gpu top list updated

on notebookcheck, the leader has been the hd 5870m crossfire. now they have updated the list and the no.1 spot now goes to the gtx480m sli.

Notebookcheck: Mobile Graphics Cards - Benchmark List

thoughts?

 

Sounds right to me. I don't see that there's anything to talk about here.

 

The point being?

 

it's more the fact is it that faster?, they both have there strong points and there weakness's. what exactly makes the sli gtx480m superior to the crossfire hd 5870's? i'm no expert and would like to know what it is exactly that makes the notebook gpu's faster.

 

GTX 480M SLI would be faster overall; I don't think there would be many areas where it would be beaten by the 5870 CF, if any at all, so Notebookcheck's listings are fine.

The list has nothing to do with the 480M's main weaknesses, like the price (a single 480M costs about as much as two 5870s, as far as I've seen, and the CrossFire configuration would definitely be faster than a single 480M), and the relatively high level of heat output/power consumption. That list is based solely on performance, and in terms of performance it makes perfect sense.

 

i see that the sli480 has roughly half the pixel shaders, but twice the memory bus speed, does this make a massive difference in gaming?

 

Nvidia Shaders != ATI Shaders. The amount of shaders Nvidia use in their 480M is monstrous for mobile GPU standards.

In terms of performance, 480M is definitely going to be faster, and as said, that list is solely based on performance, it doesn't care whether your GPU can make coffee or whatsoever.

 

What AMD calls a shader is very different to what Nvidia calls a shader, and in just looking at the memory bus widths you're ignoring that memory bandwidth is determined by memory clock speeds (and type) as well as bus width - the GTX 480M has its GDDR5 clocked at 600MHz (the 1200 and 2400 figures are not the actual clock speeds), which is only a slight advantage in memory bandwidth (20%) vs the 5870's 1000MHz + 128-bit bus. Besides that, the two manufacturers differ in the way they use that bandwidth, so that doesn't tell you much either.

On the whole, comparing the raw memory and shader figures of these two cards is pointless since their architectures are completely different. However, all the benchmarks I've seen so far for Fermi suggest the 480M will have the advantage in performance over the 5870; not much in DX10, but with a solid lead in DX11.

 

cheers for clearing the shader situation up. might have to google there architectures and have a read up. along with how each of them use there memory..


cheers

 

Sadly, they don't give away much. I remember reading that ATI compresses textures better, which leads to less heavy VRAM usage, but that was some time ago and I'm not sure if it still holds true.

Read the earlier articles on each architecture for an overall picture - probably reviews of the desktop 5870 and 480, from somewhere like AnandTech.

 

To get a rough idea of the difference between Nvidia and AMD shaders, most people divide AMD shaders by 5. So having half the 'shaders' of an AMD, like lackofcheese said, is gargantuan.

 

Yes, massive number of shaders + ridiculously low clock speeds seems to be working for NVIDIA.

 

In the chart below, ATI calls each w,y,z,w,t portion a separate shader (5 total.) Nvidia counts their own as 1.




Besides the chart, it's even more difficult to compare fairly as Nvidia shaders usually run faster (2-2.5X the GPU clock.)

 

i still think its wrong to make GTX480M SLI top.. for christ's sake , its not even out or in a notebook yet unlike 5870m CF...

 

Yeah, it's complicated. In the case of the 480M the shaders run at 850MHz vs 700MHz in the Mobility Radeon 5870, though.

As your diagram shows, the difference in performance will vary depending on the type of work the GPU is doing. AMD's GPUs have significantly more raw computational power, but their architecture also makes it more difficult to actually use this power - consider the 1.12 TFLOPS vs 598.4 TFLOPS in the 5870 vs the 480M, and yet it seems that the 480M will be the faster card.

On the whole, based on the Fermi benchmarks I've seen, it would seem that what Nvidia calls a shader is worth something like 2.3 times what AMD calls a shader when it comes to DirectX 10 gaming performance. DirectX 11 performance is different, because tessellation is handled by AMD's dedicated tessellators, so it doesn't really reflect on shader power. On the other hand, in terms of raw FLOPS, each AMD shader is actually equal to an Nvidia shader, which is why AMD's figures for FLOPS are always much higher. However, as the above figure suggests, only something like half of this raw power is brought to bear, on average.

Notebookcheck isn't about being 100% accurate, and never has been. Putting up cards that haven't been properly released yet based on unknown benchmark sources, or even just guesses, is typical of them. If you don't like that approach, find another site, or make your own.

 

Is there even a production machine that you can get the 480M-SLI configuration?

 

There's also the fact that most comparisons put the 480 GTX slightly ahead of the 5870, but we're dealing with the mobile versions here. I think NVidia's may be less ham-stringed--it is a very underclocked GT100 chip with a lot of shader processors disabled, whereas AMD's mobile 5870 is based on the desktop 5770.

Anyone more in the know please correct me...

 

Yeah, the 480M is still a GF100, with 352 shaders compared to 480 in the GTX 480, but the shader clocks are also 850MHz vs 1401MHz in the desktop card - i.e. ~73.3% of the shaders with ~60.7% of the clock speed, and hence ~44.5% of the overall raw power.

With the 5870, it's half the shaders, but the clock speed is 82.4% that of the desktop card, and so that's 41.2% of the overall raw power.

As you can see, while Nvidia's laptop card is slightly less hamstrung, the difference isn't so large. Some might say that shader count matters more than clock speeds, but that seems baseless to me - half the shaders at twice the clock speed are capable of getting the same amount of work done.