Overclocking the Ivy Bridge HD4000 IGP?

[Fjodor2001]

Hi,

Overclocking the Ivy Bridge HD4000 IGP has been tried and documented here:

http://semiaccurate.com/2012/04/23/overclocking-intels-hd-4000/

Apparently they achieved a 39% overclock, still the frame rate only improved by about 16%. How can that be? Why didn't the frame rate increase by about 39%? Is it because the memory bandwidth is the bottleneck, or something else?

 

[Kristijonas]

Yes, HD4000 uses RAM for memory bandwidth. Upgrading RAM to faster ones will make integrated graphics faster. Fast dual-channel RAM + overclocking could possibly increase fps by 33% in some cases, I think.

 

[borisvodofsky]

Interesting, but it's like me trying to overclock a geforce 4 Ti 4200 to 4600.

Not all that useful, still cool though. I had to swap the capacitors on my Ti4200 because the old ones blew due to the overclocking

 

[podspi]

Interesting, but it's like me trying to overclock a geforce 4 Ti 4200 to 4600.

Not all that useful, still cool though. I had to swap the capacitors on my Ti4200 because the old ones blew due to the overclocking

It was useful at the time :biggrin:
Free performance is free performance...! I wonder what the bottleneck is if an increase in clockspeed didn't push up performance much. Has anyone seen anything written about HD4000's scaling w/ clockspeed?

 

[Kristijonas]

It was useful at the time :biggrin:
Free performance is free performance...! I wonder what the bottleneck is if an increase in clockspeed didn't push up performance much. Has anyone seen anything written about HD4000's scaling w/ clockspeed?

I doesn't work that way. Performance increases as much as you up the clock speed. It doesn't just stop at some point just because memory bandwidth starts bottlenecking it. I'd say just overclock it to the 100% stable max. That's what I did with intel HD 2000 on i3 2120. And I felt the difference. Intel HD 4000 should not be underestimated. It can play many games on decent graphics smoothly. For many casual players that are into strategy and RPG games it can really be enough.

 

[borisvodofsky]

It was useful at the time :biggrin:
Free performance is free performance...! I wonder what the bottleneck is if an increase in clockspeed didn't push up performance much. Has anyone seen anything written about HD4000's scaling w/ clockspeed?

It doesn't have nuff' shaders, clock speed won't matter as much if you can't handle the high volume of effects. :ninja:

But memory bandwidth does help significantly. Try going 2133 :biggrin:

 

[LOL_Wut_Axel]

Hi,

Overclocking the Ivy Bridge HD4000 IGP has been tried and documented here:

http://semiaccurate.com/2012/04/23/overclocking-intels-hd-4000/

Apparently they achieved a 39% overclock, still the frame rate only improved by about 16%. How can that be? Why didn't the frame rate increase by about 39%? Is it because the memory bandwidth is the bottleneck, or something else?

Because, as I've said some time before, Intel's IGP architecture cares more about core configuration than clock speeds. That's why there's such a small difference in performance between the mobile Core i3's HD 3000 and the Core i7's.

 

[borisvodofsky]

Because, as I've said some time before, Intel's IGP architecture cares more about core configuration than clock speeds. That's why there's such a small difference in performance between the mobile Core i3's HD 3000 and the Core i7's.

burr.. wha? I need elaboration on this core configuration difference.

 

[IntelUser2000]

It doesn't have nuff' shaders, clock speed won't matter as much if you can't handle the high volume of effects. :ninja:

Memory bandwidth is one reason:
http://www.computerbase.de/artikel/arbeitsspeicher/2012/test-welchen-ram-fuer-intel-ivy-bridge/5/

Sometimes, the bottleneck is very strange. Exclusively for shared systems like iGPUs, it may be a CPU bottleneck, but not in a traditional sense. In low TDP chips, maybe the CPU core doesn't have enough power allocated. Or maybe the shared system itself is crippling the CPU with longer latencies, conflicts, less bandwidth, etc.
http://www.anandtech.com/show/4083/...core-i7-2600k-i5-2500k-core-i3-2100-tested/13

Another reason is not fully optimized drivers.
http://ht4u.net/reviews/2011/treibervergleich_intel_2361/index3.php

You can see the new drivers allow in some cases, HD Graphics 2000's performance to be almost on par with the HD Graphics 3000. Also, the HD Graphics 3000 gains better with newer drivers than the 2000.

 

[Fjodor2001]

Memory bandwidth is one reason:
http://www.computerbase.de/artikel/arbeitsspeicher/2012/test-welchen-ram-fuer-intel-ivy-bridge/5/

Sometimes, the bottleneck is very strange. Exclusively for shared systems like iGPUs, it may be a CPU bottleneck, but not in a traditional sense. In low TDP chips, maybe the CPU core doesn't have enough power allocated. Or maybe the shared system itself is crippling the CPU with longer latencies, conflicts, less bandwidth, etc.
http://www.anandtech.com/show/4083/...core-i7-2600k-i5-2500k-core-i3-2100-tested/13

Another reason is not fully optimized drivers.
http://ht4u.net/reviews/2011/treibervergleich_intel_2361/index3.php

You can see the new drivers allow in some cases, HD Graphics 2000's performance to be almost on par with the HD Graphics 3000. Also, the HD Graphics 3000 gains better with newer drivers than the 2000.

Thanks. You mentioned 3 different reasons, but I could only see the first one having a possible relation to overclocking the IGP, see details below:

1. I could see the connection between overclocking the IGP and the memory bandwidth being a bottleneck (i.e. the first case you mentioned). Because all things equal, increasing the memory frequency improved the frame rate. If the memory bandwidth would not be a bottleneck, increasing it should not have had any effect.

2. However, the second case regarding the CPU being a bottleneck, I could not find any explanation in the article you linked to mentioning or clarifying that, or saying it has anything to do with overclocking the IGP.

3. And then finally you mentioned the drivers being the problem. However I could not find any connection to overclocking the IGP there. All the article said was that with newer drivers, you got higher frames per second regardless of whether using the HD2000 or HD3000. It did not mention any connection to overclocking the IGP. So couldn't it just be the result of the newer drivers having more optimized SW, thereby running faster just like any other SW that is optimized?

 

[Fjodor2001]

So to summarize my previous post, the only reason I've found so far that explains why overclocking the IGP by X % doesn't result in X % higher FPS seems to be limited memory bandwidth. Unless I've missed something?

If so, does that mean that it should be possible to achieve 39% higher FPS by overclocking the Ivy Bridge IGP as mentioned in my original post (if the memory is overclocked enough so it no longer becomes a bottleneck)? In that case it should bring IGP in IB up to about the same performance level as is expected by the Haswell GT2 IGP, right?

And if that is the case, is Intel reaching a limit to how much faster their IGP can get? I.e. there is no use increasing the number of EUs or IGP clock, since the memory bandwidth is the bottleneck anyway?

 

[ShintaiDK]

If I OC my GTX 680 core with 20%. The card doesnt get 20% faster either

 

[MacGyverSG1]

The only way to really know is to experiment. I have read that increasing system memory speed will help Intel's IGP performance (also overclocking IGP speed). 1600MHz seems to be the usual IGP overclock. Try running some game benchmarks at 1333, 1600, 1866, 2133, 2400 memory speeds (if possible).

Grab some of that Samsung 30nm RAM. It can be run from 1066-2666. It pretty cheap also.

When I finally get my new computer running (i7-3770K), I am going to try out all the memory speeds I can (have 16GB of Samsung 30nm).

I just need to pick a motherboard. :hmm:

 

[MacGyverSG1]

Check this article out...http://semiaccurate.com/2012/04/23/overclocking-intels-hd-4000/

 

[dagamer34]

So to summarize my previous post, the only reason I've found so far that explains why overclocking the IGP by X % doesn't result in X % higher FPS seems to be limited memory bandwidth. Unless I've missed something?

If so, does that mean that it should be possible to achieve 39% higher FPS by overclocking the Ivy Bridge IGP as mentioned in my original post (if the memory is overclocked enough so it no longer becomes a bottleneck)? In that case it should bring IGP in IB up to about the same performance level as is expected by the Haswell GT2 IGP, right?

And if that is the case, is Intel reaching a limit to how much faster their IGP can get? I.e. there is no use increasing the number of EUs or IGP clock, since the memory bandwidth is the bottleneck anyway?

Overclocking is never a linear increase in performance.

 

[Fjodor2001]

Check this article out...http://semiaccurate.com/2012/04/23/overclocking-intels-hd-4000/

Check out the link in the original post in this thread, does it look familiar? :whiste:

 

[LOL_Wut_Axel]

burr.. wha? I need elaboration on this core configuration difference.

EUs. HD 3000 has 12 and HD 4000 16.

 

[Fjodor2001]

Overclocking is never a linear increase in performance.

True, but the goal is always to get as close to that as possible. Clearly a 16% FPS increase for a 39 % IGP overclock as in the article in my original post is not very good. It ought to be possible to improve on that...

 

[Fjodor2001]

The only way to really know is to experiment. I have read that increasing system memory speed will help Intel's IGP performance (also overclocking IGP speed). 1600MHz seems to be the usual IGP overclock. Try running some game benchmarks at 1333, 1600, 1866, 2133, 2400 memory speeds (if possible).

Grab some of that Samsung 30nm RAM. It can be run from 1066-2666. It pretty cheap also.

When I finally get my new computer running (i7-3770K), I am going to try out all the memory speeds I can (have 16GB of Samsung 30nm).

I just need to pick a motherboard. :hmm:

I agree. However I have not bought my IB system yet, so I cannot do that. If anyone else can do some experiments as you described above that would be really interesting though. If it turns out that memory overclock will help when also doing IGP overclock, I might decide to get some "overclocking friendly" RAM for my system...

 

[Don Karnage]

The only way to really know is to experiment. I have read that increasing system memory speed will help Intel's IGP performance (also overclocking IGP speed). 1600MHz seems to be the usual IGP overclock. Try running some game benchmarks at 1333, 1600, 1866, 2133, 2400 memory speeds (if possible).

Grab some of that Samsung 30nm RAM. It can be run from 1066-2666. It pretty cheap also.

When I finally get my new computer running (i7-3770K), I am going to try out all the memory speeds I can (have 16GB of Samsung 30nm).

I just need to pick a motherboard. :hmm:

I've seen people struggle to hit 1866 with that samsung memory. If you want truly fast memory grab this
\
http://www.newegg.com/Product/Produc...82E16820231587

Rated for 2400 but does 2800 easily.

 

[IntelUser2000]

All the article said was that with newer drivers, you got higher frames per second regardless of whether using the HD2000 or HD3000.

Yes, that's true. But the gains with HD 3000 are higher as you can see from that review. Sometimes driver bottlenecks limit higher end parts from performing better.

2. However, the second case regarding the CPU being a bottleneck, I could not find any explanation in the article you linked to mentioning or clarifying that, or saying it has anything to do with overclocking the IGP.

Sorry, I forgot to explain in the midst of changing the post multiple times.

That may have been a mistake of mine linking that post. But that HT4U link with driver tests also supports theory #2. You can see the gains are greatest in lowest settings, where the CPU bottleneck is greatest. They talked about reduced CPU cycles with Sandy Bridge running drivers, and that may have been further improved in the newer drivers.

I've ran a Crysis benchmark of the improved drivers. The range of improvement is greatest at 800x600 Low with gains being over 30% and progressively decreasing as details and resolutions increase. At highest settings and resolutions the gains are nonexistent.

 

[MacGyverSG1]

Check out the link in the original post in this thread, does it look familiar? :whiste:

My bad. I did not even see it. That's what happens when you speed read through all the posts.

 

[MacGyverSG1]

I've seen people struggle to hit 1866 with that samsung memory. If you want truly fast memory grab this
\
http://www.newegg.com/Product/Produc...82E16820231587

Rated for 2400 but does 2800 easily.

I have read a few reviews and threads on the Samsung Green 30nm and most have no trouble hitting 2133MHz with better timings than the Trident stuff. 2400MHz stable doesn't seem to hard either, but 2133 is probably the sweet spot.

I haven't read much good about the Trident stuff. 2400MHz is fine (or lower), but not much headroom without lossening the timings too much. I suppose newer stock may get better over time.

I really like the low profile of the Samsung. The heatsinks are not even needed for DDR3 RAM. The Samsung overclocked is warm to the touch.