Nvidia Fermi versus radeon 5800 benchmarks out!

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

ZimZum

Golden Member
Aug 2, 2001
1,281
0
76
He is getting more and more mad because his "broke and cant be fixed" article is going to be funny on launch day when it sells out.

Given that reports put the number of cards that will be available at launch will be under 10k. Its a virtual certainty it will sell out. But that won't prove or disprove any of Charlie's claims about the major issues with Fermi. Nvidia could put a ham sandwich in a fancy box and people would still buy it.
 

OCGuy

Lifer
Jul 12, 2000
27,227
36
91
Given that reports put the number of cards that will be available at launch will be under 10k. Its a virtual certainty it will sell out. But that won't prove or disprove any of Charlie's claims about the major issues with Fermi. Nvidia could put a ham sandwich in a fancy box and people would still buy it.

If the ham sandwich out-performs whatever the competition is selling, and is worth it for the price, I would buy it too. But I would try not to eat it.
 

bryanW1995

Lifer
May 22, 2007
11,144
32
91
you think that in 2 months we'll have a sandwich gtx 495 when they have publicly stated that they won't be available in volume for several months? also, what have you heard about potential 5890 and 5950? I haven't heard jack squat about those cards or even if they'll be made. Why would amd bother to release a respin if they couldn't beat the gtx 480? I see the logic of gtx 460 however, and I wouldn't be surprised to see it out fairly soon.
 

Keysplayr

Elite Member
Jan 16, 2003
21,209
50
91
If the ham sandwich out-performs whatever the competition is selling, and is worth it for the price, I would buy it too. But I would try not to eat it.

Last ham sandwich I've eaten had quite a bit of bandwidth. It actually hit about 4000 tbps (taste buds per second). Maybe a little pepper would bring it over the top! I'll give it a go with my next sammich.
 

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
You mean the 8800?

No, the 6800. nVidia has supported tesselation for a long time, I'm not sure why there is a ton of hype about it now, the only new thing is MS finally decided that DX was worthy of getting an update to support it.

That since you can use the shaders to do tesselation that mean NVIDIA had a tesselation unit?

Fermi does not have a tesselation unit. It handles tesselation via shader hardware, as the 6800 did. My point is rather simple, nVidia has supported the feature for quite a few years now, they simply didn't bother promoting it as it had no developer support. The only difference between their implementation and ATi is that ATi went the retro route and used dedicated hardware which is in stark contrast to their hype about the vast superiority of flexible units. Personally I think there are very clear trade offs no matter which way you go, it just goes against their PR hype of a few years ago.
 

GaiaHunter

Diamond Member
Jul 13, 2008
3,628
158
106
No, the 6800. nVidia has supported tesselation for a long time, I'm not sure why there is a ton of hype about it now, the only new thing is MS finally decided that DX was worthy of getting an update to support it.
The link you gave is about the 8800.
 

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
The link you gave is about the 8800.

Yep, wrong doc link, can't find the 6800 one at the moment although looking for it I ran across the old nV3x's tesselation support also under Gelato(full shader based render path, way slower then real time)- although it was far inferior to TruForm it was around back then too.
 

Grooveriding

Diamond Member
Dec 25, 2008
9,107
1,260
126
Feel bad for those who waited 6 months for a card 15% faster than a 5870 for a likely higher price. And more sorry for those who will wait another two months for a dual gpu fermi card that will be 15% faster than a 5970, and likely more expensive.

Doubly sorry for them when ATI releases their new arch. end of this year and they wait another six months for NV to catch up again :D
 

ScorcherDarkly

Senior member
Aug 7, 2009
450
0
0
He is getting more and more mad because his "broke and cant be fixed" article is going to be funny on launch day when it sells out.

Nvidia fans are going to look like morons whenever Fermi sells out on the first day. After having a field day with ATi's supply problems, you can't turn around and claim Nvidia selling out the first day is a result of fantastic sales/performance/architecture/etc, whenever they only shipped 10k units. But, I predict that's exactly what you'll do, and look like a fool in the process. Of course that's nothing new for the green team these days, so what's the difference, right?
 

DefRef

Diamond Member
Nov 9, 2000
4,041
1
81
Apple could put a ham sandwich in a fancy box and people would still buy it, paying twice as much as other ham sandwiches cost and to make it as useful as a regular ham sandwich, installing a 2nd ham sandwich on it.
ftfy.
 

evolucion8

Platinum Member
Jun 17, 2005
2,867
3
81
Last ham sandwich I've eaten had quite a bit of bandwidth. It actually hit about 4000 tbps (taste buds per second). Maybe a little pepper would bring it over the top! I'll give it a go with my next sammich.

But isn't adding pepper might bottleneck your tongue's ROP? You will be taste interpolation limited!!
 

ViRGE

Elite Member, Moderator Emeritus
Oct 9, 1999
31,516
167
106
Fermi does not have a tesselation unit. It handles tesselation via shader hardware, as the 6800 did. My point is rather simple, nVidia has supported the feature for quite a few years now, they simply didn't bother promoting it as it had no developer support. The only difference between their implementation and ATi is that ATi went the retro route and used dedicated hardware which is in stark contrast to their hype about the vast superiority of flexible units. Personally I think there are very clear trade offs no matter which way you go, it just goes against their PR hype of a few years ago.
2 things:

1) I'm not sure where this "NVIDIA doesn't have a tesselator" meme comes from. It's very clear in the GF100 articles that NVIDIA has a hardware tesselator; 16 of them in fact.

2) DX11 tessellation impacts the shaders by design. Hull and Domain shading in the tessellation process runs on the shaders. So you're going to impact the shaders on any architecture when using tessellation.
 

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
1) I'm not sure where this "NVIDIA doesn't have a tesselator" meme comes from. It's very clear in the GF100 articles that NVIDIA has a hardware tesselator; 16 of them in fact.

You must be talking about the polymorph engine which also handles vertex fetch, transformation, vertex, hull, domain and geometry shaders, transformation and setup. I would call it a bit of a stretch to call that dedicated tesselation hardware, wouldn't you?

2) DX11 tessellation impacts the shaders by design. Hull and Domain shading in the tessellation process runs on the shaders. So you're going to impact the shaders on any architecture when using tessellation.

Of course, but anytime you have functional units dedicated to one task it is going to offer superior performance for that singular tast in an ideal case. If a nV part is not utilizing tesselation the polymorph unit can be used for other shader tasks, ATi's setup can't. When ATi is handling tesselation they are using less of their overall shader resources then nVidia. It is a tradeoff and one I wouldn't say is right or wrong one way or the other, but it is very much against the hype ATi was pushing for unified shaders not all that long ago.
 

ViRGE

Elite Member, Moderator Emeritus
Oct 9, 1999
31,516
167
106
You must be talking about the polymorph engine which also handles vertex fetch, transformation, vertex, hull, domain and geometry shaders, transformation and setup. I would call it a bit of a stretch to call that dedicated tesselation hardware, wouldn't you?
Not at all. Everything I've read on the matter implies that the entire PolyMorph engine is a fixed-function block. Everything there is a fixed function process with transistors dedicated to it (and Hull/Domain shading being sent to the SPs).

sm.png


Of course, but anytime you have functional units dedicated to one task it is going to offer superior performance for that singular tast in an ideal case. If a nV part is not utilizing tesselation the polymorph unit can be used for other shader tasks, ATi's setup can't. When ATi is handling tesselation they are using less of their overall shader resources then nVidia. It is a tradeoff and one I wouldn't say is right or wrong one way or the other, but it is very much against the hype ATi was pushing for unified shaders not all that long ago.
The PolyMorph engine doesn't handle shader tasks, so there's no trade-off taking place.
 

BenSkywalker

Diamond Member
Oct 9, 1999
9,140
67
91
The first stage begins by fetching vertices from a global vertex buffer. Fetched vertices are sent to the SM for vertex shading and hull shading. In these two stages vertices are transformed from object space to world space, and parameters required for tessellation (such as tessellation factor) are calculated. The tessellation factors (or LODs) are sent to the Tessellator.

In the second stage, the PolyMorph Engine reads the tessellation factors. The Tessellator dices the patch (a smooth surface defined by a mesh of control points) and outputs a mesh of vertices. The mesh is defined by patch (u,v) values, and how they are connected to form a mesh.

The new vertices are sent to the SM where the Domain Shader and Geometry Shader are executed. The Domain Shader calculates the final position of each vertex based on input from the Hull Shader and Tessellator. At this stage, a displacement map is usually applied to add detailed features to the patch. The Geometry Shader conducts any post processing, adding and removing vertices and primitives where needed. The results are sent back to the Tessellation Engine for the final pass.

In the third stage, the PolyMorph Engine performs viewport transformation and perspective correction. Attribute setup follows, transforming post-viewport vertex attributes into plane equations for efficient shader evaluation. Finally, vertices are optionally “streamed out” to memory making them available for additional processing.

On prior architectures, fixed function operations were performed with a single pipeline. On GF100, both fixed function and programmable operations are parallelized, resulting in vastly improved performance.

http://www.bjorn3d.com/read.php?cID=1778&pageID=8317
 

ViRGE

Elite Member, Moderator Emeritus
Oct 9, 1999
31,516
167
106

Tempered81

Diamond Member
Jan 29, 2007
6,374
1
81
What difference would it make if tessellation was fixed function or a programmable operation? What matters is that it exists and it's 10x faster than ATI's rendition which has been around since 2006 or earlier.
 

SlowSpyder

Lifer
Jan 12, 2005
17,305
1,001
126
You must be talking about the polymorph engine which also handles vertex fetch, transformation, vertex, hull, domain and geometry shaders, transformation and setup. I would call it a bit of a stretch to call that dedicated tesselation hardware, wouldn't you?



Of course, but anytime you have functional units dedicated to one task it is going to offer superior performance for that singular tast in an ideal case. If a nV part is not utilizing tesselation the polymorph unit can be used for other shader tasks, ATi's setup can't. When ATi is handling tesselation they are using less of their overall shader resources then nVidia. It is a tradeoff and one I wouldn't say is right or wrong one way or the other, but it is very much against the hype ATi was pushing for unified shaders not all that long ago.


Maybe you could clarify this for me... I know Fermi has their polymorph engine, but it seems to be a bit of a mystery in if it actually is a 'hardware tesselator' in the more traditional sense. Do you know if the shaders still do the 'heavy lifting' for tesselation? Or does Fermi have a number of tesselators that are similar to AMD's (but just more of them)?

If Fermi is using the shaders I would think there would be times when that would hurt performance, then again times when it would offer superior performance depending on the workload. But overall I'd think the shaders would typically be saturated already, so that would ask them be used for tesselation instead of other tasks. Obviously if this will drop FPS from 130 to 110 it won't be a big deal. But in cases where you go from 35FPS to 27FPS it would make a difference... but then again AMD's parts take a hit as well.

Anyway, I thought maybe you could give a yes/no answer and clear this up a bit. Thanks.
 
Last edited: