Here's what AMD didn't want us to see - HAWX 2 benchmark

[OCGuy]

lets be fair here, what is apple to apple anyway?

AMD thinks apple to apple is 6870 vs 4601G, and who is comparing that?
Batmen disable AA on ATi card, where is the apple?
6870 consumes half of the power than 470, what is the apple?

the standards are made for the end-users and developers, thats whats important.

I compare the finished product, not what it takes to achieve that finished product.

If AMD is lowering IQ (not proven yet, I have a feeling we will see articles on this in the coming days to either prove or dispell) in order to appear faster in benchmarks, then we are not getting an "apples to apples" comparison.

 

[Lonyo]

This sholdnt surprise anybody. AMDs own graph shows their tesselation performance above factor 10ish starts to arrive near the 5000 series with these chips. In other words the tesselator is still weak compared to Nvidia's 400 series. It should be interesting to see if the 6900 series fixes this issue. But considering Tessealation was one of the 3 big additions in DX11 it is disappointing AMD didnt address this issue with these cards.

"Weak", maybe. Less powerful, certainly.

In other DX11 games which use tessellation, the HD6870 outperforms the GTX460 1GB OC (Metro 2033), loses by a small margin (Civ 5), is about equal (Stalker: COP, varies by resolution, HD6870 catches up as res increases), loses (Dirt 2, beats standard clocked GTX460).

So when other developers use DX11, it seems for their implementations (of DX11 as a whole, because they typically use more than just tess), the AMD tessellators on the HD6850/70 are sufficient for it to just about hold station around its given price point, relative to the GTX460.

The significant difference in this one test is therefore an anomaly, and makes AMD's claims that the implementation in the game could be improved seem reasonably valid, since such a difference in performance does not appear in other real world uses of DX11 features.
Also it's interesting to note that both cards perform acceptable well even at 1920x1200 with 8xAA, both getting over 60fps.

 

[Lonbjerg]

A lot of the FUD/defence for AMD in this thread iare debunked in this thread:

http://forums.anandtech.com/showthread.php?t=2113133

Funny how when AMD want to lower the I.Q. it dandy fine...had this been NVIDIA...forks and torches would have been out already...double standards are strong in this thread...

 

[Skurge]

"Weak", maybe. Less powerful, certainly.

In other DX11 games which use tessellation, the HD6870 outperforms the GTX460 1GB OC (Metro 2033), loses by a small margin (Civ 5), is about equal (Stalker: COP, varies by resolution, HD6870 catches up as res increases), loses (Dirt 2, beats standard clocked GTX460).

So when other developers use DX11, it seems for their implementations (of DX11 as a whole, because they typically use more than just tess), the AMD tessellators on the HD6850/70 are sufficient for it to just about hold station around its given price point, relative to the GTX460.

The significant difference in this one test is therefore an anomaly, and makes AMD's claims that the implementation in the game could be improved seem reasonably valid, since such a difference in performance does not appear in other real world uses of DX11 features.
Also it's interesting to note that both cards perform acceptable well even at 1920x1200 with 8xAA, both getting over 60fps.

Yeah, AMD even said that reviewers compare it to other games that have tess in them and that they would see an anomaly. It seems AMD was right.

 

[Red Storm]

I compare the finished product, not what it takes to achieve that finished product.

If AMD is lowering IQ (not proven yet, I have a feeling we will see articles on this in the coming days to either prove or dispell) in order to appear faster in benchmarks, then we are not getting an "apples to apples" comparison.

So then you would agree that we should wait for HAWX 2 to become a finished product before judging its performance, no? 🙂

 

[Skurge]

A lot of the FUD/defence for AMD in this thread iare debunked in this thread:

http://forums.anandtech.com/showthread.php?t=2113133

Funny how when AMD want to lower the I.Q. it dandy fine...had this been NVIDIA...forks and torches would have been out already...double standards are strong in this thread...

Really? Show me the forks and torches when the Distant LOD in Crysis was lower on the 480 than the 5870 and how about metro2033 when the 480 had lower quality textures than the 5870?

Yeah, it didn't happen. Who has double standards now?

 

[OCGuy]

Meh. nV bought nV-specific optimizations. It's the same marketing game as MS paying for timed exclusives on DLC, or Sony paying for an exclusive in-game character.

That wasn't your attitude on proprietary features when it came to Physx being locked out on AMD :

If you can't compete on quality and value, take your ball and run home crying. Stay classy, nvidia.


(9 of my last 10 card purchases were nvidia, until the 4870 last year.)

 

[dguy6789]

Nope, doesn't matter. AMD wants to water down DX11 so it "appears" faster.

🙄

"AMD has demonstrated to Ubisoft tessellation performance improvements that benefit all GPUs, but the developer has chosen not to implement them in the preview benchmark. For that reason, we are working on a driver-based solution in time for the final release of the game that improves performance without sacrificing image quality."

Careful, your bias is showing since you made the post above being fully aware of the quote I just posted.


Personal insults are not acceptable.

Re: "Careful, your bias is showing since you made the post above being fully aware of the quote I just posted."

These types of "in your face" negative comments are purely inflammatory and serve no positive purpose in a technical forum such as the VC&G.

Moderator Idontcare

 

[Skurge]

I just saw the benches.

I would hate to be playing a benchmark of an unreleased game maxed out at 1920x1200 8xAA at 70FPS with unoptimized drivers (as AMD have said). The fact that a 5770 is within 20fps of a 5870 should show you that it needs some work.

Seriously, what kind of bench is this? almost 200FPS at 1920x1200 with a GTX480? Does it even represent real gameplay?

Another thing, why do the Geforces lose 60FPS going from 0x to 8xAA, but the Radeons lose only about 15fps?

This thing looks iffy to me.

 

[AnandThenMan]

This is not a proper bench, and it's not even a game you can play. It contains code highly optimized by Nvidia. I'm sure AMD could easily produce a bench that shows their cards in the best possible light, but that would be a waste of time and sleazy to boot.

 

[Madcatatlas]

Sleazy indeed. Shame on you Nvidia, shame on you. (not that it was unexpected lol)

 

[Scali]

Yeah, AMD even said that reviewers compare it to other games that have tess in them and that they would see an anomaly. It seems AMD was right.

Funny how "AMD said" is taken as a truth by most people on this forum. You might want to think that over a bit more...

 

[Lonyo]

Funny how "AMD said" is taken as a truth by most people on this forum. You might want to think that over a bit more...

So... you're saying it's untrue?
So... where's the evidence against it? Because most benchmarks show there isn't a huge discrepancy in DX11 mode between the HD6870 and the GTX460. Which suggests that AMD's line that it's an anomaly is a tenable claim.

 

[Skurge]

Funny how "AMD said" is taken as a truth by most people on this forum. You might want to think that over a bit more...

Well, feel free to show me a game where the 6800 series performs as poorly as that one that also uses tessellation, not even in heaven on extreme is that gap that huge from a 6870 to a 460, intact they are equal.

I took it as truth as soon as I saw the benchmarks, I don't see anything wrong with that.

 

[Scali]

So... you're saying it's untrue?

Yes I am, read the other thread.

So... where's the evidence against it?

TessMark, Unigine Heaven, DX11 Detail Tessellation sample (AMD's own contribution to the DX SDK!), GPU Caps Viewer... you name it. Plenty of benchmarks showing a huge difference between nVidia and AMD when it comes to tessellation performance. HAWX 2 is just yet another benchmark demonstrating the same problem that AMD has.

 

[AnandThenMan]

TessMark, Unigine Heaven, DX11 Detail Tessellation sample (AMD's own contribution to the DX SDK!), GPU Caps Viewer... you name it. Plenty of benchmarks showing a huge difference between nVidia and AMD when it comes to tessellation performance. HAWX 2 is just yet another benchmark demonstrating the same problem that AMD has.

You play all of those games on a regular basis?

 

[Scali]

You play all of those games on a regular basis?

Read what I said. They all demonstrate a tessellation bottleneck. Stop being in denial. This isn't about games or benchmarks or whatever. This is about hardware limitations.

I think we're now seeing that there are indeed plenty of paid AMD shills on the forum as well. Hard evidence is denied left and right, AMD's PR spin is held as absolute truth, and nVidia is being blamed for everything. The cat is out of the bag guys, give it up.

 

[Lonyo]

Yes I am, read the other thread.



TessMark, Unigine Heaven, DX11 Detail Tessellation sample (AMD's own contribution to the DX SDK!), GPU Caps Viewer... you name it. Plenty of benchmarks showing a huge difference between nVidia and AMD when it comes to tessellation performance. HAWX 2 is just yet another benchmark demonstrating the same problem that AMD has.

Where's the anomaly?
No one would deny that in benchmarks that AMD has worse tessellation performance, but it has never applied to any real game ever until HAWX 2. Therefore HAWX 2 is an anomaly.

In the meantime we recommend you hold off using the benchmark as it will not provide a useful measure of performance relative to other DirectX® 11 games using tessellation.

So, where are the GAMES which show AMD's claim to be untrue?
I think you showed your own claim to be untrue (that AMD's was untrue) by only being able to specify synthetic benchmarks.

If saying something which turns out to be accurate is PR spin, then I must have gone to another dimension. Maybe you should have read what AMD said before saying they lied.

 

[dguy6789]

Yes I am, read the other thread.



TessMark, Unigine Heaven, DX11 Detail Tessellation sample (AMD's own contribution to the DX SDK!), GPU Caps Viewer... you name it. Plenty of benchmarks showing a huge difference between nVidia and AMD when it comes to tessellation performance. HAWX 2 is just yet another benchmark demonstrating the same problem that AMD has.

So you're saying it's normal for a 5870 and 5770 to perform similarly in this benchmark and AMD's comment should be considered nonsense despite a near 100% difference between the two cards in performance in nearly every other benchmark including ones you mentioned. Yeah, there can't possibly be anything wrong with the benchmark.

 

[Skurge]

Read what I said. They all demonstrate a tessellation bottleneck. Stop being in denial. This isn't about games or benchmarks or whatever. This is about hardware limitations.

I think we're now seeing that there are indeed plenty of paid AMD shills on the forum as well. Hard evidence is denied left and right, AMD's PR spin is held as absolute truth, and nVidia is being blamed for everything. The cat is out of the bag guys, give it up.

They all demonstrate a bottleneck on 5series cards, AMD wasn't talking about them, AMD was talking about the the 6800series.

 

[AnandThenMan]

Read what I said. They all demonstrate a tessellation bottleneck. Stop being in denial. This isn't about games or benchmarks or whatever. This is about hardware limitations.

Oh, I see. So it's not about games that I can actually buy and enjoy. It's about beating the tessellation drum to try and make AMD look bad. At least you're being honest.

 

[brybir]

Yes I am, read the other thread.



TessMark, Unigine Heaven, DX11 Detail Tessellation sample (AMD's own contribution to the DX SDK!), GPU Caps Viewer... you name it. Plenty of benchmarks showing a huge difference between nVidia and AMD when it comes to tessellation performance. HAWX 2 is just yet another benchmark demonstrating the same problem that AMD has.

Well to be fair, the things you listed are all benchmarks that demonstrate maximum theoretical performance of a particular feature. His point was show where it matters in other actual games, and I think so far, we have seen that it only makes a big difference in HAWX 2, so in that sense it appears to be an anomoly compared to other current games that use tesselation.

But within that you do have a good point in that while current implementations that are properly done perform equally well on both vendors cards, its possible over the next year more tesselation heavy games will come out that will do better on Nvidia's hardware as it stresses the tesselators more.

But, only time will tell, and with the ~1year refresh cycle, you never spend very long worrying about those problems.

 

[AnandThenMan]

I think we're now seeing that there are indeed plenty of paid AMD shills on the forum as well. Hard evidence is denied left and right, AMD's PR spin is held as absolute truth, and nVidia is being blamed for everything. The cat is out of the bag guys, give it up.

Seriously? Now you're making accusations?

 

[Evilhomer99]

Who the hell plays HAWX?

 

[Scali]

They all demonstrate a bottleneck on 5series cards, AMD wasn't talking about them, AMD was talking about the the 6800series.

AMD's own charts show that at best the tessellation (under synthetic conditions) is 2x 5000-series as fast, and dropping off to pretty much 1x 5000-series at tessellation factors above 11 (which is not all that high, TessMark considers 16x normal, and then you can still clearly see a bit of aliasing).
So 6000-series still suffers from a big tessellation bottleneck... slightly better than the 5000-series perhaps, but then again, that is exactly what the HAWX2-figures are showing, correct? They correlate just fine with AMD's own tessellation chart.