AMD Polaris architecture – GCN 4.0

[Ferrari100]

+1 they should have shown us just how it is, rather than showing us something that has to be clarified and analysed just to get to the result.

The whole idea of it is to get people talking and make people aware of the product. Its actually a very good marketing technique in any business. Especially when you know you have a sure-fire hit product.

 

[D.P.]

That explains a lot, the "51% GPU utilisation" cited at Computex was wrong
It is 83%, 17 percentage points short of full scaling.

So cut the FPS in half 62.6 x0.5 = 31.5 and add the 17% = 36.5 FPS.

Math fail
36.5x1.83 = 66.8FPS which is faster than they measured.

If 1 GPU provide Y FPS and 2GPU provides 1.83xY and you want to know the performance of a single GPU, Y, then simply divide by 1.83

62.5/1.83= 34FPS.

 

[humbug]

That explains a lot, the "51% GPU utilisation" cited at Computex was wrong
It is 83%, 17 percentage points short of full scaling.

So cut the FPS in half 62.6 x0.5 = 31.5 and add the 17% = 36.5 FPS.

i think thats wrong, its 17% of two GPU's so when scaling to one its 17% from each GPU.

2x 83% = 166% - 2x 100% (each GPU full scaling) = well... 200%, difference = 36 percentage points, so lets do it again.

Cut the FPS in half; 62.5 x0.5 = 31.25 and add the 36% = 42.5 FPS.

 

[D.P.]

i think thats wrong, its 17% of two GPU's so when scaling to one its 17% from each GPU.

2x 83% = 166% - 2x 100% (each GPU full scaling) = well... 200%, difference = 36 percentage points, so lets do it again.

Cut the FPS in half; 62.5 x0.5 = 31.25 and add the 36% = 42.5 FPS.

Complete and utter math fail again, but sure, just keep re-iterating rubbish until you get the FPS you want.

 

[bru]

That explains a lot, the "51% GPU utilisation" cited at Computex was wrong

Well if that is the case and what they showed at the announcement is incorrect then somebody needs firing.

You absolutely have to get the fact and figures on your presentation slides correct, even if your trying to be devious by using non linear scales on your graphs or some such, the actual figures you present should always be correct.

Myself I'm not so sure about the whole mess, it just seems like a cluster *&^%$ to me.

 

[humbug]

i think thats wrong, its 17% of two GPU's so when scaling to one its 17% from each GPU.

2x 83% = 166% - 2x 100% (each GPU full scaling) = well... 200%, difference = 36 percentage points, so lets do it again.

Cut the FPS in half; 62.5 x0.5 = 31.25 and add the 36% = 42.5 FPS.

No thats wrong, you can't convert two into one, it was right the first time, so about 36.5 FPS

 

[humbug]

Well if that is the case and what they showed at the announcement is incorrect then somebody needs firing.

You absolutely have to get the fact and figures on your presentation slides correct, even if your trying to be devious by using non linear scales on your graphs or some such, the actual figures you present should always be correct.

Myself I'm not so sure about the whole mess, it just seems like a cluster *&^%$ to me.

Yeah, ^^^^ they complicated it to such an extent they got it wrong themselves.

 

[D.P.]

Well if that is the case and what they showed at the announcement is incorrect then somebody needs firing.

You absolutely have to get the fact and figures on your presentation slides correct, even if your trying to be devious by using non linear scales on your graphs or some such, the actual figures you present should always be correct.

Myself I'm not so sure about the whole mess, it just seems like a cluster *&^%$ to me.

Someone does deserve firing. The AotS benchmark doesn't even tell you diddly squat about GPU utilization. It tells you the percentage time the GoU was the bottleneck vs the CPU. For The results the 480 xfire was GPU bound 51% of the time, but you have no idea what the GPU utilization was in either the 51% GPU bound or the 49% CPU bound parts. In the GOU bound time the GPU utilization could have been much less than 100%, merely getting constrained by something like geometry throughput or ROPs.

 

[CAT-THE-FIFTH]

We already know the basic specs - 2304 shaders at 1266MHZ. That should plonk it close to an R9 390X.

I don't expect any higher.

 

[humbug]

What they need to do is do an official run with one card, no enigma silliness and publish that.

 

[D.P.]

No thats wrong, you can't convert two into one, it was right the first time, so about 36.5 FPS

Stop embarrassing yourself with elementary school math errors.

 

[CAT-THE-FIFTH]

We will find out in a few weeks,and I expect a week to launch we will get the press deck leaked or something similar.

 

[humbug]

Stop embarrassing yourself with elementary school math errors.

Stop jumping about on you're high-chair food tray, you not what you think you are.

 

[bru]

We will find out in a few weeks,and I expect a week to launch we will get the press deck leaked or something similar.

Shush you, you'll start the whole paper launch debate again.

 

[CAT-THE-FIFTH]

Shush you, you'll start the whole paper launch debate again.

AMD and Nvidia seem to like doing that nowadays!!

 

[Gregster]

Why don't you just run the AoTS bench on your 970 Humbug? You did join in on the first bench thread, so I assume you have it. It would be a better comparison also.

 

[Lokken86]

Wish I had the game to try.

 

[CAT-THE-FIFTH]

Wish I had the game to try.

Wish I had a GPU good enough to try!

 

[Lokken86]

Wish I had a GPU good enough to try!

You'll have a 480 soon enough

 

[CAT-THE-FIFTH]

You'll have a 480 soon enough

I might actually wait and see what the GTX1060/GTX1060TI brings to the table first. But I do feel tempted.

One of my mates has an ancient HD7870XT and he probably is going to buy one at launch as long as performance is around what we expect.