AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[misuspita]

There is no r9, actually we don't know what the commercial names will be, but from Lisa Su we know that this RX 480 is middle of the road chip, priced at 200. She said there are other, between 100-300. That price range would be covered by Polaris only, be it 10 or 11. No Vega in that price range that we have been told yet. It may well be, but they didn't let that info slip.

 

[jpiniero]

Why do you think that? There's absolutely no reason the full die can't be $299. This is a Pitcairn-sized chip.

I think AMD is done having small differences in price between models, although if the 8 GB model is $230 then I guess. Either way it does seem like the 480 is it for desktops for the time being until Vega arrives and they will just fill in rebrands below the 4 GB 480. The C4 model mentioned is more likely a Mobile Polaris 10 model now but we will have to see.

 

[oussama-tn]

In the interest of simplicity I think they will call the full chip rx485 if it exists.

Vega will be rx490/495 and fury/fury x branding.

that's a bit too much for high end cards.

 

[Mopetar]

A full polaris would be 480x an r9 490 would be cut version of Vega no?

That's always been my thinking. I don't know where people keep getting the idea that we'll see a 490 before Vega. Looking back over the last several generations there's always been a different chip for the x80/x90 lineups or the x800/x900 series before that.

In the interest of simplicity I think they will call the full chip rx485 if it exists.

485 doesn't seem likely either based on AMD naming history. The x85 designation has been used when the have a refresh with a newer part. The 280 and 280X were both Tahiti, and the 285 was Tonga. The did the same with the 265 as well, with a Pitcairn part replacing the Bonaire 260/260X.

 

[poofyhairguy]

A full polaris would be 480x an r9 490 would be cut version of Vega no?

Based on history, yes.

 

[happy medium]

I thought the next Fury X would be the Vega 10
and the next vanilla Fury/Nano would be the Vega 11?

What part is the 390/390x replacement?

 

[Mopetar]

I thought the next Fury X would be the Vega 10
and the next vanilla Fury/Nano would be the Vega 11?

What part is the 390/390x replacement?

Small Vega is likely the 490/490X chip.

All of the current Fury cards are based on Fiji, so it doesn't make any sense for them to be split for the 400 series.

 

[Sweepr]

Unlike Maxwell -> Pascal where they regressed in IPC, I don't expect GCN 3 -> GCN 4 to regress.

I don't get it. If you compare the Geforce GTX 980 to the Geforce GTX 1070 (closest comparison) there's a 25.3% GFLOPS rate advantage for the latter, yet the Pascal VGA is ~41% faster at 1440P according to TPU.

 

[oussama-tn]

I thought the next Fury X would be the Vega 10
and the next vanilla Fury/Nano would be the Vega 11?

What part is the 390/390x replacement?

I think the 390/390x was made to fill the 300-450 segment since Fiji was new a chip with hbm wich makes them expensive to manufacture and sell

 

[Glo.]

You do realise that card has an 8 pin connector?

AIB card to be super clocked or RX 490 hiding in plain sight?

29th June :sneaky:

I see only 6 pin connector on Sapphire GPU.

 

[cen1]

And about those temps at load ?

After a run of Unigine Heaven at ultra 1440p it never went over 65 degrees C.

 

[kraatus77]

I don't get it. If you compare the Geforce GTX 980 to the Geforce GTX 1070 (closest comparison) there's a 25.3% GFLOPS rate advantage for the latter, yet the Pascal VGA is ~41% faster at 1440P according to TPU.

980= 1200mhz x 2 x 2048 = 4.9 tflops ( actual clock)

1070= 1800mhz x 2 x 1920 = 6.9 tflops (actual clock)

6.9/4.9= 40.8 % more tflops for 40.8% more performance, where are you getting that 25.3% difference lol.

and tpu's 1070was running at 1873mhz, so it's actually lower in ipc.

there's no ipc increase for pascal. period.

 

[oussama-tn]

After a run of Unigine Heaven at ultra 1440p it never went over 65 degrees C.

Thanks!!

 

[Sweepr]

980= 1200mhz x 2 x 2048 = 4.9 tflops ( actual clock)

1070= 1800mhz x 2 x 1920 = 6.9 tflops (actual clock)

Curious to see where are you getting 1800 MHz 'actual clock' from, considering typical boost is only 1683 MHz (base is 1506 MHz). Anyway, even this calculation confirms what I'm questioning, no 'IPC regressions' here. Thanks!

 

[tential]

980= 1200mhz x 2 x 2048 = 4.9 tflops ( actual clock)

1070= 1800mhz x 2 x 1920 = 6.9 tflops (actual clock)

6.9/4.9= 40.8 % more tflops for 40.8% more performance, where are you getting that 25.3% difference lol.

and tpu's 1070was running at 1873mhz, so it's actually lower in ipc.

there's no ipc increase for pascal. period.

We know it's a lower IPC. Only a select few don't want to believe it. The way to make it painfully obvious is oc both cards by the same amount of mhz and you'll see maxwell gains performance.

 

[selni]

980= 1200mhz x 2 x 2048 = 4.9 tflops ( actual clock)

1070= 1800mhz x 2 x 1920 = 6.9 tflops (actual clock)

6.9/4.9= 40.8 % more tflops for 40.8% more performance, where are you getting that 25.3% difference lol.


there's no ipc increase for pascal. period.

There are, but they may be more geared towards compute (or GP100 cores really are quite different to GP104).

https://devblogs.nvidia.com/parallelforall/inside-pascal/

TLDR is NV claiming improved scheduling and some new instructions. It's not just a straight die shrink (again unless GP104 is quite a different architecture, not just a different layout), but it doesn't appear to matter very much for gaming so far.

 

[happy medium]

Small Vega is likely the 490/490X chip.

All of the current Fury cards are based on Fiji, so it doesn't make any sense for them to be split for the 400 series.

SO your saying no 490 cards till at least the end of the year?
WHy would AMD do that? that's suicide!

they must have a full Polaris 10 or something, that don't make sense.

 

[kraatus77]

Curious to see where are you getting 1800 MHz 'actual clock' from, considering typical boost is only 1683 MHz (base is 1506 MHz). Anyway, even this calculation confirms what I'm questioning, no 'IPC regressions' here. Thanks!

did you even read tpu review ? if i go by everything right there's a regression.
and there's no regression when i leave that 73mhz advantage from 1070.


have you ever owned any nvidia gpu after kepler ? all gpus run 7-8% above their stated boost clocks. all of them ( unless there's huge temps). but thanks for confirming your 25% difference was straight up fud :thumbsup:.

besides low ipc isn't really a bad thing if you can't accept this fact (it hurts i know), as long as there's frequency, core increase. and that's what matters.

 

[selni]

SO your saying no 490 cards till at least the end of the year?
WHy would AMD do that? that's suicide!

they must have a full Polaris 10 or something, that don't make sense.

That was always the plan - polaris was aimed at mainstream/performance with vega being the enthusiast release following shortly after. It's not very different to how NV has done it - the mid size cores coming well before big kepler/maxwell/pascal.

It's possible GP104 was a little faster than expected, but it's been priced such that it doesn't really matter.

 

[Sweepr]

have you ever owned any nvidia gpu after kepler ? all gpus run 7-8% above their stated boost clocks.

Unfortunately for you Kepler is irrelevant here. You failed to prove that Geforce GTX 1070 FE was running at max boost clocks all the time during TPU testing so let me help you with some real Pascal data from Hardware Canucks:

The card ran between 1750-1800 MHz most of the time during their 10-min testing. Now let's take this piece from your single useful post in this discussion:

Originally Posted by kraatus77 View Post
980= 1200mhz x 2 x 2048 = 4.9 tflops ( actual clock)

1070= 1800mhz x 2 x 1920 = 6.9 tflops (actual clock)

6.9/4.9= 40.8 % more tflops for 40.8% more performance

And there you have it. Thanks again!

all of them ( unless there's huge temps). but thanks for confirming your 25% difference was straight up fud :thumbsup:.

Could be FUD, but here's another possibility. Maybe I just took the only numbers I found, as NVIDIA only lists data for base clocks and you're jumping straight to personal attacks.

 

[kraatus77]

Could be FUD

it was. thanks

 

[96Firebird]

So in conclusion, no regression in IPC found.

 

[Mopetar]

SO your saying no 490 cards till at least the end of the year?
WHy would AMD do that? that's suicide!

they must have a full Polaris 10 or something, that don't make sense.

Their roadmap has had Vega as late 2016 / early 2017 for months now so I can't see how this is news to you if you've been paying attention.

There likely will be a full Pascal die but it will be a 480X and I don't expect it will be much more than 25% performance over the 480 based on what we know so far.

AMD will be fine not having any high end cards because they'll own everything below that for several months.

 

[kraatus77]

So in conclusion, no regression in IPC found.

what i don't understand is, when i leave 73mhz out, nobody bats an eye. and goes as fas as to another review to show different clocks.


for the last time, show me within a review same delta between performance and tflops of 980/1070. if you can't than it's proven there's a ipc regression.

 

[tential]

Unfortunately for you Kepler is irrelevant here. You failed to prove that Geforce GTX 1070 FE was running at max boost clocks all the time during TPU testing so let me help you with some real Pascal data from Hardware Canucks:

The card ran between 1750-1800 MHz most of the time during their 10-min testing. Now let's take this piece from your single useful post in this discussion:



And there you have it. Thanks again!



Could be FUD, but here's another possibility. Maybe I just took the only numbers I found, as NVIDIA only lists data for base clocks and you're jumping straight to personal attacks.

Except the card in use wasn't the one from hardware Canucks so this whole post is useless.
The only thing it confirms is that these cards struggle to sustain stable clockspeeds.