[Guru3d] Radeon RX 470 Benchmarks

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

crisium

Platinum Member
Aug 19, 2001
2,643
615
136
Look, you still do not understand. Going wider with lower clocks is a design decision deliberately done by AMD.

Not necessarily. That's an assumption as much as any of mine. Plenty of GF criticism out there.

It does _NOT_ mean you can overclock Polaris to insane levels.

My main issue with your argument is, that you are taking perf/area as basis - an irrelevant metric - and are only happy (see you 1700Mhz overclock statement) until perf/area of Polaris matches Pascal. Not going to happen.

That's not my expectation. I want AMD to at least match the relative perf/area gains going to 14nm that Nvidia made going to 16. This is my whole point being disappointed. Nvidia was already ahead on 28nm. It appears to me that they may have gotten more ahead.
 

sirmo

Golden Member
Oct 10, 2011
1,012
384
136
$149 reference card beating a $500 MSI gtx780 I bought 2 1/2 years ago. Pretty insane.
 

Elixer

Lifer
May 7, 2002
10,376
762
126
That's not my expectation. I want AMD to at least match the relative perf/area gains going to 14nm that Nvidia made going to 16. This is my whole point being disappointed. Nvidia was already ahead on 28nm. It appears to me that they may have gotten more ahead.
Why don't you wait for the actual high-end performance part (Vega) instead of trying to compare a mid-range part (Polaris) to some 'expected' level of performance?
 

Lyfer

Diamond Member
May 28, 2003
5,842
2
81
We need more benchmarks to conclude anything. Unless these two benchmarks are cherry picked. If benchmarks are consistent in other titles this is one amazing $150 card.
 

sirmo

Golden Member
Oct 10, 2011
1,012
384
136
Why don't you wait for the actual high-end performance part (Vega) instead of trying to compare a mid-range part (Polaris) to some 'expected' level of performance?
Yeah that guy's arguments are pretty nonsensical. There are "uncore" components which don't shrink like the cores do (memory controller for one).. Also comparing perf/area is not easy as the cost of larger dies isn't linear. The yield gets exponentially worse the larger you go.

We have no idea if rx480 is a bin either. r9 390x was a bin too, we never actually saw the full uncut part in cards.. making any sort of comparison impossible.

Seems like another Nvidia fanboy grasping at straws at this point.
 
Last edited:

Piroko

Senior member
Jan 10, 2013
905
79
91
What I take away from this: Both Polaris and Pascal seem to target around 60W per 100mm². That would cap both architectures at around 450mm² if you want to stay within 300W, unless you drop clocks (=area efficiency).
 

maddie

Diamond Member
Jul 18, 2010
4,722
4,627
136
What I take away from this: Both Polaris and Pascal seem to target around 60W per 100mm². That would cap both architectures at around 450mm² if you want to stay within 300W, unless you drop clocks (=area efficiency).
Using HBM2 relieves a lot of that limitation. You can easily get above 550mm², especially as the top tier will probably have 12-16 GB.
 

sirmo

Golden Member
Oct 10, 2011
1,012
384
136
Probably that the 750 Ti offered close to GTX480 performance four years after Fermi launched with an even larger power and price advantage.
Thermi was a horrible card though. gtx780 was a pretty power efficient and decent card at the time.
 

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,414
8,356
126
You're right I'm working with incomplete info. That's the whole point, working with what I have. I think it's fun to work with theoreticals and make guesswork from a mix of rumour and press releases to see if AMD is making gains (no fun allowed armsdealer).

Nv clocked their new cards 50%+ over their old, although the max OC is closer to 40%. AMD is only +20% over the 390X based on 480 numbers. That itself is the main reason for the leaked performance gains, and lower wattage does not necessarily mean they have the skies limit for frequencies. For now I am working with the numbers we have and assuming a similar overclocking margin that the competition has until proven otherwise

There are frequency limits from the architecture too that can be somewhat independent of power consumption, e.g. Pitcairn uses much fewer watts but doesn't get higher clocks than Tahiti. A lot of Shaders that can't (relatively) clock high is perhaps why we don't see better gains. Those Shaders take up die spice - frequency does not, so getting performance with more and more shaders takes up more and more space - again that's my worry the hard limit of a die size.

1600-1700Mhz+ max OC (with voltage tuning) on a typical aftermarket would shut me up fast
icon10.gif
since that's a 40% gain over typical Hawaii and means AMD is really stressing PPW and there is plenty of room for performance per die. We shall see. Remember, I want to be convinced I'm wrong so I appreciate it (I want a big fat Vega that competes with the next Ti).

meh, GCN makes more sense to go wider than to go faster. big deal. silicon isn't that expensive. and silicon power consumption increases with the square of the frequency. frankly, the power limits are more of hard target than die sizes. the PCIe spec delivers power in discrete steps - the slot by itself can deliver only so much before additional connectors are needed, and each additional connector can deliver only so much. and knowing OEMs and consumers, they'll strongly prefer to match those steps.

and on the notebook side that stuff gets even worse. better to deliver parts that match already available cooling and power delivery systems than to require an OEM to re-engineer things.

(note, i'm not saying AMD is necessarily doing these things)

and keep in mind, AMD has to buy a certain amount of silicon. so, there's a bunch of factors shading toward go a bit bigger than go a bit faster.

so, quit with the sky is falling act :)
 
Last edited:

f2bnp

Member
May 25, 2015
156
93
101
470 looks to be a great card for most people. Plenty of performance for 150$, reminds me very very much of the 4850. It was an outstanding card in terms of price/perf.

If the 480 matches 390x/980, it'll be a nice card. I would have liked it to be a little faster overall, maybe closer to Fury Nano/Fury, but it is fine as long as its along these lines. Now what really remains to be seen is whether or not it overclocks well enough. I wanna be optimistic, 7850, 7950, 7970 were originally clocked very conservatively, so to me it would make sense that AMD may want to do the same thing to be as power efficient as possible.
If that is true and a 1266MHz (seems to be the clock for 480) clocks up to 1500 or 1600, now that would be something else, it would be pretty close to Fury X and 980Ti at this point. People often tout just how awesome the 980Ti was/is, but truth, when using stock clocks, the Fury X was very competitive and they traded punches very often. What made this card was OC, it clocks like a champ. So, hopefully, something like this applies to the 480. If not, that's okay by me, I'll just be slightly disappointed in that.

There's not much else to do right now other than wait for more "official" leaks or just wait for the reviews to come out. Some of the speculation here is pretty lame if I may say so. Just chill and wait, we're all excited here ;).
 

OatisCampbell

Senior member
Jun 26, 2013
302
83
101
Thermi was a horrible card though. gtx780 was a pretty power efficient and decent card at the time.

It's pretty good to beat the 290 for $150, but that is the mom n' pop/paper route market.

I've had a 290 for more than 2 years, I wanted to see the $300-$350 card that obsoletes FuryX.

:(

THAT is the card that generates some excitement on an enthusiast board in my opinion.

The only reason I care about this at all is it might help keep AMD in the market. :'(
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
meh, GCN makes more sense to go wider than to go faster. big deal. silicon isn't that expensive. and silicon power consumption increases with the square of the frequency.

Incorrect. Power usage increases with square of voltage, but linear with frequency.

If the architecture and circuitry inherently clocks higher, then it won't need extra voltage over one that inherently clocks less.
 

crisium

Platinum Member
Aug 19, 2001
2,643
615
136
frankly, the power limits are more of hard target than die sizes

But aren't both hard limits in a practical sense. If, say, the top chip of Company 1 was 630mm2, and the top chip of Company 2 needs 800m2+ to match it, is that even possible? I think that's the true barrier as the expense of development, manufacture, and failure rates might be completely unacceptable. Then you have 500W TDP 295X just doing its thing and using however much power it pleases.
icon10.gif
Even though the heat has 876mm2 to dissipate over, that comes down more to quality of coolers I would think.

Seems like another Nvidia fanboy grasping at straws at this point.

It seems you have it all figured out.

vvv
 
Last edited:
Aug 11, 2008
10,451
642
126
We need more benchmarks to conclude anything. Unless these two benchmarks are cherry picked. If benchmarks are consistent in other titles this is one amazing $150 card.

Totally agree. We dont even have cards in the same segment to compare. And the NDA is not even released on the AMD cards. Way too early to make any conclusions.
 

sirmo

Golden Member
Oct 10, 2011
1,012
384
136
It's pretty good to beat the 290 for $150, but that is the mom n' pop/paper route market.

I've had a 290 for more than 2 years, I wanted to see the $300-$350 card that obsoletes FuryX.

:(

THAT is the card that generates some excitement on an enthusiast board in my opinion.

The only reason I care about this at all is it might help keep AMD in the market. :'(
Maybe I am different from the rest, but a card like rx480 really excites me. Reminds me of the 48xx or 58xx.. cards. The next leap in mainstream graphics.

This raises the bar for the market as a whole instead of just yet another GPU for the top 10%. I will still probably get a Vega or a Volta card, but rx480 is more interesting as it pertains to the community as a whole.
 

maddie

Diamond Member
Jul 18, 2010
4,722
4,627
136
There's not much else to do right now other than wait for more "official" leaks or just wait for the reviews to come out. Some of the speculation here is pretty lame if I may say so. Just chill and wait, we're all excited here ;).
That's optimistic. I would say some are experiencing horror at AMD releasing the RX models. Years of progress about to be lost.
 

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,414
8,356
126
Incorrect. Power usage increases with square of voltage, but linear with frequency.

Ah, quite right

Britishgentleman.jpg



But aren't both hard limits in a practical sense. If, say, the top chip of Company 1 was 630mm2, and the top chip of Company 2 needs 800m2+ to match it, is that even possible? I think that's the true barrier as the expense of development, manufacture, and failure rates might be completely unacceptable.

That's still not a hard point. That's an asymptotic limit rather than a defined inflection point. You can tell when you use fuzzy terms like "might be unacceptable."

We like to talk about a lot of things in this industry as "barriers" when they're more "milestones".
 

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
Incorrect. Power usage increases with square of voltage, but linear with frequency.

If the architecture and circuitry inherently clocks higher, then it won't need extra voltage over one that inherently clocks less.

As a whole he s right, to increase frequency by a ratio R current must be increased with the same ratio wich is proportional to the square of voltage,
to sumarize increasing frequency by said ratio R mandate to increase voltage by sqrt(R), the resulting power increasement is R^2.
 

ocre

Golden Member
Dec 26, 2008
1,594
7
81
Why don't you wait for the actual high-end performance part (Vega) instead of trying to compare a mid-range part (Polaris) to some 'expected' level of performance?

We need more benchmarks to conclude anything. Unless these two benchmarks are cherry picked. If benchmarks are consistent in other titles this is one amazing $150 card.
Totally agree. We dont even have cards in the same segment to compare. And the NDA is not even released on the AMD cards. Way too early to make any conclusions.

Yeah that guy's arguments are pretty nonsensical. There are "uncore" components which don't shrink like the cores do (memory controller for one).. Also comparing perf/area is not easy as the cost of larger dies isn't linear. The yield gets exponentially worse the larger you go.

We have no idea if rx480 is a bin either. r9 390x was a bin too, we never actually saw the full uncut part in cards.. making any sort of comparison impossible.

Seems like another Nvidia fanboy grasping at straws at this point.

All these comments are pretty funny considering the article in the OP....which, of course has a chart with the supposed 470 in the middle of the graph. Probably didn't notice that they had the 1080, 1070 and way too many more to list. Doesn't make much sense...

Let's just pretend this isn't a thread in reference to an article that puts a 470 in a chart with pascal cards. Why are we even talking about Polaris then, I mean I don't get what is wrong with projecting and guessing when pretty much that is all we can do with the limited info we have at this point. I think it is more so that perhaps it was just this members projections that you didn't like.
Maybe you don't realize but you now need to inject these comments in every discussion about polaris, I mean if you were to be consistent.

Because, one cannot say the most basic things in regards to polaris with this kind of thinking. I am sure you guys feel comfortable saying polaris will be a good value...but you see, that would be a violation all the same.

Some "expected performance" is absolutely needed if you have the slightest feeling that polaris will be a good value. We can only expect, we are short on verifiable data. We are supposed to expect, that was AMDS market strategy.
 

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
Going by the Firestrike database it scores similar to 280X 1150MHz with same 5960X. Am I off in not being that impressed in terms of a new node? Unless 470 is extremely cut down to 24 CU (1536 cores). Then it becomes technically impressive but also would be the end of the 2nd tier extra bang for the buck AMD has been providing.