several new components never before featured on AMD GPUs, which will improve the chip's efficiency to cope with complex graphics and general processing workloads. "

 

[Fluffmeister]

Carefully read between the lines.

"The GPUs will still employ GCN compute units, but will augment them with several new components never before featured on AMD GPUs, which will improve the chip's efficiency to cope with complex graphics and general processing workloads. "

Indeed, even AMD refer to Vega as 5th generation GCN...

https://www.docdroid.net/PucBhT9/gdc2017-advanced-shader-programming-on-gcn.pdf.html#page=35

 

[efikkan]

I been reading the post all over again and I'm tired ...

Normally a refesh of any product is marginally better at some aspects, but RX 580 is in fact a worse product than RX 480, since it sacrifies a lot in terms of efficiency for a very marginal performance gain. That's not opinions, but facts.

Now why is efficiency important? Since both vendors are able to create GPUs of roughly the same size, and GPUs are more or less going to struggle over 250W, efficiency becomes the deciding factor for performance, and is the end also price. GP102 and Vega10 are GPUs of the same size, so if Vega are to come close in performance, it would have to be very close in efficiency. Right now the efficiency-gap between Polaris and Pascal is massive, and Vega would have to cut off >80% of that gap to even be close to competing. This is why it's very unrealistic that Vega will even compete in the high-end.

 

[ratirt]

Normally a refesh of any product is marginally better at some aspects, but RX 580 is in fact a worse product than RX 480, since it sacrifies a lot in terms of efficiency for a very marginal performance gain. That's not opinions, but facts.

Now why is efficiency important? Since both vendors are able to create GPUs of roughly the same size, and GPUs are more or less going to struggle over 250W, efficiency becomes the deciding factor for performance, and is the end also price. GP102 and Vega10 are GPUs of the same size, so if Vega are to come close in performance, it would have to be very close in efficiency. Right now the efficiency-gap between Polaris and Pascal is massive, and Vega would have to cut off >80% of that gap to even be close to competing. This is why it's very unrealistic that Vega will even compete in the high-end.

And you still compare Polaris to Pascal. You haven't answer anything. You say efficiency and that's all. For me 580 is superior to 480 and it is a good refresh. It has better OC potential, is faster etc. So if you were to buy a video card now from a lower range tier and you were choosing between 480 vs 580 you would choose 480 I suppose? Well my answer is 580 since it is better for the same price. For me it is simple.

 

[efikkan]

And you still compare Polaris to Pascal. You haven't answer anything. You say efficiency and that's all. For me 580 is superior to 480 and it is a good refresh. It has better OC potential, is faster etc. So if you were to buy a video card now from a lower range tier and you were choosing between 480 vs 580 you would choose 480 I suppose? Well my answer is 580 since it is better for the same price. For me it is simple.

And what should I compare then? Besides Polaris and Pascal?
If RX 580 had the same thermals as RX 480 it would be fine, but it sacrifices a lot of efficiency for a very minimal gain, any reasonable person understands this. And the OC potential is lower for RX 580, since it's essentially just an overclocked RX 480.

 

[ratirt]

And what should I compare then? Besides Polaris and Pascal?
If RX 580 had the same thermals as RX 480 it would be fine, but it sacrifices a lot of efficiency for a very minimal gain, any reasonable person understands this. And the OC potential is lower for RX 580, since it's essentially just an overclocked RX 480.

So you say I'm not reasonable? thanks bro that was very cool. Thermals are lower than 480 with new cooler and heat sink. I would rather better performance not thermals or lower wattage. but that's just me I guess.

 

[Vayra86]

So you say I'm not reasonable? thanks bro that was very cool. Thermals are lower than 480 with new cooler and heat sink. I would rather better performance not thermals or lower wattage. but that's just me I guess.

By that rationale any watercooled card is also a much better GPU because it runs cooler. Your logic is completely flawed, and indeed it is not reasonable. You need to look beyond your stupid RX580 because that is entirely not what this topic is about. 'Thermals are lower with a bigger heatsink' LOL comment of the day.

 

[ratirt]

By that rationale any watercooled card is also a much better GPU because it runs cooler. Your logic is completely flawed, and indeed it is not reasonable. You need to look beyond your stupid RX580 because that is entirely not what this topic is about. 'Thermals are lower with a bigger heatsink' LOL comment of the day.

water cooler is a premium and all know that. you don't and that's just sad
And no. I don't need to look beyond my stupid rx 580 cause it's just the comparison of the2 480 and 580 no other. Period. you compare those 2 and that is what I wanted to share. You are locked with NV(envy) of the 2. so please. please don't tell me how do I live already know that

 

[medi01]

The "leaked" bench look like epic.fail

http://www.3dmark.com/spy/1544741

 

[efikkan]

Add the benchmark to a comparison, and it will reveal the GPU name as "687F:C1", which is the same engineering sample of Vega10 AMD demonstrated running Doom.
Keep in mind the clocks of the final product might be higher than 1200 MHz.

 

[EarthDog]

The "leaked" bench look like epic.fail

http://www.3dmark.com/spy/1544741

You posted the result.. cool.

For the slow, me apprently, tell me why that link does anything more than what was previously linked?

 

[uuuaaaaaa]

The "leaked" bench look like epic.fail

http://www.3dmark.com/spy/1544741

Add the benchmark to a comparison, and it will reveal the GPU name as "687F:C1", which is the same engineering sample of Vega10 AMD demonstrated running Doom.
Keep in mind the clocks of the final product might be higher than 1200 MHz.

HBM2 @700MHz on 2 stacks will give less bandwidth than 4 stacks of HBM1 @ 500MHz + we already know that the Radeon Instinct MI25 must be clocked around 1525MHz to meet AMD's specs. I still believe that AMD will deliver this time, they have no other option.

 

[efikkan]

HBM2 @700MHz on 2 stacks will give less bandwidth than 4 stacks of HBM1 @ 500MHz + we already know that the Radeon Instinct MI25 must be clocked around 1525MHz to meet AMD's specs. I still believe that AMD will deliver this time, they have no other option.

I encourage people to have realistic expectations, it's not like this card is going to be a GTX 1080Ti killer.

Lower memory bandwidth than Fury X is actually quite possible. Not only did Fury X fail to utilize all it's bandwidth, Vega will also employ slightly better compression and various other improvements to reduce bandwidth requirements.

 

[medi01]

I encourage people to have realistic expectations, it's not like this card is going to be a GTX 1080Ti killer.

Well, it's not like it would be unreasonable to expect 500mm2 chip to beat 47?mm2 chip, but everything that has been shown so far, is rather to underwhelming (beating 1080 at Doom vulkan,, followed by pathetic event where we learned Vega will be called Vega, give me a break) for that to materialize.

we already know that the Radeon Instinct MI25 must be clocked around 1525MHz to meet AMD's specs

That's true, but what we see runs at 1200 and is... pathetic.

was previously linked

I possibly have missed what was linked previously.

 

[EarthDog]

Does size always matter?

 

[Caring1]

Does size always matter?

Every time

 

[ratirt]

Every time

What about small but freaking crazy animal ??

BTW: Do we know the exact date of the Vega release? From what I've read so far it's not that great. I hope they are wrong. I'm just waiting for Vega it would be not so good if I get to be disappointed. :/