Clarkdale IGP benchmark

Oh wow, that is a nice improvement. Still the kicker will still be whether or not the drivers are upto scratch, the X4500 scored well in a few games but there's plenty titles where its performance is simply atrocious or the game flat out won't play without major issues. At least it looks like they've finally nailed the hardware, I never thought I'd see the day when Inte;'s integrated parts comfortably outperformed the low end discrete offering from Nvidia and ATI.

With Fusion hopefully hitting this year, it looks like the low end of the PC market is set to see a significant boost in graphics performance. Don't know where this leaves the low end discrete market.
 
Last edited by a moderator:
If you squint a bit, you can read the minimum framerate for the WIC benchmark -- the Clarkdale picked up that win by having far better minimum rates than the other two. It also outperformed the others in synthetic benches (3DMark Vantage and RTHDRIBL) and RE:5. I'm not sure what that last game was (Eagle Hit the Sky?) and 3DMark'06 was basically a tie with the 4350 - but still above the G210.

Overall, they've got a solid base platform; here's hoping they don't castrate it with terrible drivers.
 
Shame we won't be seeing this IGP in netbooks anytime soon, as so long as the hardware video decoding is upto scratch, it'd potentially offer the same sort of package as Nvidia's ION did, which was a great value add over the GMA 950 based netbooks if you ask me.
 
Indeed; it will be curious to see power consumption figures in the production hardware. Or more specifically, perf / mw when up against the 4350 and G210.
 
The game named "Eagle Hit the Sky" is actually HAWX in more commonly understood title. :)
 
Last edited by a moderator:
that's with a crippled version to boot :oops:

pentium has a 533MHz GPU and PC8500 memory, core i3 will sport a 733MHz GPU and PC10600 memory.

it's still DX10, so maybe it's merely a GMA X4500 with more units.
that's clever from Intel, I never doubted they could scale the design, so here is it. not only cutting grass under ION and 785G, but even making the cheap OEM cards on their platform moot.

it's both good news and bad news. bad news is it's a huge dent in the competition, good news is this will help PC gaming.
also AMD Fusion will destroy it with but by not an actual huge margin.
 
that's with a crippled version to boot :oops:

pentium has a 533MHz GPU and PC8500 memory, core i3 will sport a 733MHz GPU and PC10600 memory.

it's still DX10, so maybe it's merely a GMA X4500 with more units.
that's clever from Intel, I never doubted they could scale the design, so here is it. not only cutting grass under ION and 785G, but even making the cheap OEM cards on their platform moot.

it's both good news and bad news. bad news is it's a huge dent in the competition, good news is this will help PC gaming.
also AMD Fusion will destroy it with but by not an actual huge margin.

That I didn't know, so this level of performance is probably what we can expect from the mobile version? That's pretty nice gaming performance for a small and low power notebook with an integrated GPU in that case.

What's the state of X4500 drivers these days?
 
*snip*

With Fusion hopefully hitting this year,

*snip*

Hey, is it 2010 already? :D



According to the Clarkdale IGP diagram from a presentation at recent IDF, it said "3x Mathbox". I'm taking that as Clarkdale having 3x more Mathbox units/performance compared to the G45 counterpart.

Is the Mathbox really that big of a bottleneck for G45?

pentium has a 533MHz GPU and PC8500 memory, core i3 will sport a 733MHz GPU and PC10600 memory.

Remember, they are using PC-12800 memory. The CPU is even overclocked. Heres to HOPING that the GPU isn't.

This gives hope Sandy Bridge's IGP will have fighting chance against Fusion.
 
I didn't pay attention to that.
thus if run at the actual memory spec given by Intel performance will be lower (dunno what's the QPI link frequency either, here it's at 2800MHz - how much bandwith is that?)

ignoring the recommended memory frequency set by Intel is a sane option but the actual consumers will let it as it is.
future low end cards will use ddr3 instead of ddr2 : both facts help mitigate results from that review.
 
I think Clarkdale uses a 6GT/s QPI, which makes it a 3GHz connection. That's 24GB/s of bandwidth. If its overclocked that will be a 36GB/s, 9GT/s. But that's just the connection to the CPU.

thus if run at the actual memory spec given by Intel performance will be lower (dunno what's the QPI link frequency either, here it's at 2800MHz - how much bandwith is that?)

That shouldn't be an issue as the regular Core i3/i5's support DDR3-1333 or even DDR3-1600(some infos I have seen say it supports XMP).

I'd like to see how the 900MHz version in the Core i5 661 performs.
 
It's intersting that Apple doesn't want the intel graphics parts in their products this generation. It might tell us something about the performance of amd and nvidia's new low end since those benches definitely fare well against the 9400 that MacBooks currently pack...
 
that's with a crippled version to boot :oops:

pentium has a 533MHz GPU and PC8500 memory, core i3 will sport a 733MHz GPU and PC10600 memory.

Don't forget there's even a much faster version at 900Mhz.

it's still DX10, so maybe it's merely a GMA X4500 with more units.

Anandtech explained nicely what changed in the graphics core of Clarkdale:

The smaller transistors enable much higher performance. While G45 had 10 shader cores, the 'dale GPU increases that to 12. A number of performance limiting issues have now been resolved, so we should see much more competitive performance from Intel's graphics.
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3648&p=6
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3648&p=6
 
it's both good news and bad news. bad news is it's a huge dent in the competition, good news is this will help PC gaming.
also AMD Fusion will destroy it with but by not an actual huge margin.

Isn't increased competition a good thing? This just means Intel's IGPs can't be ignored any more and the other guys need to step up. It's going to be very hard to justify any sort of entry level card with this kind of performance from an on-die GPU though.
 
From the article:

As the GeForce 210, Radeon HD 4350 and the G6950 integrated GPU are all entry-level product, so we chose the Entry (entry level) test model of the processor performance in order to prevent a bottleneck I will be super-CPU FSB to 200MHz, frequency to 4.0GHz, the memory as DDR3 1600MHz, to maintain consistency of the platform to be tested.

Unfortunately that means this preview is not an apples to apples comparison and the logic that the CPU could be a potential bottleneck for such enty level cards seems slightly flawed.

As mentioned previously in this thread by overclocking the system bus and CPU these numbers are not going to be indicitive of how good Clarkdale IGP is on its own compared to the competition. More of a platform preview than an IGP vs enty level discrete comparison.
 
We recently saw intel offloading gpu work to the cpu in Vantage and others http://techreport.com/articles.x/17732 , so would have prefered these benchmarks on a less overclocked, more entry level cpu...

You can see the Vantage result without the offloading in their other review.

http://translate.googleusercontent....gle.ca&usg=ALkJrhgQjPbNyGsm5DvH1jW4zZNfxYdDeA

GPU score: 6192 vs 5213

Still better than 210's 4265 and 4350's 4088

Yes, I do wish they did a more fair review though. :p What's the logic in thinking IGP will be bottlenecked by the CPU? :rolleyes:
 
According to the Clarkdale IGP diagram from a presentation at recent IDF, it said "3x Mathbox". I'm taking that as Clarkdale having 3x more Mathbox units/performance compared to the G45 counterpart.

Is the Mathbox really that big of a bottleneck for G45?
A good question, but it seems plausible. Both nvidia and AMD can execute those functions (rcp, transcendentals etc) at a rate of ~1/5 that of normal instructions.
One extended math unit however is seriously slow, operates on one scalar at a time, and instruction throughput is 1/1 (rcp) to 1/12 (worst case sin/cos) per clock. G45 has 10 normal execution units (which are afaik 4-wide physically), so you can see that to get a similar ratio for these instructions as AMD and nvidia do you'd need a LOT of these math units. I don't know how many mathbox units there are in G45 but it is not even possible to be anywhere close that. Or maybe I do, the manual mentions execution units are further divided into rows, each row gets its own l1 cache and MathBox, but it doesn't mention how many rows there are - I guess since it's 10 execution units that must be either 2 or 5 rows hence MathBox units - if intel tripled MathBox units that'll probably mean that there were only 2 rows, and 6 now with Ironlake, and execution units would therefore needed to be increased to at least 12.
Even with 6 MathBox units for 12 (4-wide) execution units, rcp throughput would only be 1/8 that of mac, and other MathBox instructions even less.
I guess someone should just measure instruction throughput on a g45 to see if it's really that low for these instructions :).

In any case that's certainly not the only change, for instance it also got a hw clip unit.
 
Last edited by a moderator:
Back
Top