ATi 5850/5870 review thread

[Janooo]

Originally posted by: BFG10K

Originally posted by: Idontcare

But Janooo isn't Cypress a wider GPU? Meaning it can process those A1/A2, etc, in parallel just as readily as two RV770's would? Everything was doubled up.

No, it?s doubled up but it can still only work on one frame at a time, unlike an AFR system. This is the inherent difference between a single GPU and AFR. With AFR, each GPU in the system can be working on a different frame concurrently.

But it doesn?t matter as much as Janooo thinks because the CPUs still have to finish working on the first frame before they start the second one. CPUs don?t work in ?AFR?, they work in ?SFR? (very loosely of course, but it gets the basic point across).

That's fine for game threads. What about driver threads? The frames are interleaved.
'When a game has profile' and it scales well, it's aware of a multi-gpu solution and it issues DX API calls that run in parallel otherwise we would not see the benefits.
My point is that a multi-gpu solution benefits more from the multi-core CPU than a single-gpu. I hope it makes sense.

 

[MODEL3]

lol,

what is going here?

AFR technique delivers higher frame rate than a 2X GPU? (everything 2X)
A monolithic 2X GPU is always delivering more than AFR technique. (or at best case equal)
A monolithic 2X GPU can finish the job for one frame at half the time of a 1X GPU. (if the game isn't CPU limited, or it is badly coded. etc...)
An AFR technique can match a 2X GPU in the best case scenario.

Like I told at 5870 launch, geometry set-up and GS is the reasons 5870 can't deliver in many cases 2X (it depends from the game engine...)

EDIT*
Below is what i told Ryan Smith:

-------------------------------------------------------

Hi Ryan,

Nice new info in your review.

The day you posted your review, i wrote in the forums that according to my perception there are other reasons except bandwidth limitations and driver maturity, that the 850MHz 5870 hasn't doubled its performance in relation with a 850MHz 4890.

Usually when a GPU has 2X the specs of another GPU the performance gain is 2X (of cource i am not talking about games with engines that are CPU limited or engines that seems to scale badly or are poor coded for example)
There are many examples in the past that we had 2X performance gain with 2X the specs. (not in all the games, but in many games)

From the tests that i saw in your review and from my understanding of the AMD slides, i think there are 2 more reasons that 5870 performs like that.

The day of your review i wrote to the forums the additional reasons that i think the 5870 performs like that, but nobody replied me.

I wrote that probably 5870 has:

1.Geometry/vertex performance issues (in the sense that it cannot generate 2X geometry in relation with 4890) (my main assumption)

or/and

2.Geometry/vertex shading performance issues (in the sense that the geometry shader [GS] cannot shade vertex with 2X speed in relation with 4890)(another possible assumption)

I guess there are synthetic benchmarks that have tests like that (pure geometry speed, and pure geometry/vertex shader speed, in addition with the classic pixel shader speed tests) so someone can see if my assumption is true.

If you have the time and you think that this is possible and you feel like it is worth your time, can you check my hypothesis please?

Thanks very much,

MODel3

-------------------------------------------------------

Then i checked the web for specific geometry set-up tests and geometry shading test but i couldn't find anything (all i found was some 3dMark Vantage tests which was not what i was talking about.

-------------------------------------------------------

I checked the web for synthetic geometry tests.
Sadly i only found 3dMark Vantage tests.
You can't tell much from them, but they are indicative.

Check:

http://www.pcper.com/article.p...783&type=expert&pid=12

GPU Cloth: 5870 is only 1,2X faster than 4890. (vertex/geometry shading test)
GPU Particles: 5870 is only 1,2X faster than 4890. (vertex/geometry shading test)

Perlin Noise: 5870 is 2,5X faster than 4890. (Math-heavy Pixel Shader test)
Parallax Occlusion Mapping: 5870 is 2,1X faster than 4890. (Complex Pixel Shader test)

All the above 4 tests are not bandwidth limited at all.
Just for example, if you check:

http://www.pcper.com/article.p...=674&type=expert&pid=8

You will see that a 750MHz 4870 512MB is 20-23% faster than a 625MHz 4850 in all the above 4 tests, so the extra bandwidth (115,2GB/s vs 64GB/s) it doesn't help at all.
But 4850 is extremely bandwidth limited in the color fillrate test (4870 is 60% faster than 4850)

Also it shouldn't be a problem of the dual rasterizer/dual SIMDs engine efficiency since synthetic Pixel Shader tests is fine (more than 2X) while the synthetic geometry shading tests is only 1,2X.

My guess is ATI didn't improve the classic geometry set-up engine and the GS because they want to promote vertex/geometry techniques based on the DX11 tesselator from now on.

 

[v8envy]

There are two other (already explored) possibilities why twice the silicon isn't doing twice as much work per unit time.

1. Power limitations. Whether software or hardware, something is throttling somewhere to avoid blowing out PWM circuits or exceeding the rated load from PCIe + 8 pin + 6 pin connectors. I consider this Unlikely.

2. Silicon or software bugs in scheduling.

I'd be willing to bet a beer the 5890 will be 2x the performance of a 4890 in everything except memory bandwidth-centric situations when it comes out.

 

[cbn]

Originally posted by: v8envy

I'd be willing to bet a beer the 5890 will be 2x the performance of a 4890 in everything except memory bandwidth-centric situations when it comes out.

What situations are memory bandwidth-centric?

Wasn't the idea of memory bandwidth being a bottleneck dismissed?

 

[cbn]

Originally posted by: MODEL3
lol,

what is going here?

AFR technique delivers higher frame rate than a 2X GPU? (everything 2X)
A monolithic 2X GPU is always delivering more than AFR technique. (or at best case equal)
A monolithic 2X GPU can finish the job for one frame at half the time of a 1X GPU. (if the game isn't CPU limited, or it is badly coded. etc...)
An AFR technique can match a 2X GPU in the best case scenario.

Like I told at 5870 launch, geometry set-up and GS is the reasons 5870 can't deliver in many cases 2X (it depends from the game engine...)

EDIT*
Below is what i told Ryan Smith:

-------------------------------------------------------

Hi Ryan,

Nice new info in your review.

The day you posted your review, i wrote in the forums that according to my perception there are other reasons except bandwidth limitations and driver maturity, that the 850MHz 5870 hasn't doubled its performance in relation with a 850MHz 4890.

Usually when a GPU has 2X the specs of another GPU the performance gain is 2X (of cource i am not talking about games with engines that are CPU limited or engines that seems to scale badly or are poor coded for example)
There are many examples in the past that we had 2X performance gain with 2X the specs. (not in all the games, but in many games)

From the tests that i saw in your review and from my understanding of the AMD slides, i think there are 2 more reasons that 5870 performs like that.

The day of your review i wrote to the forums the additional reasons that i think the 5870 performs like that, but nobody replied me.

I wrote that probably 5870 has:

1.Geometry/vertex performance issues (in the sense that it cannot generate 2X geometry in relation with 4890) (my main assumption)

or/and

2.Geometry/vertex shading performance issues (in the sense that the geometry shader [GS] cannot shade vertex with 2X speed in relation with 4890)(another possible assumption)

I guess there are synthetic benchmarks that have tests like that (pure geometry speed, and pure geometry/vertex shader speed, in addition with the classic pixel shader speed tests) so someone can see if my assumption is true.

If you have the time and you think that this is possible and you feel like it is worth your time, can you check my hypothesis please?

Thanks very much,

MODel3

-------------------------------------------------------

Then i checked the web for specific geometry set-up tests and geometry shading test but i couldn't find anything (all i found was some 3dMark Vantage tests which was not what i was talking about.

-------------------------------------------------------

I checked the web for synthetic geometry tests.
Sadly i only found 3dMark Vantage tests.
You can't tell much from them, but they are indicative.

Check:

http://www.pcper.com/article.p...783&type=expert&pid=12

GPU Cloth: 5870 is only 1,2X faster than 4890. (vertex/geometry shading test)
GPU Particles: 5870 is only 1,2X faster than 4890. (vertex/geometry shading test)

Perlin Noise: 5870 is 2,5X faster than 4890. (Math-heavy Pixel Shader test)
Parallax Occlusion Mapping: 5870 is 2,1X faster than 4890. (Complex Pixel Shader test)

All the above 4 tests are not bandwidth limited at all.
Just for example, if you check:

http://www.pcper.com/article.p...=674&type=expert&pid=8

You will see that a 750MHz 4870 512MB is 20-23% faster than a 625MHz 4850 in all the above 4 tests, so the extra bandwidth (115,2GB/s vs 64GB/s) it doesn't help at all.
But 4850 is extremely bandwidth limited in the color fillrate test (4870 is 60% faster than 4850)

Also it shouldn't be a problem of the dual rasterizer/dual SIMDs engine efficiency since synthetic Pixel Shader tests is fine (more than 2X) while the synthetic geometry shading tests is only 1,2X.

My guess is ATI didn't improve the classic geometry set-up engine and the GS because they want to promote vertex/geometry techniques based on the DX11 tesselator from now on.

Do you think optimizing for tesselation could affect classic geometry performance?

 

[MODEL3]

Originally posted by: Just learning

Originally posted by: MODEL3
lol,

what is going here?

AFR technique delivers higher frame rate than a 2X GPU? (everything 2X)
A monolithic 2X GPU is always delivering more than AFR technique. (or at best case equal)
A monolithic 2X GPU can finish the job for one frame at half the time of a 1X GPU. (if the game isn't CPU limited, or it is badly coded. etc...)
An AFR technique can match a 2X GPU in the best case scenario.

Like I told at 5870 launch, geometry set-up and GS is the reasons 5870 can't deliver in many cases 2X (it depends from the game engine...)

EDIT*
Below is what i told Ryan Smith:

-------------------------------------------------------

Hi Ryan,

Nice new info in your review.

The day you posted your review, i wrote in the forums that according to my perception there are other reasons except bandwidth limitations and driver maturity, that the 850MHz 5870 hasn't doubled its performance in relation with a 850MHz 4890.

Usually when a GPU has 2X the specs of another GPU the performance gain is 2X (of cource i am not talking about games with engines that are CPU limited or engines that seems to scale badly or are poor coded for example)
There are many examples in the past that we had 2X performance gain with 2X the specs. (not in all the games, but in many games)

From the tests that i saw in your review and from my understanding of the AMD slides, i think there are 2 more reasons that 5870 performs like that.

The day of your review i wrote to the forums the additional reasons that i think the 5870 performs like that, but nobody replied me.

I wrote that probably 5870 has:

1.Geometry/vertex performance issues (in the sense that it cannot generate 2X geometry in relation with 4890) (my main assumption)

or/and

2.Geometry/vertex shading performance issues (in the sense that the geometry shader [GS] cannot shade vertex with 2X speed in relation with 4890)(another possible assumption)

I guess there are synthetic benchmarks that have tests like that (pure geometry speed, and pure geometry/vertex shader speed, in addition with the classic pixel shader speed tests) so someone can see if my assumption is true.

If you have the time and you think that this is possible and you feel like it is worth your time, can you check my hypothesis please?

Thanks very much,

MODel3

-------------------------------------------------------

Then i checked the web for specific geometry set-up tests and geometry shading test but i couldn't find anything (all i found was some 3dMark Vantage tests which was not what i was talking about.

-------------------------------------------------------

I checked the web for synthetic geometry tests.
Sadly i only found 3dMark Vantage tests.
You can't tell much from them, but they are indicative.

Check:

http://www.pcper.com/article.p...783&type=expert&pid=12

GPU Cloth: 5870 is only 1,2X faster than 4890. (vertex/geometry shading test)
GPU Particles: 5870 is only 1,2X faster than 4890. (vertex/geometry shading test)

Perlin Noise: 5870 is 2,5X faster than 4890. (Math-heavy Pixel Shader test)
Parallax Occlusion Mapping: 5870 is 2,1X faster than 4890. (Complex Pixel Shader test)

All the above 4 tests are not bandwidth limited at all.
Just for example, if you check:

http://www.pcper.com/article.p...=674&type=expert&pid=8

You will see that a 750MHz 4870 512MB is 20-23% faster than a 625MHz 4850 in all the above 4 tests, so the extra bandwidth (115,2GB/s vs 64GB/s) it doesn't help at all.
But 4850 is extremely bandwidth limited in the color fillrate test (4870 is 60% faster than 4850)

Also it shouldn't be a problem of the dual rasterizer/dual SIMDs engine efficiency since synthetic Pixel Shader tests is fine (more than 2X) while the synthetic geometry shading tests is only 1,2X.

My guess is ATI didn't improve the classic geometry set-up engine and the GS because they want to promote vertex/geometry techniques based on the DX11 tesselator from now on.

Do you think optimizing for tesselation could affect classic geometry performance?

The tessellator can take basic geometric shapes and break them up into smaller parts.
It can also take these smaller parts and reshape them to form geometry that is much more complex and that more closely approximates reality.

The usual procedure is this:

The classic geometry set-up engine (geometry & vertex assempler) (this is a fixed function part)
is generating the vertex data (also doing some basic geometry based calculations that the geometry/vertex shader is to slow to do)
and is sending data (patch control points) to the Vertex shader to do some programable geometry based staff.

If the DX11 game engine is taking advantage of the tesselator unit (which is a fixed function unit, geometry shader which is implemented in the programable unit doesn't have the power to do tesselation fast enough)
the vertex shader sends the data (transformed control points) to the Hull shader and the Hull shader decide how much tesselation will be required.
Then the tessellator is taking the outputs from the hull shader and generating the added geometry, then it sends the points generated to the Domain Shader, which can apply operations to them.
And the procedure continious with other stages of the DX11 pipeline like Geometry shader/rasterizer/Pixel shader/Output Merger etc...

You can check youtube for some tesselation based techniques and what the results will look like.

http://www.youtube.com/watch?v=-eTngR6M37Q

 

[bunnyfubbles]

Originally posted by: v8envy
I'd be willing to bet a beer the 5890 will be 2x the performance of a 4890 in everything except memory bandwidth-centric situations when it comes out.

with 7Gbps GDDR5 chips on the way (ie a possible 1750MHz on 256bit bus = 224GB/sec), you may be wrong about the bandwidth if the 5890 comes with those instead of the 5Gbps ones that the 5870 has.