Has the AMD RX 490 appeared in AOtS benchmarks online?

[HEXUS]

GPU with AMD-like Device ID of 687F:C18 performs on a par with Nvidia GTX 1080 cards.

Read more.

[DanceswithUnix]

If it is Vega, it could be all-sorts.

If I were in charge of Vega, I would have tried to make it bump compatible with Fury as that has very similar capabilities and connectivity. That means at release you have partner boards ready to go including ITX boards, water cooled and dual gpu variants.

4096 shaders seems a bit weak. The RX480 has a few more shaders than R9 380 and a ~30% higher clockspeed giving ~50% more performance. Compare this to Fury, you have the same GCN1.2 in Fury and can expect a similar clock bump, but Fury also has 4096 shaders so no gain there. So it mainly comes down to clock speed, that doesn't seem enough.

[scaryjim]

I dunno, isn't AotS one of the games on which AMD is closer to nvidia anyway? Clock speed alone - assuming Vega can clock in at the same ~1250MHz boost clocks that Polaris hits - would give 25% increase over Fury X. Based on recent GTX 1080 reviews that would push Vega 10 up to somewhere around GTX 1080 performance in many games - particularly those which already favour AMD.

Plus, remember that Vega 10 is the smaller Vega die; Vega 10 with 2 stacks of HBM2 would basically be Fury X+ and trade blows with the GTX 1070/1080, with a much faster Vega 11 card to follow later...?

[[DW]Cougho]

Plus, remember that Vega 10 is the smaller Vega die; Vega 10 with 2 stacks of HBM2 would basically be Fury X+ and trade blows with the GTX 1070/1080, with a much faster Vega 11 card to follow later...?

Is that right? I thought 10 was always the bigger die and 11 was the cut-down 'midrange' die?

If the mid-range Vega can compete with the GTX1080 then it all sounds good, but if this is the big Vega and it's only competing with the GTX1080 then I'm disappointed.

[DanceswithUnix]

... with a much faster Vega 11 card to follow later...?

I think that is what people really want to see. Vega 10 is sounding like a meh product, a cut down 1070 competitor might be interesting but I suspect the full part will be just the wrong side of the 1080. Who is going to pay top dollar for second place?

[Percy1983]

... with a much faster Vega 11 card to follow later...?

I think that is what people really want to see. Vega 10 is sounding like a meh product, a cut down 1070 competitor might be interesting but I suspect the full part will be just the wrong side of the 1080. Who is going to pay top dollar for second place?

It all comes down to price, its 95% of the performance for 80% of the cost its a winner.

[scaryjim]

Is that right? I thought 10 was always the bigger die and 11 was the cut-down 'midrange' die? ...

Nope - although it confused a lot of people at the time too AMD's new dies are numbered chronologically, apparently - with Polaris they developed the large die first (10) then created a cut-down version (11); with Vega they are doing the smaller die first (10) then scaling it up (11) - or at least that was the last rumour I heard.

If the 512GB/s bandwidth is right that actually supports Vega 10 being the small die: with HBM2 transfer rates you hit 512GB/s at 2 stacks, and 1TB/s at four stacks. It looks like Samsung is concentrating on 4-hi HBM2 stacks which will mean 4GB/stack, so a 2 stack card would give 8GB of VRAM at the same overall bandwidth as Fury X, making a potential Vega 10 RX 490 pretty much a die-shrunk Fury X with some clock-speed tuning. That'd leave the top end open for a proper enthuisast Vega 11 card with a lot more shaders and 4 stacks of HBM2, for 16GB @ 1TB/s...

[jigger]

Cougho;6011]

Plus, remember that Vega 10 is the smaller Vega die; Vega 10 with 2 stacks of HBM2 would basically be Fury X+ and trade blows with the GTX 1070/1080, with a much faster Vega 11 card to follow later...?

Is that right? I thought 10 was always the bigger die and 11 was the cut-down 'midrange' die?

If the mid-range Vega can compete with the GTX1080 then it all sounds good, but if this is the big Vega and it's only competing with the GTX1080 then I'm disappointed.

I thought the performance of the GTX1080 was good. The price and features on offer was very disappointing.

[[DW]Cougho]

I thought the performance of the GTX1080 was good. The price and features on offer was very disappointing.

The performance of the GTX1080 is fine, but it's also Nvidias mid-sized core. As opposed to the Titan X which is based on their larger core.

If AMD need their large core just to compete with a 1080 then then Vega is a failure in my eyes.

And thanks ScaryJim - I suppose it makes sense they would be named chronologically, I don't know why I didn't consider that.

[sykobee]

Note that Vega will be using GFX9 design, which may mean some redesign that enabled higher clocks, or higher IPC, etc.

I would assume a smaller Vega that competes with 1070 and 1080, and a larger Vega for competing with Titan and Pro uses.

[CAT-THE-FIFTH]

My main worry is if AMD is using HBM2 instead of GDDR5 and GDDR5X for its GTX1070 and GTX1080 competitors - it might mean Nvidia is able to fight any price war easier,unless OFC the HBM2 and interposer integration process has being cheaper and less unwieldy to do since th Fury series was launched.

[jigger]

Why would anyone care what the size of the die is.

[CAT-THE-FIFTH]

If it is Vega, it could be all-sorts.

If I were in charge of Vega, I would have tried to make it bump compatible with Fury as that has very similar capabilities and connectivity. That means at release you have partner boards ready to go including ITX boards, water cooled and dual gpu variants.

4096 shaders seems a bit weak. The RX480 has a few more shaders than R9 380 and a ~30% higher clockspeed giving ~50% more performance. Compare this to Fury, you have the same GCN1.2 in Fury and can expect a similar clock bump, but Fury also has 4096 shaders so no gain there. So it mainly comes down to clock speed, that doesn't seem enough.

A 4096 shader part was leaked a while back and I suspect it might be the same GPU in the XBox Scorpio - noticed how the PS4 PRO has the same number of shaders as Polaris 10.

I suspect the same can be said of Bonaire too - the XBox One GPU has the same number of shaders too(but some are disable).

I expect AMD has been using the consoles to help spread out R and D costs for its own PC graphics lines.

Why would anyone care what the size of the die is.

For the end user its not important,but OFC it does affect how much margins AMD makes and how easily they can compete with Nvidia in price too.

[DanceswithUnix]

My main worry is if AMD is using HBM2 instead of GDDR5 and GDDR5X for its GTX1070 and GTX1080 competitors - it might mean Nvidia is able to fight any price war easier,unless OFC the HBM2 and interposer integration process has being cheaper and less unwieldy to do since th Fury series was launched.

If AMD only need 2 stacks of HBM2 to get their 8GB, then that might not be so bad for price.

That doesn't guarantee parts volume though, and in the 1070 market that might be important. I wonder how a single stack 4K shaders 4GB card would perform. Should still be faster than Fury, either that or AMD would want a shader disabled part which wouldn't cut costs as much.

[spolsh]

Fingers crossed for this.

[CAT-THE-FIFTH]

If AMD only need 2 stacks of HBM2 to get their 8GB, then that might not be so bad for price.

That doesn't guarantee parts volume though, and in the 1070 market that might be important. I wonder how a single stack 4K shaders 4GB card would perform. Should still be faster than Fury, either that or AMD would want a shader disabled part which wouldn't cut costs as much.

Well the XBox Scorpio is rumoured to be using a 4096 shader GPU but has GDDR5 or GDDR5X. I wonder if Vega 11 has both GDDR5 or GDDR5X and HBM2 compatability??