backtop


Print 42 comment(s) - last by Jkm3141.. on Aug 7 at 6:17 PM

Two chipsets, two graphics cores

DailyTech has received a July 2006 Intel specification update on G965 and Q965 Express chipsets that reveals final details of Intel’s GMA X3000 and GMA 3000 integrated graphics cores. Intel originally intended for the G965 and Q965 Express chipsets to have the same graphics core but have since changed its mind. The result is two separate graphics core—GMA X3000 for G965 Express and GMA 3000 for Q965 Express. As Intel is appealing to the mainstream consumer with G965 Express the GMA X3000 graphics core will have greater graphics processing capabilities.

G965 Express with its GMA X3000 graphics core will support DirectX 9c, DirectX 10 and OpenGL 1.5 features. The supported features include:
  • Hardware vertex shader model 3.0
  • Hardware pixel shader model 3.0
  • 32-bit and 16-bit full precision floating point operations
  • Up to 8 multiple render targets
  • Occlusion Query
  • 128-bit floating point texture formats
  • Bilinear, trilinear and anisotropic mipmap filtering
  • Shadow maps and double sided stencils
There’s no mention of the amount of vertex or pixel shaders available on GMA X3000 graphics cores. However, the shaders are fully programmable which can be adapted to varying amounts of vertex or pixel shaders. As previously reported the GMA X3000 will have a 667 MHz graphics core clock with support for high dynamic range and Intel Clear Video Technology for enhanced video playback.

Q965 Express chipsets will receive a less powerful graphics core in the form of the GMA 3000. GMA 3000 will meet the minimums to support Microsoft’s upcoming Windows Vista Premium with Aero Glass interface with support for DirectX 9c and OpenGL 1.4 plus. The following features are supported:
  • Software vertex shader model 2.0/3.0
  • Hardware pixel shader model 2.0
  • 32-bit and 16-bit fixed point operations
  • Up to 8 multiple render targets
  • Occlusion query
  • 128-bit floating point texture formats
  • Bilinear, trilinear and anisotropic mipmap filtering
  • Shadow maps and double sided stencils
With the exception of hardware pixel shader model 2.0 support the GMA 3000 graphics core has similar specifications as the outgoing GMA 950 graphics core. Since Intel is catering Q965 Express chipsets towards business users as part of its vPro platform initiative it doesn’t exactly need gaming capable graphics power.

Availability of G965 Express based products is expected sometime this month while Q965 Express based products will have a formal launch in early September.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: Comparison
By ltcommanderdata on 8/5/2006 10:42:54 PM , Rating: 3
I'm actually thinking that with proper driver support the GMA X3000 could easily be the fastest IGP. Based on the diagrams Intel has provided, indications point to the GMA X3000 having 8 unified shaders. This makes sense intuitively too since the extra room from the 130nm to 90nm transition could be used to double shaders from the 4PS in the GMA950. Now both ATI's X700 based IGP and nVidia's 7300 based IGP look to have 4PS + 2VS. In contrast, the GMA X3000 has 8 unified shaders. Also, it's clocked at 667MHz which is very aggressive although I haven't heard how high ATI and nVidia are clocking their solution. In any case, from a hardware perspective, Intel's solution is definitely superior.

All it needs is Intel to take full advantage of it with drivers. If done properly, the GMA X3000 should be able to compete with the X1300 HM. The X1300HM is only clocked at 400MHz, has 4PS, 2VS, and has 6.4GB/s of memory bandwidth. In comparison the GMA X3000 is clocked at 667MHz, has 8 unified shaders, and has 10.7GB/s of bandwidth with dual channel DDR2 667 memory. Granted that it's bandwidth is shared, but it has a lot more available to begin with than the X1300HM. The GMA X3000 can also take advantage of Fast Memory Access in the G965 and it's shaders are also multithreaded to keep processing in the event that a thread stalls (probably a development of Hyperthreading). These features should help offset the disadvantages of shared memory. I can also see the GMA X1300 taking on the vanilla X1300 which is only slocked slightly higher than the X1300 HM at 450MHz and has 8GB/s of memory bandwidth. Again it all depends on whether the drivers realize the potential of the GMA X3000, but the potential is certainly there.


RE: Comparison
By coldpower27 on 8/5/2006 11:43:31 PM , Rating: 3
Yes just as how from a hardware perspective, the GMA950 is more powerful then the Xpress 200 IGP. With the 4PS vs 2PS argument there.

It doesn't matter from a theoretical standpoint which one is quicker, what matters is how the solution performs in reality not in theory. Judging from Intel's past attempts at IGP, performance has been anemic at best in comparison to Nvidia/ATI based solutions.

I certainly hoping this next attempt is improved. There are two big IFs here IF Intel can gets it driver team for it's IGP together and IF they have increase their IPC on their pipelines to the levels of the Nvidia/ATI counterparts, then Intel has a fighting chance. If not it's doubtful performance will be very high.

Well it depends what you will be comparing these solutions to, the 7300 LE and X1300 HM are adequate comparisons, from a theoretical standpoint the X3000 should be quicker but I expect the X1300 HM and 7300 LE to win out in the end.

7300 LE has a a tad more memory bnadwidth, then the X1300 HM, with 667MHZ vs the 500MHZ utilized on the X1300 HM, though both have 64Bit Memory Interfaces.

You specs aren't entirely correct.

X1300 HM
450MHZ Core
500MHZ Memory on 64Bit = 4.0GB/s of Isolated Memory Bandwidth

And the 7300 LE has access to 5.3GB/s of Isolated Memory Bandwidth.

Also you got to keep in mind the G965 Chipset will have access to Dual Channel DDR2-800 for 12.8GB/s of Shared Bandwidth between the processor and other peripherials.

Both the 7300 LE and X1300 HM, have their own isolated Memory buffers of 128MB capacity, in addtion to being able to utilize some system memory, while the X3000 has to be completely reliant on system memory alone. Comparing shared bandwidith vs isolated bandwidth is hardly apples to apples doesn't matter if there is alot more of it, if most of it is meant for the processor itself.

Your doing alot of speculating yourself, it is very unlikely for IGP solutions to reach performance levels of even the current generation lowest end discrete cards, if they do that would be quite a monummental feat in itself. I am expecting Geforce 6200 TC-16 level performance as a starting point, any higher is basically icing on the cake. Look at the Xpress 200 vs the X300 SE, which is basically the lowest end possible R3xx derived card to my knowledge, the X300 SE is undoubtablely superior.

The X3000 certainly has alot of potential, but whether on not Intel executes on this particular IGP solution is another story, as well, we don't know that actual pipeline configuration of the X3000 8 Shaders Units there isn't enough information without the TMU's as well.










RE: Comparison
By ltcommanderdata on 8/6/2006 3:22:25 AM , Rating: 3
quote:
Yes just as how from a hardware perspective, the GMA950 is more powerful then the Xpress 200 IGP. With the 4PS vs 2PS argument there.

This circumstance was slightly different though since the GMA950 didn't have any hardware VS or T&L which severely limited it's performance (to say the least) even if it had 4PS to the Xpress 200's 2PS + 1VS.

http://www.digit-life.com/articles2/mainboard/ati-...

The GMA950 actually does surprisingly well in Doom 3 and UT2004 although the other games are disappointing compared to the Xpress 200. I guess it's a matter of game dependance on VS, which the GMA950 lacks, and on driver support.

quote:
You specs aren't entirely correct.

X1300 HM
450MHZ Core
500MHZ Memory on 64Bit = 4.0GB/s of Isolated Memory Bandwidth

Your right, my specs were a bit off. I quoted a 400MHz core for the X1300HM while it is 450MHz, however, the standard RAM speed is 400MHz x2 on a 64-bit interface for 6.4GB/s of bandwidth.

http://www.beyond3d.com/reviews/powercolor/x1300/i...

Various manufacturers change their RAM speeds though so your numbers might be right for the cards you were looking at. nVidia gives their manufacturers even more flexibility and I've even seem models with 32-bit memory interfaces. Generally the 7300LEs are all over the place so it's hard to determine.

Anyways, in terms of architecture the GMA X3000 is designed around a shared memory environment so it isn't affected as much. The GMA X3000 uses PowerVR SGX technology and tiling which isn't as dependant on memory bandwidth and probably includes larger internal buffers to offset that. In constrast, the RV515 and the G72 are designed for larger amounts of memory bandwidth so they are unlikely to have the larger buffers to compensate when that bandwidth is taken away. Interestingly the PowerVR SGX architecture supports usage of HDR and antialiasing at the same time like ATI, but it'd probably be too slow to use.

In any case, you're right that we'll have to wait for the actual products and finalized drivers. Still I am hopeful since Intel looks to be taking graphics seriously for once. This is the product/architecture that AMD and ATI went to all the trouble to merge over afterall.

The GMA X3000 White Paper is here if anyone wants it:
http://download.intel.com/design/chipsets/applnots...

(On the TMU note, I'm going to bet on 4 of them. I can't believe that Intel would include less than the GMA 950. 8 unified shaders and 4 TMUs seems about right.


RE: Comparison
By IntelUser2000 on 8/6/2006 6:59:44 PM , Rating: 2
quote:
This circumstance was slightly different though since the GMA950 didn't have any hardware VS or T&L which severely limited it's performance (to say the least) even if it had 4PS to the Xpress 200's 2PS + 1VS.


quote:
The GMA X3000 uses PowerVR SGX technology and tiling which isn't as dependant on memory bandwidth and probably includes larger internal buffers to offset that.


Although the tiling based seems similar to PowerVR based architectures, I doubt its based on the latest SGX. The SGX derived ones are supposed to be on Intel's Stanwood, which is successor to 2700G multimedia accelerator used in PDA's. Stanwood is now cancelled however. It was also rumored that next generation graphics core from Intel for PDAs feature shader technology(sounds like SGX). PowerVR mentions

"Maximum effective pixel fillrate performance from 200Mpix/sec to 1200Mpix/sec @ 200MHz with even higher Z and stencil fill rate and polygon throughput from 2Mpoly/sec to 13.5Mpoly/sec @ 200MHz. Performance depends on core and configuration selected."

I highly doubt its the same thing.


RE: Comparison
By ltcommanderdata on 8/6/2006 9:11:53 PM , Rating: 2
quote:
Although the tiling based seems similar to PowerVR based architectures, I doubt its based on the latest SGX. The SGX derived ones are supposed to be on Intel's Stanwood, which is successor to 2700G multimedia accelerator used in PDA's. Stanwood is now cancelled however. It was also rumored that next generation graphics core from Intel for PDAs feature shader technology(sounds like SGX).

It makes sense that the GMA X3000 uses PowerVR SGX technology since Intel already has a license for it. The GMA 950 and before all used Zone Rendering which is an implementation of PowerVR's earlier tiling architectures. PowerVR SGX specifically adds unified shader support and this is exactly what the GMA X3000 adds over the GMA 950 so it can't just be a coincidence.

The Inquirer also reports that the G965 uses PowerVR SGX technology:
http://www.theinquirer.net/default.aspx?article=33...

Obviously it's an "Intelized" implementation, but it makes sense that PowerVR SGX is used since I can't see Intel spending the resources to develop a completely need architecture from the ground up when they already have a license to a perfectly good one.


RE: Comparison
By IntelUser2000 on 8/6/2006 7:03:59 PM , Rating: 2
quote:
This circumstance was slightly different though since the GMA950 didn't have any hardware VS or T&L which severely limited it's performance (to say the least) even if it had 4PS to the Xpress 200's 2PS + 1VS.


Forgot to reply to the above.

According to Sudhian, Geforce 2 MX/Radeon 64 T&L cores are actually slightly less powerful than Athlon XP 2500+ CPU. In order for GMA X3000's hardware T&L to be substantially better than GMA950's software T&L, it will have to beat Pentium D 960 running at 3.6GHz. The counter-argument may be that the CPU is also doing other tasks and the dedicated hardware T&L in X3000 is faster, but I doubt it'll be better on the GMA X3000 if the hardware T&L fails to outperform the Pentium D 960.


RE: Comparison
By coldpower27 on 8/7/2006 12:40:20 AM , Rating: 2
quote:
This circumstance was slightly different though since the GMA950 didn't have any hardware VS or T&L which severely limited it's performance (to say the least) even if it had 4PS to the Xpress 200's 2PS + 1VS.
quote:


http://techreport.com/reviews/2004q4/radeon-xpress...

To be fair the Xpress 200 IGP only seems to offload transform and lighting capabilities, the rest of the vertex shader functions are handled by the system processor. So it's not a Full 1 VS Unit.

I would hardly say that the 7300 GS or X1300 Pro are designed for large amounts of bandwidith. Certainly not G72 as it's highest part is only ~ 6.4GB/s or so. Maybe RV515 a tad, but these are low end cards, they are designed to do as well as possible without bandwidth. I dunno I was going by the 7300 LE's on Newegg, and it seems they are still on 64Bit Interfaces with either 533MHZ/667MHZ which isn't too much of a drop from the 7300 GS official 810MHZ.

Regarding ATI/AMD it will remain to be seen, if they themselves can give us their unified platform so to speak in 2008, I welcome their attempts on the issue.

Like I said, I am not giving Intel the benefit of the doubt, given how poorly their past 2 attempts have performed.





"I mean, if you wanna break down someone's door, why don't you start with AT&T, for God sakes? They make your amazing phone unusable as a phone!" -- Jon Stewart on Apple and the iPhone














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki