[HardOCP] Crysis 3 Benchmarks

[Termie]

Link: http://www.hardocp.com/article/2013/03/12/crysis_3_video_card_performance_iq_review

HardOCP has just published its in-depth performance review of Crysis 3. As usual, HardOCP looks only at higher-end configurations - 7870, 660Ti, 7950 Boost, 670, 7970 Ghz, 680, 7970 Crossfire, and 680 SLI.

A couple of interesting points:
(1) While the 7970 and 680 are dead even, the 670 does much better than the 7950 Boost in these benchmarks, which is an odd result. Perhaps it's leaning hard on the extra shaders in the 7970, whereas the 680's extra shaders aren't as critical. Edit: As suggested below, the issue may in fact be that while the 680 and 670 have the same memory bandwidth, there's a significant difference in memory bandwidth between the 7950 and 7970.
(2) HardOCP is finding very high Crossfire and SLI scaling. Previous benchmarks showed less than ideal scaling.

Note: there's an existing thread referencing these results, but it's specific to the HD7950 and 660Ti comparison.

 

[Elfear]

Thanks for posting Termie. Seems weird that the 7950 Boost is performing so much lower than the 7970Ghz card. In other games the difference is much smaller.

 

[guskline]

Termie: Check your PM

 

[chimaxi83]

Not surprising, TXAA is bad:

The third mode is the worse and most useless mode in this game, TXAA. If you thought FXAA blurs your textures, you will be shocked at how much TXAA blurs everything in the game. Looking at this game with TXAA enabled is like looking at your display with Vaseline smeared all over it. It causes a drastic reduction in texture quality, object quality, grass, tree, vegetation quality, and character quality. It doesn't matter if you use the TXAA 1 or TXAA 2 mode, both are nasty, visually speaking.

TXAA does improve to a great extent alpha texture aliasing and specular or shader aliasing. In fact, it is the best at doing it, but at the cost of completely destroying your gameplay experience. TXAA should be avoided, you should never use it in Crysis 3.

 

[BallaTheFeared]

Not really when you consider AMD took over >.<

 

[Plimogz]

Is TXAA ever much less blurry? The last I remember reading into it was with regards to The Secret World and the verdict there seemed to be that it also blurred the visuals pretty badly.

Regardless, I must say that it is pretty bold to blame AMD for the underwhelming performance of an NV feature, considering how often the exact opposite argument has been made by some and adamantly defended by others. The "there could've been some intermediate level of scripted particle effects run off of the CPU" debate from the Hawken physx thread comes to mind, for one.

 

[3DVagabond]

Link: http://www.hardocp.com/article/2013/03/12/crysis_3_video_card_performance_iq_review

HardOCP has just published its in-depth performance review of Crysis 3. As usual, HardOCP looks only at higher-end configurations - 7870, 660Ti, 7950 Boost, 670, 7970 Ghz, 680, 7970 Crossfire, and 680 SLI.

A couple of interesting points:
(1) While the 7970 and 680 are dead even, the 670 does much better than the 7950 Boost in these benchmarks, which is an odd result. Perhaps it's leaning hard on the extra shaders in the 7970, whereas the 680's extra shaders aren't as critical.
(2) HardOCP is finding very high Crossfire and SLI scaling. Previous benchmarks showed less than ideal scaling.

Note: there's an existing thread referencing these results, but it's specific to the HD7950 and 660Ti comparison.

Wonder why that was? :sneaky:

Anyway, thanks for posting the complete story.

 

[aaksheytalwar]

It is really just a I told you so with the 7970 ghz and 7950 boost. I always knew that the 7970 was worth it, much more than how the 6970 was compared to the 6950.

 

[ICDP]

It is really just a I told you so with the 7970 ghz and 7950 boost. I always knew that the 7970 was worth it, much more than how the 6970 was compared to the 6950.

I disagree, the 7950 (BE) is seriously underclocked compared to a 7970 GE and comparing both at stock doesn't give an accurate comparison. I have tested both cards extensively and found the HD7950 to be so close that I cut my losses and sold the HD7970 (and GTX680).

The stock clocks are 925/1250 compared to 1050/1450. That alone would account for between 15-20% of the performance difference.

Clock a HD7950 at GE speeds of 1050/1450 and it has around 7% less performance of a HD7970 for only 70% of the cost. Of the 3 HD 7950s I have tested not one got less than 1150 core clock so 1050 is preTty much guaranteed IMHO.

 

[Dark Shroud]

You can buy 7950s factory over clocked at 1100mhz and they bring it on the performance.

Yet no one ever really runs benches with those 7950s.

 

[Silverforce11]

It is really just a I told you so with the 7970 ghz and 7950 boost. I always knew that the 7970 was worth it, much more than how the 6970 was compared to the 6950.

Reduction in 7950 shaders alone give neglible performance difference vs 7970, its the big clock speed difference. The 670 and 680 both boost to similar speeds.

 

[Leadbox]

Reduction in 7950 shaders alone give neglible performance difference vs 7970, its the big clock speed difference. The 670 and 680 both boost to similar speeds.

The low memory clocks on the 7950 and 7870 must be whats hurting their performance, 5Ghz and 4.8Ghx respectively

 

[bgt]

I am running my 7950 @1250/1250 without a problem. The core speed makes the difference.

 

[Termie]

The low memory clocks on the 7950 and 7870 must be whats hurting their performance, 5Ghz and 4.8Ghx respectively

This is a very good theory. This would also explain why the 670 and 680 are so close - they have the same memory bandwidth.

Looks like Crysis 3 may be a good candidate for VRAM overclocking!

 

[Ferzerp]

And of course the obligatory reference to CFX issues.

With the Radeon HD 7970 GHz Edition CrossFire however, we found SMAA MGPU 2X provided what looked like fast enough framerates, but the actual feel of the game was choppy and sluggish at this setting. It was as if we weren't getting the full potential out of CrossFire acceleration anymore with SMAA MGPU 2X selected. Our only option then was to lower the setting one notch to SMAA Low 1X. At this setting, it was perfectly smooth, and faster in framerate. As you'll also find out later in this evaluation, SMAA Low 1X takes no performance hit at all, it is as if you are running with AA disabled performance wise, so that is why it is faster than GTX 680 SLI with SMAA MGPU 2X mode.

 

[Ares202]

This is a very good theory. This would also explain why the 670 and 680 are so close - they have the same memory bandwidth.

Looks like Crysis 3 may be a good candidate for VRAM overclocking!

Here is a good review to test to your theory

http://hexus.net/tech/reviews/graphics/40613-evga-geforce-gtx-670-ftw/

the 670 FTW has the same core speed as a GTX 680 - but has an extra 200mhz on the memory providing greater bandwidth than the 680, overall the 670 FTW outpeforms the standard 680 on most of the tests despite its crippled shaders.

 

[blackened23]

Here is a good review to test to your theory

http://hexus.net/tech/reviews/graphics/40613-evga-geforce-gtx-670-ftw/

the 670 FTW has the same core speed as a GTX 680 - but has an extra 200mhz on the memory providing greater bandwidth than the 680, overall the 670 FTW outpeforms the standard 680 on most of the tests despite its crippled shaders.

Oh so you want to compare an overclocked card to a non overclocked card. That is always the basis for a sound comparison.

 

[BallaTheFeared]

He's just showing that a slight memory boost went a long way, where it's slower than the 680 still it's slower than the 670 top which has higher core but lower memory, and faster than both when bandwidth was the bottleneck.

 

[Termie]

Here is a good review to test to your theory

http://hexus.net/tech/reviews/graphics/40613-evga-geforce-gtx-670-ftw/

the 670 FTW has the same core speed as a GTX 680 - but has an extra 200mhz on the memory providing greater bandwidth than the 680, overall the 670 FTW outpeforms the standard 680 on most of the tests despite its crippled shaders.

Oh so you want to compare an overclocked card to a non overclocked card. That is always the basis for a sound comparison.

He's just showing that a slight memory boost went a long way, where it's slower than the 680 still it's slower than the 670 top which has higher core but lower memory, and faster than both when bandwidth was the bottleneck.

Just to clarify, Ares202 was on the right track, but there was a problem with the methodology. While the GTX680 and GTX670 FTW have the same base clock, their boost clocks are quite different, so the 670 FTW actually gets an advantage not only in memory clock speed but also in core clock speed.

We'd have to have a true clock-for-clock core comparison to show that in some cases, more memory bandwidth can make up for less shader power.

In most situations, it takes a 670 core well above 1200MHz boost plus a memory overclock to match a stock 680.

 

[BallaTheFeared]

Darn you boost 1.0~!!!

 

[Ares202]

Oh so you want to compare an overclocked card to a non overclocked card. That is always the basis for a sound comparison.

overclocking is a pretty linear increase, so yes

Just to clarify, Ares202 was on the right track, but there was a problem with the methodology. While the GTX680 and GTX670 FTW have the same base clock, their boost clocks are quite different, so the 670 FTW actually gets an advantage not only in memory clock speed but also in core clock speed.

Fair point

 

[Will Robinson]

The low memory clocks on the 7950 and 7870 must be whats hurting their performance, 5Ghz and 4.8Ghx respectively

Radeons historically respond better to engine overclocking while leaving little margin to play with memory speed.
Tahiti is no exception and it really does need to be cranked up some over stock to really make it shine.

 

[BallaTheFeared]

It just seems strange to me that a card like the 7950 can match a 670 ans then continue to see scaling past 300 GB/s against a card with less.

Would titan be as fast as it is with just gk104 on a 384 bus??? I don't think so...

 

[omeds]

Not surprising, TXAA is bad:

TXAA looks amazing in this game if you bump up the sharpening filter (r_sharpening).

The sharpening filter being on by default even when not using post AA is also why they thought Transparency MSAA wasn't doing a whole lot.. Lower the sharpening filter and its as good as an other game, like normal.

 

[SirPauly]

http://www.gamerconfig.eu/command/crysis-3/r_Sharpening

r_Sharpening - Crysis 3

Controls the level of the post-processing image sharpening filter. Higher values give more sharpening, but too high values will create artifacts. A value of 0.0 is off

If some don't like the default image with TXAA -- they can sharpen it!