Intel's latest graphic chipset specification update

D

DavidC

Regular
Hi guys. I just registered at Beyond3D. The topics at B3D seems pretty interesting.

Anyway onto the topic: http://download.intel.com/design/chipsets/specupdt/31305403.pdf

If you check the links above, the page mentions one sad truth about the new integrated chipsets from Intel:

-Core frequency of 667MHz
-1.33Gp/s fill rate

It does match the news from the Inquirer: http://www.theinquirer.net/default.aspx?article=31239

all that you will get is a two-pipeline GPU equipped with four to six Shader units.

We might expect good performance in the unified shader part, but for fixed function fill rate, its actually lower than the GMA950, and on par with GMA900. That might explain the poor performance in the benchmarks.

Hopefully, the other architectural features like hardware geometry, increased DVMT size, and good final drivers can overcome the lower fill rate(and exceed it).

Here's the most recent benchmark(14.24 drivers): http://www.pconline.com.cn/diy/evalue/evalue/main/0608/856535_4.html
946GZ(unknown drivers): http://www.pconline.com.cn/diy/evalue/evalue/main/0608/851863_5.html
 
  • Like
Reactions: Geo
If intel is very exact about the GMA X3000's Specs, that could also mean the chip only has two ROPs. Given the limitations in integrated solutions there'd never be enough bandwidth to saturate four or more ROPs with current DDR2-RAM's frequencies. Especially if you also add texture fetch bw and the cpu competing with the chipset for bw.
 
this is as much fillrate as a voodoo5 6000. slightly higher than a radeon 9600, much higher than geforce 6150. I wouldn't have been surprised if the chip had only one ROP and one TMU.
don't forget : the GPU runs at a pretty high frequency, it's intended for playing 1024x768 and lower, and fillrate is going to be limited by memory bandwith anyway (even lower than that of DDR2, as it's limited by the FSB, which will still be at 800 on the low end. on the AMD side it's no better, it can't be higher than HTT's bandwith, right?)

I think the GPU is quite poor because of its shader units. they have to do everything (geometry, pixel shading) including some stuff previously done the fixed-function way (triangle setup) (well, according to a thread I read there), and they are probably less efficient than finely tuned separate vertex/pixel shaer pipelines, so I guess there's not enough of them.

in my opinion they should put more ALU (let's say 12 or 16), and 64MB of 64bits GDDR3 on the motherboard for the framebuffer and cache (afterall, there was a time when most branded PCs had their integrated 1MB VRAM on the mobo)
 
this is as much fillrate as a voodoo5 6000. slightly higher than a radeon 9600, much higher than geforce 6150. I wouldn't have been surprised if the chip had only one ROP and one TMU.
Click to expand...

IHVs way too often still use Mega- (or Giga-) Pixels instead of Mega-Texels when quoting multitexturing fillrates. It could easily also be two TMUs with only one ROP, whereby it would be interesting to know what each ROP (irrelevant of the amount) is exactly capable of.

in my opinion they should put more ALU (let's say 12 or 16), and 64MB of 64bits GDDR3 on the motherboard for the framebuffer and cache (afterall, there was a time when most branded PCs had their integrated 1MB VRAM on the mobo)
Click to expand...

It would be also interesting to see what the transistor budget of that IGP is, considering its claimed D3D10 compliance. More would be obviously better, but IGPs should have a second acronym: DCGP as in dirt cheap graphics processors ;)
 
Last edited by a moderator:
It would be also interesting to see what the transistor budget of that IGP is, considering its claimed D3D10 compliance. More would be obviously better, but IGPs should have a second acronym: DCGP as in dirt cheap graphics processors
Click to expand...

I think one of the publications by Japanese Site PC Watch said 3x the transistor count over the previous generation in the 3D core. It was translated so it may be wrong, and we don't know the transistor count of the previous gen core.
 
Blazkowicz said:
and fillrate is going to be limited by memory bandwith anyway (even lower than that of DDR2, as it's limited by the FSB, which will still be at 800 on the low end.
Click to expand...

I'm quoting myself as I think I was wrong about that, so forget it :)

Ailuros said:
It would be also interesting to see what the transistor budget of that IGP is, considering its claimed D3D10 compliance. More would be obviously better, but IGPs should have a second acronym: DCGP as in dirt cheap graphics processors ;)
Click to expand...

sure, but I tend to see two kinds of IGP : the crappy ones (integrated VIA, SiS, Intel "eXXtreme graphics 2"), and the better ones, which are actually intended for some gaming (nforce IGP and radeon 9100 IGP in their respective times, geforce 6100 and 6150).
that G965 is thus quite a disappointment :)
 
I haven't checked performances yet of GMA X3000 but I talked extensively about it with its main architect. Intel is very smart when it comes to efficiently implement math units, even if ATI/NV don't like to hear that. Of course great math units is not enough to make a great graphic core.

It has full DX10 support but Intel will probably not expose it because they probably won't have time to QA it before the next chipset is out with a tweaked architecture. Supporting DX10 on this chipset is not a priority so they probably won't unless the next one is late and main customers ask for DX10 support.
 
You must log in or register to reply here.
Back
Top