[Guru3d] Radeon RX 470 Benchmarks

[4K_shmoorK]

http://www.guru3d.com/news-story/radeon-rx-470-benchmarks.html

In the slide-deck that was released yesterday some benchmark numbers have been, well almost hidden. But they are there. I added them into two charts to check out.

Let me clearly state that the benchmarks have been performed by AMD so we cannot verify quality settings. The scores have been derived from the footnotes of the PDF:

Polaris 10 2.8x performance per watt: Testing conducted by AMD Performance Labs as of May 10, 2016 on the AMD Radeon™ RX 470 (110w) and AMD Radeon™ R9 270X (180w),ona test systemcomprising i7 5960X @ 3.0 GHz 16GB memory, AMD Radeon Software driver 16.20 and Windows 10.. Using 3DMark Fire Strike preset 1080p the scores were 9090 and 5787 respectively. Using Ashes ofthe Singularity 1080P High, the scores were 46 fps and 28.1 fps respectively. Using Hitman 1080p High, the scores were 60 fpsand 27.6 fps respectively. Using Overwatch1080p Max settings, the scoreswere 121 fps and 76 fps respectively. Using Performance/Board power, the resulting average across the 4 different titles was a perf per watt of 2.8X vs the Radeon R9 270X. Test results are not average andmay vary.

Good to see is that they used a test-setup that matches ours:

Testing conducted by AMD Performance Labs as of June 10, 2016 on the Radeon™ RX 480 8GB and R9 380 on a test system comprising Intel i7-5960X, 16GB DDR4-2666 RAM, GigabyteX99-UD4, Radeon Software Crimson Edition 16.1.1, and Win10 64-bit. Using the Steam VR Performance test, Radeon RX 480 scored 6.3 and Radeon R9 380 scored 3.6. Performance may vary based onuse of latest driver versions. Test results are not average and may vary.

BTW theres also mention of a mobile part called AMD Radeon RX 480M based on Polaris 11:

Testing conducted by AMD Performance Labs as of May 10, 2016 on 3DMark 11 and 3DMark Firestrikeusing a test system comprising of an i7-4600M, 8GB, AMD Radeon driver 16.20. AMD Radeon R9 M280X (14CUs) scored 5700 and 3500 with a board power of 82W. AMD Radeon RX 480M (16CUs) scored 7200 and 4070 with a board power of 35W. Using Performance/Board power, the resulting average across the 2 different titles was a perf per watt of 2.8X vs the Radeon R9 M280X .

The RX 470 would use roughly 110W, benchmarks used are 3DMarl FireStrike, Ashes of the Singularity, Hitman and Overwatch. The card is only being compared to a 270X, however FireStrike is normalized and was placed into our regular chart:

 

[HannooFX]

Impressive performance for the price R9 290 performance for $150

 

[crisium]

Unimpressive that a 14nmFF 2048SP card (my best guess) cannot beat a 28nm 2560SP card.

 

[rainy]

Unimpressive that a 14nmFF 2048SP card (my best guess) cannot beat a 28nm 2560SP card.

For 110W TDP card it's indeed a disaster performance. /s

 

[dark zero]

Decent. Really decent and is only the reference card. Wondering if the best card can reach Fury Vanilla levels.

 

[Bacon1]

FYI you might want to request to close this thread, once Sweeper adds the information to his you'll get an infraction for creating a "duplicate" thread.


Mod callouts are not allowed
Markfw900

 

[crisium]

redacted

 

[Bacon1]

We'll see what the wattage is, but based on the purported die size and SP count, it is a complete and utter disaster, and leaves no confidence that the company is capable of competing in DX11 whatsoever in any technical metric.

It shows it using 110 watts in the first set of graphs.

This is the $150(?) 470 not 480 and it is competing with a 290 and above 780.

 

[crisium]

That's the TDP its rated at from the AMD slides. It will use less than 110W if they follow the same rating as Pitcairn (it did not use anywhere close to 180W).

All Guru3d did was take the AMD slides and put it into a graph. They added their own Firestrike scores from a similar build, so YMMV, but it isn't looking good if we are to take AMD at their word.

 

[Armsdealer]

? They are actually showing up with their 2.8x perf / watt. It wasn't just "ceo talk" What's not to like? No way nvidia has an 80w gtx 970

 

[khon]

So performance close to 970 GTX for $150 at 110W, and this is bad somehow ?

 

[dark zero]

Oh boy... crisium's gloom destroys mine hard. I don't like AMD way to do anything, but this one impressed me. Is a very decent product with OC potential (rumors sais that Asus will launch a 1600 Mhz version)

 

[crisium]

Have some more gloom. Keep in mind I'm not criticizing the price, just abyssmal gains on a new node relative to the other company.

I can only make guesses based on rumoured die sizes. But this performance is straight from the horse’s mouth.

It appears that AMD has not matched the die size gains from the new node. P10 vs Hawaii and GP104 vs GM200 demonstrate this clearly as the ratios are similar, actually likely slightly more favorable to AMD yet they gain far less. The 25% cut 1070 still draws even (slightly beats both stock) with the fully enabled GM200. The only 11% (guessing….) cut 470 seems to draw even with the 9% cut 290? Ouch. If AMD matched Nvidia’s gains, a 11% cut P10 should be easily faster than fully enabled Hawaii.

Shaders too, same story as the die size. Nvidia’s old 1920 beats beats at stock and draws a 2816 980 Ti when both OC. Hey hey, here comes AMD’s new 2048 and it draws the 2560.

Power consumption we shall see. But if the GP104 improvement over Maxwell stay mostly true to GP106 then Nvidia will have absolutely no problem exceeding 970 performance at under 100 W too.

If you are interested in cards at this price don’t let me convince you otherwise because this is a price-to-performance winner here for now since they are beating GTX 1060 and 1050 to market. I’m simply expressing that from a technical perspective this should give you no confidence that AMD can properly compete at the top with bigger size chips unless you really believe in the OCing headroom and/or HBM2 for Vega.

 

[LTC8K6]

Impressive performance for the price R9 290 performance for $150

If it actually had 290 performance, they would likely advertise it that way, and not compare it to a 270X, imo.

I'm guessing that it is actually in the 270X ballpark overall.

 

[Thala]

Unimpressive that a 14nmFF 2048SP card (my best guess) cannot beat a 28nm 2560SP card.

How many SPs is completely irrelevant. The only interesting metric is perf/W.
In this particular case the 14nm gain is in the reduction of voltage thus reducing power significantly and not in bumping up the clock.

just abyssmal gains on a new node relative to the other company.

And more of the same bullshit. Where did you see abyssmal gains in perf/W? All i can see are gains of the factor 2 and more.
Show me where the other company beats this in perf/W by a significant margin. (That is if you have the power numbers for RX470)

 

[MajinCry]

You sound like you're comparing different architectures, in the same way that CPU lads used ta with GHz.

 

[khon]

If it actually had 290 performance, they would likely advertise it that way, and not compare it to a 270X, imo.

I'm guessing that it is actually in the 270X ballpark overall.

They compare it to the 270 because it is the 470, really not that complicated.

 

[Armsdealer]

Have some more gloom. Keep in mind I'm not criticizing the price, just abyssmal gains on a new node relative to the other company.

AMD has been slapped in mobile and mass market. While we're sitting around wondering whether rx480 is 390x +/-20% the real intention of p10/11 - efficiency gains - has already been a massive success. nVidia is for once categorically behind amd in efficiency.

 

[rainy]

If it actually had 290 performance, they would likely advertise it that way, and not compare it to a 270X, imo.

I'm guessing that it is actually in the 270X ballpark overall.

Everyone knows that AMD is terrible in marketing and this is just another case.

 

[linkgoron]

If it actually had 290 performance, they would likely advertise it that way, and not compare it to a 270X, imo.

I'm guessing that it is actually in the 270X ballpark overall.

Nvidia did the same thing. They compared the 1080 to the 980. Now AMD is doing the same thing - 470 to 270x.

 

[LTC8K6]

They compare it to the 270 because it is the 470, really not that complicated.

Yeah, there's that.

But 280X/285 owners need a reason to upgrade.

If you show/tell me it has 290 class performance, I know it's faster than my 280X/285 card.

If you compare it to a 270X, I'm not sure it beats my 280X/285 card.
It sorta' sounds like it would be a side grade to a 280X/285 when you compare it to a 270X.

 

[crisium]

You sound like you're comparing different architectures, in the same way that CPU lads used ta with GHz.

No. I am not comparing Polaris to Pascal.

I am comparing Polaris to Hawaii. I am also comparing Pascal to Maxwell.

Then I am clearly seeing (keep in mind based half on rumours half on AMD's figures) that Nvidia has further pulled away from AMD.

I'm not a technical wizard, in fact quite the opposite. If you think the rumoured 230-265mm^2 or so Polaris 10 should have this massive deficit to the 314mm^2 Pascal while also concluding that that this deficit when compared to 28mm is equal or has lessened, not worsened, then convince me please. I want to believe.

Of course if it's a much smaller die, then w/e. Just working with what I can.

 

[LTC8K6]

Nvidia did the same thing. They compared the 1080 to the 980. Now AMD is doing the same thing - 470 to 270x.

I don't care what NV did or does, actually.

 

[LTC8K6]

With those charts we have seen, it looks like the 480 is 390X territory, and the 470 is 390 territory, in 3D Mark?

 

[VulgarDisplay]

All the tests we've seen so far have had drivers that are months old on an unreleased card.

Why can't we all just wait for benchmarks with launch drivers for the card and see what it can actually do before we spread fud.

Sent from my XT1575 using Tapatalk