Radeon RX 480 vs Geforce GTX 970: CPU Scaling

[Sweepr]

Yep, I posted their 480 CPU scaling as the 2nd post in this thread, but it is getting ignored.

It is getting ignored because it's useless for comparisons. If you have more gaming results comparing both VGAs with APU, Core i3, Pentium and Celeron CPUs that many people actually use with $200 VGAs, feel free to add them.

I hope reviewers include low-end CPU testing in Geforce GTX 1060 reviews as well.

 

[Bacon1]

It is getting ignored because it's useless for comparisons. If you have more gaming results comparing both VGAs with APU, Core i3, Pentium and Celeron CPUs that many people actually use with $200 VGAs, feel free to add them.

I hope reviewers include low-end CPU testing in Geforce GTX 1060 reviews as well.

I didn't realize that the 2500k was too powerful these days, thought it was one of the most popular CPUs out there.

The review you posted didn't include Pentium or Celeron CPUs, so guess that means it is useless as well. Guess you should remove it.

Oh but here are tests that include Pentium, I3, Celeron and more:

https://www.computerbase.de/2016-05/intel-core-i7-6950x-6800k-test/7

And whats that, the Fury X scales fine against the 980 TI...

Hell in a CPU heavy game (Total War) it goes from 5% slower with a i5-5675C (faster than the i7s on 980 TI) to 4% faster on the A10-7860K (slowest for both cards).

So if anything, you could say that Nvidia has more CPU scaling issues than AMD.

But I don't expect the results from 26 CPUs to make it into your OP.

Witcher 3 with a 960 vs 380 AMD goes from 16% faster @ top end CPU to 23% faster on an AMD FX chip.

So all I see are results showing that AMD scales better on lower end CPUs than Nvidia. :hmm:

 

[BSim500]

I recently came across a review that compared gpu perf with different CPUs. They had pair of GPUs with relatively close performance from both vendors and a lot of CPUs. It showed perfectly that there is no "amd driver overhead killing OCed haswells" effect.

Try looking beyond just one site testing just one game:-
http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/cpu_960/f4b_1920u.png
http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/cpu_960/gta5p_1920h.png
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/8.bmp.jpg
http://images.eurogamer.net/2013/articles//a/1/7/3/2/6/6/2/2.bmp.jpg
http://i.imgur.com/RpS8qfG.png

Also mentioned in Eurogamer review guidelines : "The standard way of benchmarking graphics cards is to pair them with an overclocked Core i7 in order to isolate pure GPU performance. However, this ignores that entry-level cards are more likely to be paired with less capable processors, so we've included Core i3 benchmarks too. You'll see from the minimum frame-rates that the GTX 750 Ti holds more of its performance than the R7 360, and that's because AMD's driver actually consumes quite a lot more CPU power than its Nvidia equivalent - something to be aware of at the budget end of the market."
http://www.eurogamer.net/articles/digitalfoundry-2016-graphics-card-upgrade-guide?page=2

Looking at OP's charts, the RX 480 is around 18% slower in FO4 & GTA V and nearer 28% slower in Crysis 3 vs the GX 970 when both are running with identical i3 CPU. (And for same games with i5-4460's, respectively around 5-15% and 21% slower). Likewise in the PCLab tests, you think it's normal a GTX 960 thrashes a 380X by 33% in Fallout 4 when matched with an i3, yet with an i7-6700K the results are normal?

The "gap" is precisely the DX11 driver overhead bottleneck effect being far worse in AMD's drivers on lower end CPU's when there isn't the surplus CPU power to "brute force" it through, causing it to bottleneck much earlier. Most tech sites "hide" this by benchmarking only with i7's "to eliminate CPU bottlenecks", which is fine for isolated analysis of a particular card. But for budget buyer's who will be using those lower end CPU's, doing only that with no additional real-world low-end CPU tests can be hugely misleading to the tune of 20-30% "missing" performance. I've seen the effect myself. When my HTPC's (i3) 7790 card was dying I "sidegraded" to a 750Ti. Despite the fact the 750Ti should be no more than +10% faster, some games were up to 35% faster. When capped at 60fps, CPU usage with the 750Ti was generally around 10-20% lower. It doesn't show up in all games, and it hits minimums more than averages, but the effect is absolutely there.

 

[Sweepr]

I didn't realize that the 2500k was too powerful these days, thought it was one of the most popular CPUs out there.

The review you posted didn't include Pentium or Celeron CPUs, so guess that means it is useless as well. Guess you should remove it.

Why the anger? In all likelihood the Radeon RX 480 would fare even worse with these chips if a Core i3 is struggling.

I'm afraid your ComputerBase tests with 2 games and no Geforce GTX 970/980 or Radeon RX 480 results doesn't add anything new either.

If you want to derail the thread even further, I suggest looking at PCLab results.

AMD VGAs do look worse relative to NVIDIA with slower CPUs, and most reviews hide this fact by using OCed Core i7.

 

[Bacon1]

Amazing how the goal posts keep changing.

Are you saying that the 480 doesn't scale similar to other GCN products, and that the 970 doesn't scale similar to maxwell either?

 

[Sweepr]

Amazing how the goal posts keep changing.

Yes, you should try staying on topic. I'm afraid it's a lost cause, everybody already knows about AMD's driver overhead.

 

[Bacon1]

My results are from last June 2016, not 6+ months ago.

I've proven that they scale fine.

Yes, you should try staying on topic. I'm afraid it's a lost cause, everybody already knows about AMD's driver overhead.

Ironic when you continue to post old results that aren't representative of the 480.

 

[BSim500]

Are you saying that the 480 doesn't scale similar to other GCN products, and that the 970 doesn't scale similar to maxwell either?

My results are from last June 2016, not 6+ months ago.

What people are saying is cherry picking the same two games that clearly don't suffer from it, then repeating the links over and over as some form of "lack of evidence due to sampling bias is anti-evidence" false-equivalence is looking silly when other games clearly do suffer from it, when AMD's DX11 driver overhead is a long-term widely known and commented on issue, and when several other sites post proof of the same issue in the same game independently.

 

[Sweepr]

I've proven that they scale fine.

ComputerBase just picked two games with less scaling problems, doesn't disprove any of the other results. And in this scene tested by PCLab AMD actually loses more performance:

 

[Bacon1]

What people are saying is cherry picking the same two games that clearly don't suffer from it, then repeating the links over and over as some form of "lack of evidence due to sampling bias is anti-evidence" false-equivalence is looking silly when other games clearly do suffer from it, when AMD's DX11 driver overhead is a long-term widely known and commented on issue, and when several other sites post proof of the same issue in the same game independently.

Are you trying to say that the OP and other results aren't cherry picked?

ComputerBase just picked two games with less scaling problems, doesn't disprove any of the other results. And in this scene tested by PCLab AMD actually loses more performance:

Yet again, results from over 6 months ago. Got anything recent?

How exactly does Total War not have scaling issues? Performance drops 50% as CPU gets weaker.

Also here is GTA seeming to scale fine:

There are some super old results (April 19, 2015) that show it scaling fine in GTA V since you wanted another game

http://www.gamersnexus.net/game-bench/1911-gta-v-cpu-benchmark-4790k-3570k-9590-more

Got any quality sources or at least updated ones?

 

[USER8000]

LOL,why are people pimping the GTX970 a few weeks before it will be replaced by the GTX1060??

I would not buy a GTX970 now - even in the UK,the RX480 4GB was only £175 from Overclockers UK,and Ebuyer has it on pre-order for £173 now. Cheapest I have seen a GTX970 is £200,and it is soon to be EOL,with Nvidia concentrating on driver updates for the Pascal cards.

Plus,if the GTX970 performs poorly in Tombraider which is the first major Nvidia DX12 title,it does not bode well for another year down the line.

 

[Bacon1]

 

[Sweepr]

http://i.imgur.com/8ZLCViT.jpg
[img]http://i.imgur.com/8DvkfIq.jpg
[img]http://i.imgur.com/gKYFnKJ.jpg
[img]http://i.imgur.com/0PhUeGo.jpg[/QUOTE]

So no Core i3, Pentium, Celeron or APU results, except the two games tested by ComputerBase which we all know have less scaling issues than the ones tested by PCLab. Why are you repeating quad-core CPU results you already posted, running out of data?

[quote]Yet again, results from over 6 months ago. Got anything recent?

[/quote]

Do you have more recent tests comparing Radeon RX 480/R9 390 and Geforce GTX 970/980 in the same games PCLab tested? Your The Witcher 3 results don't prove anything, they most likely tested a different scene, example:

[img]http://pclab.pl/zdjecia/artykuly/chaostheory/2016/01/gfx/cpu_970/w3f_1920u.png

Also here is GTA seeming to scale fine:

It does.

 

[Magic Carpet]

LOL,why are people pimping the GTX970 a few weeks before it will be replaced by the GTX1060??

I would not buy a GTX970 now - even in the UK,the RX480 4GB was only £175 from Overclockers UK,and Ebuyer has it on pre-order for £173 now. Cheapest I have seen a GTX970 is £200,and it is soon to be EOL,with Nvidia concentrating on driver updates for the Pascal cards.

Only can think of two reasons why somebody would still choose a Maxwell GPU. Windows XP support and VGA output. None of which should be the reason today, but maybe it could matter for someone.

 

[BSim500]

LOL,why are people pimping the GTX970 a few weeks before it will be replaced by the GTX1060??

No-one's "pimping" the 970 specifically. The main purpose of the thread is to point out the woeful lack of testing of the new mid-range dGPU's on anything short of an i7-6700K (which is significant when the results don't scale downwards equally on lesser CPU's for low/mid range target markets). The further away a component is from high end, the more people want to see it match up in real world scenarios (in addition to regular "bottleneck isolation" tests). Neither the driver overhead issue nor complaints are new, but the laziness / lack of common sense of tech review sites stays the same (and not just GPU reviews).

Are you trying to say that the OP and other results aren't cherry picked? Got any quality sources or at least updated ones?

There are plenty more examples. Dozens. On multiple sites:-
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/5.bmp.jpg
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/2.bmp.jpg
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/7.bmp.jpg
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/8.bmp.jpg

Likewise your claim "It doesn't show up on APU's" is kinda back to front, when they're shown to be the worst hit:-
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/9.bmp.jpg
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/a.bmp.jpg

The date of the article is irrelevant when the issue is ongoing and visible even in RX 480 reviews. That's just an excuse to try and exclude any data set for purely arbitrary reasons. I'm not going to go digging any more for a 20th, 50th, 100th piece of data, when I get the feeling you can't find what you're "looking" for, because you simply don't want to look... :sneaky: Now you're spamming off irrelevant i5-2500K charts which really aren't "budget" or "low end" CPU's and obviously won't show up the issue. If you genuinely don't understand the issue, then the biggest red flag is this from the DX12 article:-

Ignore DX12 and Mantle (black & blue bars) and just focus on the red & orange bars. Compare 750Ti vs R9 290X. That's not hardware, that's AMD's DX11 driver overhead. The effect of it gets hidden on high-end CPU's but rears it's ugly head on budget ones. Shouting "AMD's have no driver overhead penalty with DX11" is borderline trolling when virtually every DX12 / Mantle article released plus multiple GPU benchmarks on low-mid CPU's openly display the contrary, and even AMD admits that's the primary reason for creating Mantle in the first place, and a large chunk of the reason AMD benefits relatively more from DX11 -> 12 than nVidia.

You need to calm down and accept there are some things AMD are better at (DX12, pricing) and some things nVidia are better at (DX11 driver overhead, power efficiency) and learn to not take every thread which doesn't offer 100% gushing praise 100% of the time to "your side" and not 100% boo "the enemy", as some threat to the continued existence of your soul essence...

 

[Bacon1]

So no Core i3, Pentium, Celeron or APU results, except the two games tested by ComputerBase which we all know have less scaling issues than the ones tested by PCLab. Why are you repeating quad-core CPU results you already posted, running out of data?

You posted 4 results with 480, I posted 4 results with 480 I just updated to show the actual charts since most people won't click on the link to read the review.

So no Core i3, Pentium, Celeron

I don't see those in your last graphs either, but they were in my other post if you missed them.

Wonder why you are pushing so hard for outdated Nvidia tech though...

 

[Bacon1]

There are plenty more examples. Dozens. On multiple sites:-
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/5.bmp.jpg
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/2.bmp.jpg
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/7.bmp.jpg
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/8.bmp.jpg

Likewise your claim "It doesn't show up on APU's" is kinda back to front, when they're shown to be the worst hit:-
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/9.bmp.jpg
http://images.eurogamer.net/2013/articles//a/1/7/3/4/1/9/6/a.bmp.jpg

Holy 2013 batman, wow... I mean really...

I posted June 2016 detailed testing and you are going back to 2013 to find CPU bottlenecks?

 

[Silverforce11]

If you want to talk about CPU scaling, you should pick data from a more trustworthy site, such as Computerbase.de that actually do test CPU scaling in some of their articles.

PCLab.Pl's results are as reliable as ABT and we all know how reliable that is... not. Every time I read PCLab, it seems like their results are skewed both ways, AMD = slower, NV = faster, comparatively compared to major tech sites. Have you not noticed the same trend?

It's the opposite of Techaltar, that skews it AMD = faster, NV = slower. lol

 

[Silverforce11]

And if you want to show AMD in a bad light, it's very easy, you pile on GameWorks or PhysX titles, like the older COD, Project Cars, RotTR etc and suddenly their GPU look utter crap, especially on a slower CPU.

Is it AMD's drivers or is it NV GameWorks & PhysX?

Hard to see such massive performance drops in neutral titles, certainly uncommon.

The answer btw, is a combination of both. AMD's single threaded DX11 driver is partly to blame, because games that are single thread bottlenecked or have features (such as PhysX) that hammers the main thread, will wreck AMD performance. That's their weakness and it's why NV is so keen on exploiting it in games that it sponsors. However, on optimized games, it's a non-issue.

 

[Sweepr]

You posted 4 results with 480, I posted 4 results with 480 I just updated to show the actual charts since most people won't click on the link to read the review.

You posted results from two games, one of which might or might not show any difference depending on the scene tested, as already exposed by PCLab.

Meanwhile, I posted Radeon RX 480 results in four different games.

Are you trying to say that the OP and other results aren't cherry picked?

Four at the OP (Radeon RX 480), another three from PCLab and even more posted by BSim500 (different VGAs), while you posted two - one of which I just invalidated above. Yes, that's what I call cherry picking.

I don't see those in your last graphs either, but they were in my other post if you missed them.

OP has Haswell Core i3 results. Radeon RX 480 would only do worse with 2C/2T CPUs.

Wonder why you are pushing so hard for outdated Nvidia tech though...

Nodoby is pushing anything, we are discussing the fact that review results might not reflect end user experience using cheaper processors, and you're getting agressive because you don't like the results.

 

[Silverforce11]

You are trying too hard. Just use Rise of the Tomb Raider, Geothermal Valley level. AMD GPUs tank in performance due to CPU bottlenecks, GPU load drops.

Or use Project Cars.

It's well known.

 

[USER8000]

No-one's "pimping" the 970 specifically. The main purpose of the thread is to point out the woeful lack of testing of the new mid-range dGPU's on anything short of an i7-6700K (which is significant when the results don't scale downwards equally on lesser CPU's for low/mid range target markets). The further away a component is from high end, the more people want to see it match up in real world scenarios (in addition to regular "bottleneck isolation" tests). Neither the driver overhead issue nor complaints are new, but the laziness / lack of common sense of tech review sites stays the same (and not just GPU reviews).

Because it is still pimping it - I have a Maxwell based card currently,and would not consider touching a GTX970 at all now. The RX480 4GB is cheaper,the RX480 8GB has more VRAM and has far more future growth potential.

Looking at the Tombraider results,the GTX980 and even the RX480 seems more consistent in DX12,and more and more of those titles will be released in the next two years.

Then you have the GTX1060 coming out in the next week or so which will be faster,have more VRAM and better DX12 support. Overclockers UK has said the Sapphire RX480 Nitro 8GB will be £250. GTX1060 6GB cards will probably be around the same and both will probably at least match a GTX980. So why does anyone care about an ancient card which has come to the end of its life??

In more and more newer games I see both leaving the GTX970 behind.

Why not compare the GTX1060 and RX480 driver overhead then - seems more valid a comparison when the GTX1060 is released. I hardly know anyone who really wants a GTX970 if they are spending money in that price range - its either an RX480 or a GTX1060 when it is released.

I remember when the HD7870 cards were out,people were making comparisons back then to the GTX570,as the latter was competitive and was a bit cheaper,and it was further priced reduced when the GTX660 was released. In the end the HD7870 and GTX660 were the longer lasting cards - the same will happen with the RX480 and GTX1060.

 

[poofyhairguy]

These charts are perfect! Thank you!

I was looking at them last night and it's clear to me that the 480 is the BEST midrange AMD card today for CPU scaling. Look how the 480 benefits less than the 390 or especially the 390x when the CPU is the modern overclocked one.

This thread is kinda funny because if anything the 480 is the best CPU resistant (if you will) card AMD has put out in years. It's a great budget card.

 

[Azix]

Not seeing what each bar color represents

 

[Bacon1]

You posted results from two games, one of which might or might not show any difference depending on the scene tested, as already exposed by PCLab.

Since you are having trouble counting

https://www.computerbase.de/2016-06/radeon-rx-480-test/7/

The 480 scales better than older cards and well... is using 6 months newer drivers than anything you've managed to find as well.