Intel 32nm Clarkdale CPU-and-GPU chip benchmarked

[qurious63ss]

Very limited numbers released but hey it's a start.


http://www.hexus.net/content/item.php?item=20291

[kl0012]

Very limited numbers released but hey it's a start.


http://www.hexus.net/content/item.php?item=20291

Thanks for the link. Clarkdale's IGP is doing surprisingly well.

[haylui]

performance is close to 785G IGP....

[qurious63ss]

Yes very well indeed. I was under the impression that the Clarkdale IGP was nothing more than a gma 4500 integrated into the same package, but it looks like it may be something completely new or could it be that moving IGP closer to CPU make that big a difference?

[Vapor]

Yes very well indeed. I was under the impression that the Clarkdale IGP was nothing more than a gma 4500 integrated into the same package, but it looks like it may be something completely new or could it be that moving IGP closer to CPU make that big a difference?

I think there are three fundamental improvements at play here:

1) Slightly upgraded IGP arch (it's about due).
2) Increased IGP clockspeeds (I think there's a process shrink and there's definitely room for an increased thermal envelope from moving it into the CPU package [that has a much better guaranteed cooler] and out of the NB [that has a dinky passive thing plus all the other stuff on the same die])
3) Increased memory bandwidth due to improved memory controller on the Clarkdale (compared to G43/45's MC).

I could be totally wrong on all three points...I have no actual knowledge of the innerworkings of Clarkdale, if I did, I couldn't talk about it

[qurious63ss]

I think there are three fundamental improvements at play here:

1) Slightly upgraded IGP arch (it's about due).
2) Increased IGP clockspeeds (I think there's a process shrink and there's definitely room for an increased thermal envelope from moving it into the CPU package [that has a much better guaranteed cooler] and out of the NB [that has a dinky passive thing plus all the other stuff on the same die])
3) Increased memory bandwidth due to improved memory controller on the Clarkdale (compared to G43/45's MC).

I could be totally wrong on all three points...I have no actual knowledge of the innerworkings of Clarkdale, if I did, I couldn't talk about it

All three IPC, freq, and memory bandwidth would definately affect performance, question now is which one helped the most or was it a little of each. I agree with #1 but not sure if it's just a "slight" improvement, I guess that depends on how big a role #2 and #3 had on the improvements.

[BulldogPO]

Indeed, looks pretty promising for Intel IGP.

[gallag]

This thing looks awesome, Big surprise on the igp performance and power consumpsion, Dual core smacking the quads around lol

[IntelUser]

I think there are three fundamental improvements at play here:

1) Slightly upgraded IGP arch (it's about due).
2) Increased IGP clockspeeds (I think there's a process shrink and there's definitely room for an increased thermal envelope from moving it into the CPU package [that has a much better guaranteed cooler] and out of the NB [that has a dinky passive thing plus all the other stuff on the same die])
3) Increased memory bandwidth due to improved memory controller on the Clarkdale (compared to G43/45's MC).

I could be totally wrong on all three points...I have no actual knowledge of the innerworkings of Clarkdale, if I did, I couldn't talk about it

The current G45 works at 800MHz while Clarkdale's IGP works at 700MHz so we can scratch #2.

[Vapor]

Sweet...considering the large improvement in performance, that puts more weight into the other categories (i.e., my first and third suggestions + everything else actually at play).

Hopefully drivers are getting improved as well.

[FischOderAal]

Increadible performance of an Intel IGP. I'm really surprised! Now I hope the mobile parts won't get a scrapped version in order to keep down the power :/

[Telperion]

I'm going with #3. I can't recall the Dr Who post, but he was singing the praises of how we'd all be surprised at how much the IGP would benefit from memory bandwidth and being on-die. Made sense at the time, and looks like these benches are backing it up.

[ajaidev]

Wow it seems that the igp has better drivers, those numbers seem better than the one posted in hkepc.

But as far as the cpu goes as a whole at around $143 "rumored" its not bad depends how much the h55 costs tough. On AMD's side i suspect the Athlon x3 and x4 to fight this foe. But price to performance is certainly with the Athlon x4 in the tests, a $99 cpu out performed a $143 cpu in most tests. I do believe the i5 dual cores will fair much better "Turbo" but at $176 base, come on really would you pay $30 more for turbo "When OCing i7 920 i have to shut turbo off".

Only the lowest end G9650 seems good in price to performance but intel has cut off its L3 cache, non the less at $87 it seems to be a bargain.

[IntelUser]

You can notice on the earlier reviews like HKEPC and PCPop that the same i3 540 got only ~8K in multi-threaded Cinebench. Now it manages to get 10K.

The IGP and drivers still have room to improve considering how far it is from launch. If you translate the HKEPC review it says that the graphics driver is one that gets the IGP "just working".

[FischOderAal]

and being on-die.

It's not really on-die, I think on-package suits it better xD

But as far as the cpu goes as a whole at around $143 "rumored" its not bad depends how much the h55 costs tough.

Yeah, Intel needs to charge less for their southbridge. Now they are charging as much for the P55 as the P45 used to cost. Ridiculous...

Maybe NVIDIA will return

[kl0012]

But as far as the cpu goes as a whole at around $143 "rumored" its not bad depends how much the h55 costs tough.

There will be also Core i3-530 2.93 GHz $123.
Also it looks like Clarkdale has an excellent overclocking potential.
4.0 GHz for only 0.832v.

[IntelUser]

Nvidia will have a problem with integrated GPU chipset with the S1156 Nehalem uArch. DMI being only 2GB/s and being shared with I/O is the biggest reason.

The big IGP surprise will be Nvidia I think, at least on AMD platforms. If the 880G only performs 15% better than current IGPs per rumor mill, then Intel has a chance to beat AMD in IGP performance(hah, think about THAT!).

[ajaidev]

There will be also Core i3-530 2.93 GHz $123.
Also it looks like Clarkdale has an excellent overclocking potential.
4.0 GHz for only 0.832v.

Ya i know but the chip used by hexus is not a Core i3-530 but a Core i3-540 @ 3.06Ghz. The overclocking potential will also be very much there with Pentium G9650, the lack of HT may even help the OC figs, but 3MB L3 sux.

If Sandy bridge is an indication where things are heading after Clarkdale, you may need to upgrade your whole set up "Mobo+CPU" because the old mobo is not compatible with the new SoC. This lack of selective upgradeable sux but what to do its the future.

BTW AMD Fusion "LLano" is suppose to have a integrated gpu as powerful as Juniper "main stream"

[onethreehill]

http://anandtech.com/cpuchipsets/showdoc.aspx?i=3648
http://www.xbitlabs.com/articles/cpu...sneakpeek.html
http://www.bit-tech.net/hardware/cpu...-performance/1
http://hothardware.com/Articles/Inte...e-CPU-Preview/
http://www.legitreviews.com/article/1091/1/

[dinos22]

i really need that bios update heheheh

[ajaidev]

@onethreehill Thanks


EDIT: Even tough Anandtech was a bit informative, all benchs are intel slides they dont represent real performance. I don't think a dual core nehalem can really out perform a Q9400, the comparison is fishy to say the least

[gOJDO]

EDIT: Even tough Anandtech was a bit informative, all benchs are intel slides they dont represent real performance. I don't think a dual core nehalem can really out perform a Q9400, the comparison is fishy to say the least

I agree that most of the benches are synthetic and are not representing real-life.

But I believe that Clarkdale at 3.33GHz will outperform a Q9400 in most of the software. It has 25% clock advantage, its core has higher IPC, it has HT, it has two ODMCs and it has turbo. So, it will not outperform Q9400 in heavilly(almost perfectly scalling) 4 threads optimised code where little memory bandwidth is required and memory accesses are rare. But it will be very close. In all other cases, ex. single-threaded and dual-threaded code, heavily bandiwdth dependent code, and partly optimised 3 and 4-threaded code, Clarkdale @3.33GHz will outperform a Q9400.

[Smartidiot89]

I think there are three fundamental improvements at play here:

1) Slightly upgraded IGP arch (it's about due).
2) Increased IGP clockspeeds (I think there's a process shrink and there's definitely room for an increased thermal envelope from moving it into the CPU package [that has a much better guaranteed cooler] and out of the NB [that has a dinky passive thing plus all the other stuff on the same die])
3) Increased memory bandwidth due to improved memory controller on the Clarkdale (compared to G43/45's MC).

I could be totally wrong on all three points...I have no actual knowledge of the innerworkings of Clarkdale, if I did, I couldn't talk about it

I think that the IGP was shrinken from 90nm all the way to 45nm actually. And honestly I am VERY surprised I never in my life thaught that Intels IGP could match AMDs offering.

EDIT: I suppose it was 45nm

[v_rr]

hexus link isn't working.

[IntelUser]

with their inferior GPU design - not likely

You are assuming that an "inferior" design will stay throughout the lifecycle of the architecture(with various modifications). With Clarkdale, the Intel will fix various weak spots which are pulling performance down in real world scenarios.

Assuming 3DMark Vantage is correct, it'll be able to exceed the scores of Nvidia Geforce 9400 IGP per HKEPC:

http://global.hkepc.com/3878/page/6#view

G45: E3508
9400 IGP: E4445
1.5x G45: ~E5300

I think that the IGP was shrinken from 90nm all the way to 45nm actually. And honestly I am VERY surprised I never in my life thaught that Intels IGP could match AMDs offering.

By the way Smartidiot89, the current G45 is already 65nm.

v_rr: I think Hexus pulled the link probably by request of Intel