Radeon RX 480 "Polaris" Launched at $199

[Mauller]

I was in Maplins today looking at the GPU's, most of them were Nvidia from 950's up to 980's of Asus branding. The 980 strix was going for £460, just looked at them all thinking that the RX480 will decimate their sales at the end of the month for under half the price. lol

 

[D.P.]

I was in Maplins today looking at the GPU's, most of them were Nvidia from 950's up to 980's of Asus branding. The 980 strix was going for £460, just looked at them all thinking that the RX480 will decimate their sales at the end of the month for under half the price. lol

People that buy a higher end card from Maplins probably don't have a clue anyway so would buy at those stupid. But this is what happens when there is a new release. OCUK are still trying to sell 980Tus at 500-550quid a pop when a 1070 will be 320-400.

 

[fs123]

I would absolute be against nvidia or AMD providing special reviewer drivers that gave anomalous performance but that isn't the case. Providing reindeers with a non-public driver under NDA to prevent sensitive I formation leaking to the public is absolutely within their rights.

The only driver reindeers would need wears a red suit and has a white beard.

 

[Beren]

Still no news... actually that in itself is an AMD win, they obviously have really good control over their NDA and it is holding up well - if they are hard launching on the 29th I would guess there are already a lot of cards out there in the wild. For no-one to have breached is a pretty strong statement.

If Fury are actually being run down that is really interesting and does point to something faster than a 480 though. I am really hoping for a faster than 480 to come out and give AMD more cards - and if it comes in at the $300 level will tempt me, I was thinking that a 480 would keep me till 1080p reaches end of life... but an extra 15% or so over that would be perfect.

 

[D.P.]

The only driver reindeers would need wears a red suit and has a white beard.

Ho Ho ho

 

[D.P.]

Still no news... actually that in itself is an AMD win, they obviously have really good control over their NDA and it is holding up well - if they are hard launching on the 29th I would guess there are already a lot of cards out there in the wild. For no-one to have breached is a pretty strong statement.

If Fury are actually being run down that is really interesting and does point to something faster than a 480 though. I am really hoping for a faster than 480 to come out and give AMD more cards - and if it comes in at the $300 level will tempt me, I was thinking that a 480 would keep me till 1080p reaches end of life... but an extra 15% or so over that would be perfect.

At least some of the leaks we get are semi-official, purposely put in the wild to drive hype.

I would be extremely surprised if AMD have an actual 490 type card to launch given the complete lack of leaks or rumors though. As you say it is possible they have managed to keep things extremely quiet but it doesn't really make sense. AMD talked about Polaris 10 and 11, but not a 3rd chip that could be the 490, unless that is the smaller Vega 11 chip which AMD said would be in 2017.

I think it is likely P10 has 40CU when whole and the 36CU is cut down for yields, or an Apple delta. But the extra 6 CUs would add up to about 7% gaming performance, that doesn't make a 490 but a 480X. And then there is the fact that AMD seemingly changed naming to RX480, so maybe that is already the X version.


There was a rumour of Vega being pushed early to October. that seems quite impossible but may have its basis in larger Polaris Chip coming out then to compete with the 1080.

 

[hominid]

DP has a point.. But not what he was expecting I think..

What this has proved is that NVidia provides 'special' drivers to reviewers that normal plebs can't get their hands on.. So why use those for reviews if they are not what users can expect, disingenuous at best.

Now it turns out in said drivers that there are some effects that aren't rendered correctly, or are omitted completely, who knows.. It would be good for someone to compare these with the previous available drivers and see if those same irregularities remain.

The review sites shouldn't allow it, but then I'd suspect that they wouldn't get many cards for 'review'..

 

[Overlag]

We are moving away from using Nvidia as a comparison to just an Nvidia thread again.

Soon we will be back to 1070/1080 comparisons like the other Polaris thread

 

[Mauller]

At least some of the leaks we get are semi-official, purposely put in the wild to drive hype.

I would be extremely surprised if AMD have an actual 490 type card to launch given the complete lack of leaks or rumors though. As you say it is possible they have managed to keep things extremely quiet but it doesn't really make sense. AMD talked about Polaris 10 and 11, but not a 3rd chip that could be the 490, unless that is the smaller Vega 11 chip which AMD said would be in 2017.

I think it is likely P10 has 40CU when whole and the 36CU is cut down for yields, or an Apple delta. But the extra 6 CUs would add up to about 7% gaming performance, that doesn't make a 490 but a 480X. And then there is the fact that AMD seemingly changed naming to RX480, so maybe that is already the X version.

There was a rumour of Vega being pushed early to October. that seems quite impossible but may have its basis in larger Polaris Chip coming out then to compete with the 1080.

More likely one of two situations.

1. The 36CU is the full part and they are using a different naming scheme this time, but there will be higher performing/clocked versions at the expense of power use and a higher price. The reference RX480 they have shown is a mix of good price/perf/efficiency as they said they would target before. but they also said that AIB's could do what they pleased with the part.

2. There is a 36-44CU version at $300 as a reference version but with a bit of a performance bump due to the extra hardware and higher base clocks. But of course AIB versions will be clocked higher for both versions if the higher CU part exists anyway.

But we also don't know the top end of this part, they have given us some base specs and performance but there is a complete possibility that they have a greater boost with Polaris parts and that it can clock higher than 1266 by a fair amount when in use. Although the max board power is 150W the parts power usage can still be a reasonable amount under this in normal use, even when clocked up.

Although they gave out some info we still know little about what the new hardware features for polaris do and how much of a performance improvement they give overall. it will more than likely be situational in terms of the performance improvement similar to SMP and single pass for Pascal etc, just that it covers a wider number of situations.

 

[AlamoX]

I think it is likely P10 has 40CU when whole and the 36CU is cut down for yields, or an Apple delta. But the extra 6 CUs would add up to about 7% gaming performance

ok need to know why 7% because you repeated it like 10 times in your last posts, if we were talking about compute capability i would understand that between pure compute and gaming the card loses efficiency, but when we are talking about the 36CU being 100% of it's gaming performance why would 4extra CU be 7% gaming performance instead of 11% ?

 

[D.P.]

ok need to know why 7% because you repeated it like 10 times in your last posts, if we were talking about compute capability i would understand that between pure compute and gaming the card loses efficiency, but when we are talking about the 36CU being 100% of it's gaming performance why would 4extra CU be 7% gaming performance instead of 11% ?

Because these things never scale linearly at all, other bottle necks come in to play. 11% would be the increase in theoretical computes but things like ROPs, TMU, cache, command processor, schedulers, geometry, tessellation processors, bandwidth all come to play.

You see this with all the cutdown cards, they are never as slow as the theoretical difference. The 1070 is much faster than the reduction in CUs would indicate, likewise the 980to to the TX etc.

So yeah, a full 40Cu would in theory be 11% faster, in realty somewhere short like 7%, all dependent on the exact game and where the bottle neck lies.

 

[AlamoX]

Because these things never scale linearly at all, other bottle necks come in to play. 11% would be the increase in theoretical computes but things like ROPs, TMU, cache, command processor, schedulers, geometry, tessellation processors, bandwidth all come to play.

You see this with all the cutdown cards, they are never as slow as the theoretical difference. The 1070 is much faster than the reduction in CUs would indicate, likewise the 980to to the TX etc.

So yeah, a full 40Cu would in theory be 11% faster, in realty somewhere short like 7%, all dependent on the exact game and where the bottle neck lies.

thats why i said 36CU being a 100% gaming performance, it takes all that into account already, so by adding another 4CU you are still taking all that into account if you scale it according to the 36CU gaming performance, which every other post does but you.
so it is possible to abb 9-14% through overclock and get a card 20-25% faster than 480, but is it worth it ? if the 480 is unexpectedly good and hovers around Fury pro performance, then it might be worth it to push 25% to trade blows with the 1070 even if it goes out of it's power envelope, but if the overclock puts it under the 1070, i rather not see them launch a card again being slower than Nvidia's card they are competing with and with worst power efficiency, even if it's cheaper.

 

[Panos]

Then why did an Nvidia employee state it was 'press only' and the reason for the discrepancy? I'm not sure why, if what you say is correct, that comment would have been made.

Press Only drivers are something NV does all the time.
Who forgot the 480 & 580 related scandals with "review only" drivers?

 

[LightSpeed]

I think it is likely P10 has 40CU when whole and the 36CU is cut down for yields, or an Apple delta. But the extra 6 CUs would add up to about 7% gaming performance, that doesn't make a 490 but a 480X. And then there is the fact that AMD seemingly changed naming to RX480, so maybe that is already the X version.

I do not think it is the X version. It is very likely that the RX480 is a cut down part, they have done exactly this in the last generation by releasing the cut down parts first in the mainstream segment (R9 380, 370). There seem to be other pointers as well but ill leave them aside for now.

So, the performance gains. You keep saying 7% but 7-8% would be from the extra CU's themselves but the actual gains would be higher.

We do not know if the RX480 is conservatively clocked but it does seem to be the case. A 390 shrink would clock to around 1.5 - 1.6 ghz pretty easily with the process change only. They could not have possibly sacrificed speed in favour of IPC to the extent that these wont do the same or similar given the massive headroom 14 FF allows over 28nm. I feel 1.26ghz is very conservative, its only around 160mhz above their 28nm big chips when the process allows 500 - 700 mhz (even more theoretically)

Secondly, since it is relatively easy to adjust the memory controller to support GDDR5x, I believe the RX480X will come with GDDR5x when the supply increases as it is very short at the moment (hence the 1080's being so scarce)

So the net gain would be a 11.1% increase in both clockspeed (1.26 -> 1.4ghz) and CU's (36 -> 40) plus 25% more bandwidth. Gaming performance gains? I wouldn't be surprised if its close to 20%.

 

[Orangey]

(Up to) 7% is a 485.

Forget about the 90s for now. You're also not factoring in the potential process problems, if that turns out true it would be the reason for the low clocks and price and also why it's an 80 with no 80X.

 

[Nutella33]

First post in two years of having an account O.0

congrats on breaking your cherry.

 

[tavtavtav]

I do not think it is the X version. It is very likely that the RX480 is a cut down part, they have done exactly this in the last generation by releasing the cut down parts first in the mainstream segment (R9 380, 370). There seem to be other pointers as well but ill leave them aside for now.

So, the performance gains. You keep saying 7% but 7-8% would be from the extra CU's themselves but the actual gains would be higher.

We do not know if the RX480 is conservatively clocked but it does seem to be the case. A 390 shrink would clock to around 1.5 - 1.6 ghz pretty easily with the process change only. They could not have possibly sacrificed speed in favour of IPC to the extent that these wont do the same or similar given the massive headroom 14 FF allows over 28nm. I feel 1.26ghz is very conservative, its only around 160mhz above their 28nm big chips when the process allows 500 - 700 mhz (even more theoretically)

Secondly, since it is relatively easy to adjust the memory controller to support GDDR5x, I believe the RX480X will come with GDDR5x when the supply increases as it is very short at the moment (hence the 1080's being so scarce)

So the net gain would be a 11.1% increase in both clockspeed (1.26 -> 1.4ghz) and CU's (36 -> 40) plus 25% more bandwidth. Gaming performance gains? I wouldn't be surprised if its close to 20%.

Being the same silicon, if there's 250mhz of headroom to clock a 480x up, wouldn't you just be able to OC a 480 up to that speed as it wouldn't be anywhere near its limit? Then the gains would be mostly from the extra physical cores.

 

[five8five]

Secondly, since it is relatively easy to adjust the memory controller to support GDDR5x, I believe the RX480X will come with GDDR5x when the supply increases as it is very short at the moment (hence the 1080's being so scarce)

Wouldn't adding GDDR5x make it too expensive - something best left for the top-end premium cards?

 

[humbug]

The 480 @ 1266Mhz is conservatively clocked to keep the power and price as low as it is.

Minimum PCB's for the budget minded.

The full fat Polaris is 30% more expensive, room for a beefier PCB, higher power and higher clocks.

 

[Jono8]

Wouldn't adding GDDR5x make it too expensive - something best left for the top-end premium cards?

Indeed. I would be incredibly surprised if AMD started kitting out a $199 card with GDDR5X.

Also there is the fact that if it is only 390/390x performance, it isnt going to need more bandwith anyway, the 8ghz clocked DDR5 will be fine.