AMD V Nvidia by Richard Huddy

[Genx87]

You would, if you were to look at a chart that also included nVidia's GPUs.
This chart is not very impressive. At its peak, it's twice as fast as the previous gen, which still isn't very fast... and with higher tessellation factors (where the real point of tessellation is) it seems to drop back down quickly, to pretty much the same poor speeds of its predecessor.

Take a look at this for example: http://www.geeks3d.com/20100826/tessmark-opengl-4-gpu-tessellation-benchmark-comparative-table/ ... You can figure out for yourself where a 1.5-2x increase would put AMD... Neglect? I think a solid yes.

Thanks, so the 5770 did have the same performance as the 5870 in tesselation. That chart then I agree doesnt look very impressive at all. Little wonder why Huddy is out pre-empting the poor tesselation results that should come down the pipe if that chart is correct.

 

[Genx87]

Considering the 6800 is not high end part for AMD, I think it's good that their new mainstream product atleast matches their flagship from last year. Also we have no idea what Cayman will do but from the 5800, this is a solid upgrade you can't take that away from them.

Looking at Scali's link the 5770 matched the highend part last generation as well.

 

[Grooveriding]

You would, if you were to look at a chart that also included nVidia's GPUs.
This chart is not very impressive. At its peak, it's twice as fast as the previous gen, which still isn't very fast... and with higher tessellation factors (where the real point of tessellation is) it seems to drop back down quickly, to pretty much the same poor speeds of its predecessor.

Take a look at this for example: http://www.geeks3d.com/20100826/tessmark-opengl-4-gpu-tessellation-benchmark-comparative-table/ ... You can figure out for yourself where a 1.5-2x increase would put AMD... Neglect? I think a solid yes.

You're talking about the 6870, a mid-range card and the answer from AMD to the GTX 460. Some early benches show it matching the 460 in tessellation in Heaven. So you also feel the GTX 460 lacks in tessellation performance ?

Do you have early info as to how the 6970 is going to perform in tessellation, is so, please enlighten us as you are again speaking in absolutes about 'neglect in tessellation now, as if you have answers and information the rest of us are not privy to. Also, please advise the percentile breakdown of how Huddy spends his time at work, as you indicated you have that information as well.

I would love some early information on the 6970, as I'm considering replacing my GTX 480s with them. Please, enlighten me, I'd be grateful to hear how they have neglected tessellation in their new single-gpu halo product.

 

[Scali]

Considering the 6800 is not high end part for AMD, I think it's good that their new mainstream product atleast matches their flagship from last year.

Yes, let's totally ignore the competition.

Or wait, let's not...
Even with 2x the tessellation performance, the 5870 (which IS a high-end part), would still not be able to perform on par with the GeForce GTX460 (which is NOT a high-end part) with tessellation.

This 'upgrade' is in the category 'too little, too late'.

Aside from that, if the 6000-series is anything like the 5000-series (which it probably is, as it's just an update of the architecture), then high-end parts will be just as bottlenecked by the tessellator as the midrange parts, so then this tessellation will be 'as good as it gets'.

 

[badb0y]

Looking at Scali's link the 5770 matched the highend part last generation as well.

Hmm, that's interesting, but it doesn't translate the same way into games does it?

Yes, let's totally ignore the competition.

Or wait, let's not...
Even with 2x the tessellation performance, the 5870 (which IS a high-end part), would still not be able to perform on par with the GeForce GTX460 (which is NOT a high-end part) with tessellation.

This 'upgrade' is in the category 'too little, too late'.

Aside from that, if the 6000-series is anything like the 5000-series (which it probably is, as it's just an update of the architecture), then high-end parts will be just as bottlenecked by the tessellator as the midrange parts, so then this tessellation will be 'as good as it gets'.

Sorry dude but you can't say that concretely till the cards are out...

 

[Scali]

Hmm, that's interesting, but it doesn't translate the same way into games does it?

I have a strong suspicion that it translates to HAWX 2, given the stress it's causing in the AMD camp.

 

[Genx87]

Hmm, that's interesting, but it doesn't translate the same way into games does it?

If the game\benchmark is tesselator limited it should.

 

[nyker96]

damn you guys are flooding this site, like DoS attack, can't get into it now. :[

 

[Scali]

To put TessMark into perspective:
Moderate is tessellation factor 8.
Normal is tessellation factor 16.
Extreme is tessellation factor 32.
Insane is tessellation factor 64.

If we take the chart that Gideon posted earlier, we see that at 'normal' setting of 16, the performance of the 'improved' 6000-series has already dropped off to nearly the level of the 5000-series.
As I said: it doesn't really improve in the range where you'd want it to improve, the peak is at low tessellation factors only. With TessMark you can clearly see quality issues with the setting of 8. Normal looks much better. Going to Extreme and beyond may be debatable... but AMD's claim that their hardware can handle any 'sensible' tessellation factor without image quality loss is clearly false.

 

[PingviN]

"TESSELATION!!!!" <---- my take on your post(s)

Really? Alrighty then, let's go the practical route:

Today: how many games use heavy tesselation?

Tomorrow (meaning, the nearby future): how many games will use heavy tesselation? Up for speculation, but I'm guessing maybe one or two. Others who adds tesselation will probably use it in a more discrete fashion, as seen in Metro 2033 and AvP. Does GF10* have a noticable advantage here over it's competitor, despite the oh-so-mighty tesselation?

In the coming few years: how many games will use heavy tesselation? I bet quite a few. But then again, wont AMD and Nvidia already have refreshed their product lines? Oh yes, they will have. Will AMDs upcoming architecture be better at tesselation? Probably.


You can't really keep on pointing to GF10*'s superior tesselation as it's irrelevant until it can actually be used. If HD68*0 comes with enough tesselation power to carry AMDs lineup until Souther Island's release in 2011, wont that be enough? Seriously?

 

[RussianSensation]

Really?
You can't really keep on pointing to GF10*'s superior tesselation as it's irrelevant until it can actually be used. If HD68*0 comes with enough tesselation power to carry AMDs lineup until Souther Island's release in 2011, wont that be enough? Seriously?

Tessellation is SM3.0 repeated over and over again. When NV had SM3.0, very few games took advantage of it. SM3.0 performance was touted by NV as important but too few games took proper advantage of it. Eventually SM3.0 became available on AMD cards and more and more games started to look prettier. But even with SM3.0 support, AMD cards still performed poorly when soft shadows and HDR were enabled. Eventually both camps had fast enough cards that these additional graphical enhancements are now taken for granted because the performance hit is minimal with today's hardware.

For tessellation to take off, both NV and AMD should have good products that can support it before developers will take that next programming step. So improved tessellation for HD6000 is a plus. However, without GF100 series being far superior to HD5000 in its use of tessellation, I doubt AMD would have taken the first step to lead the industry into that direction or even bother improving it in HD6000 series.

No offence to AMD, but even my Radeon 8500 had simple tessellation (TruForm) and in the last 9 years ATI has done a horrible job in promoting or pushing this new method. The point is it's not irrelevant to push key features on the market. It's a chicken and an egg scenario.

 

[flopper]

This 'upgrade' is in the category 'too little, too late'.

Aside from that, if the 6000-series is anything like the 5000-series (which it probably is, as it's just an update of the architecture), then high-end parts will be just as bottlenecked by the tessellator as the midrange parts, so then this tessellation will be 'as good as it gets'.

ur seem to get to be correct on that crystalball gazing

 

[Grooveriding]

Yes, let's totally ignore the competition.

Or wait, let's not...
Even with 2x the tessellation performance, the 5870 (which IS a high-end part), would still not be able to perform on par with the GeForce GTX460 (which is NOT a high-end part) with tessellation.

This 'upgrade' is in the category 'too little, too late'.

Aside from that, if the 6000-series is anything like the 5000-series (which it probably is, as it's just an update of the architecture), then high-end parts will be just as bottlenecked by the tessellator as the midrange parts, so then this tessellation will be 'as good as it gets'.

http://translate.googleusercontent....le.com&usg=ALkJrhjyX3zKszyxzNIZNBlf6RZmR9bCCA

6870 is faster than GTX 460 on the Heaven Benchmark with Extreme tessellation.

I guess your certainty was misplaced.

 

[Lonbjerg]

http://translate.googleusercontent....le.com&usg=ALkJrhjyX3zKszyxzNIZNBlf6RZmR9bCCA

6870 is faster than GTX 460 on the Heaven Benchmark with Extreme tessellation.

I guess your certainty was misplaced.

Lets look again:

http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/17

 

[blastingcap]

Lets look again:

http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/17

I think he was talking about extreme tessellation running the benchmark and not when you use GPUs for nothing but tessellation. In that case, see:

http://benchmarkreviews.com/index.p...k=view&id=628&Itemid=72&limit=1&limitstart=11

Feel free to gloat about PhysX and CUDA and 3D gaming, though; AMD still has not caught up in those areas.

 

[HurleyBird]

Aside from that, if the 6000-series is anything like the 5000-series (which it probably is, as it's just an update of the architecture), then high-end parts will be just as bottlenecked by the tessellator as the midrange parts, so then this tessellation will be 'as good as it gets'.

Now that 6870 review are out it seems that your fundamental assumption is wrong. From [H]:

The 6900 series will have a superset of features compared o the 6800 series. This means that there will be features and architecture differences between 6900 and 6800 series. This allows AMD to take more chances on the high-end enthusiast class GPUs and architecture things different, to really step up performance that enthusiasts demand

It sounds like there is a lot more going on in Cayman than simply a scaled up Barts core. I have to assume that tessellation is one of the areas we'll see differences. If not I'll be very surprised, as well as disappointed.

 

[blastingcap]

Now that 6870 review are out it seems that your fundamental assumption is wrong. From [H]:

It sounds like there is a lot more going on in Cayman than simply a scaled up Barts core. I have to assume that tessellation is one of the areas we'll see differences. If not I'll be very surprised, as well as disappointed.

Cayman is supposed to have a significantly stronger tessellator than Barts, something like +33 to +50&#37;.

 

[HurleyBird]

Cayman is supposed to have a significantly stronger tessellator than Barts, something like +33 to +50&#37;.

Where did you get that from? As far as I'm concerned all previous Cayman rumors are out the window. I'll be pretty surprised if Cayman isn't at least 200% Barts in tesselation seeing as its a more advanced and somewhat new architecture. It also makes sense why HD68xx, the new *mid level* GPU is not focusing on improving tessellation performance on *extreme* settings. Barts is nothing if not meticulously focused and optimized for its specific place in the market.

 

[blastingcap]

Where did you get that from? As far as I'm concerned all previous Cayman rumors are out the window. I'll be pretty surprised if Cayman isn't at least 200&#37; Barts in tesselation seeing as its a more advanced and somewhat new architecture. It also makes sense why HD68xx, the new *mid level* GPU is not focusing on improving tessellation performance on *extreme* settings. Barts is nothing if not meticulously focused and optimized for its specific place in the market.

When you've been tracking rumors as long as I have (2 months), you pick up a few things and get a sense for who is reliable. I can't point you to a source right now (I'm too tired.. there is probably a slide out there about this topic though, feel free to search through the 51-page monster thread on forums for it) but if you want you can ask Silverforce. Or Neliz at B3D, who has been dead-accurate throughout all of this, right down to the suggested etail price of $179/239. Keep in mind that Cayman is not an upscaled Barts; Cayman has a different architecture with different shaders and more features.

You are right that 33-50% may be an underestimation though.

 

[Lonbjerg]

Cayman is supposed to have a significantly stronger tessellator than Barts, something like +33 to +50%.

That is still behind "Fermi's" tessellation power...FYI.

 

[blastingcap]

That is still behind "Fermi's" tessellation power...FYI.

http://forums.anandtech.com/showpost.php?p=30642508&postcount=165

 

[PingviN]

Tessellation is SM3.0 repeated over and over again. When NV had SM3.0, very few games took advantage of it. SM3.0 performance was touted by NV as important but too few games took proper advantage of it. Eventually SM3.0 became available on AMD cards and more and more games started to look prettier. But even with SM3.0 support, AMD cards still performed poorly when soft shadows and HDR were enabled. Eventually both camps had fast enough cards that these additional graphical enhancements are now taken for granted because the performance hit is minimal with today's hardware.

For tessellation to take off, both NV and AMD should have good products that can support it before developers will take that next programming step. So improved tessellation for HD6000 is a plus. However, without GF100 series being far superior to HD5000 in its use of tessellation, I doubt AMD would have taken the first step to lead the industry into that direction or even bother improving it in HD6000 series.

As long as consoles wont be able to run it, you can be damn sure the main bulk of developers wont bother implementing it. Both AMD and Nvidia has potent enough DX10, still, DX10 is not used. I'm saying Fermis tesselation advantage is irrelevant, because it doesn't show in anything but a few benchmarks. Tesselation is nice and all, and will probably do much for graphics in the future, but by then, there will be new GPUs on the field.

If developers were genuinly interested in tesselation, there would be more games by now out using it.

 

[Scali]

Lets look again:

http://www.anandtech.com/show/3987/...enewing-competition-in-the-midrange-market/17

Yup, exactly as I predicted: the tessellation boost is in too low tessellation factors only, so it basically is no improvement over 5870 in the extreme benchmark.

Also, why are we comparing to GTX460 to 5870/6870?
In benchmarks without tessellation, 5870/6870 are considerably faster than GTX460, so why are some people pretending that it's good that 5870/6870 are *only* as 'fast' as the GTX460 in tessellation benchmarks? It should be competing with GTX470 at least.

 

[Scali]

I have to assume that tessellation is one of the areas we'll see differences.

So my assumption is wrong because you have a different assumption?

If not I'll be very surprised, as well as disappointed.

For me it's the other way around: I'd be surprised if they have developed a parallel tessellator, and for some reason decided NOT to put it in the 6800-series (especially since their competitor puts parallel tessellators in their entire range).

 

[Lonbjerg]

Yup, exactly as I predicted: the tessellation boost is in too low tessellation factors only, so it basically is no improvement over 5870 in the extreme benchmark.

Also, why are we comparing to GTX460 to 5870/6870?
In benchmarks without tessellation, 5870/6870 are considerably faster than GTX460, so why are some people pretending that it's good that 5870/6870 are *only* as 'fast' as the GTX460 in tessellation benchmarks? It should be competing with GTX470 at least.

Now I am just waiting for how long it will take before we see a driver from AMD, that introduces tessellation cheats per default...to make their performance look better...