CPU/IGP tests

[bgt]

strange the 2500K does not have this big difference when using winrar and furmark? So the IGP does rely more on the cpu with the i7. Or the HT is interfering? If this continues to Haswell then you get even more dependency of the cpu side for the IGP? It shows the i7 definitely needs a real GPU to stay speedy.

 

[ShintaiDK]

The 3770K is alot faster and got alot faster IGP as well. Both will quickly get memory bandwidth starved when running 2 applications thats both very bandwidth hungry.

Your AMD setup would suffer the exact same case, had the CPU just been faster. And it screws the results that you use 1866Mhz memory there as well when you wish to cross compare them as you do.

 

[Idontcare]

the 4th value is nearly the same as the 1st so not really a change. Winrar just suffers when doing them both. Will try it with linx.

And while we are at it, don't forget to do the natural permutation here of running all three simultaneously as well

Hit the IGP, IMC, and Cores all at the same time. Cage match style :twisted:

(is it too early to make a request that you run crystaldiskmark in the background too? and a handful of torrents perhaps? )

 

[bgt]

Test
1. Linx
2. Furmark
3. Linx+Furmark

i7-3770K: 30Gflops, 351fps, 19Gflops(?)+ 265fps with separate GPU: 30Gflops, 491fps, 28Gflops + 491fps
A10-5700: 15Gflops, 421fps, 12Gflops + 365fps
FX 8350: 30Gflops, 541fps, 29Gflops + 516fps (separate GPU)

if you look at the numbers it shows the Intel IGP is very CPU dependent. Not nice for a laptop. AMD's IGP is quite indepent of the CPU(downside=more powerdraw).

 

[SPBHM]

interesting results, perhaps it's more related to the shared l3? but your concern about the laptops with Intel IGP, I don't know... it's hard to imagine this sort of load in any regular software... also the IGP and laptop users are hardly the ones loading CPU and GPU 100% at the same time!?

 

[ShintaiDK]

Test
1. Linx
2. Furmark
3. Linx+Furmark

i7-3770K: 30Gflops, 351fps, 19Gflops(?)+ 265fps with separate GPU: 30Gflops, 491fps, 28Gflops + 491fps
A10-5700: 15Gflops, 421fps, 12Gflops + 365fps
FX 8350: 30Gflops, 541fps, 29Gflops + 516fps (separate GPU)

if you look at the numbers it shows the Intel IGP is very CPU dependent. Not nice for a laptop. AMD's IGP is quite indepent of the CPU(downside=more powerdraw).

You still havent addressed the memory bandwidth issue. Yet you keep blaming the CPU as a conclusion?

Try test the 3770K with 1866Mhz memory as well and compare it to 1600. If it was CPU related the results will be roughly the same as on 1600. You already use 1866Mhz memory on the 5700, so it shouldnt be that hard.

You need to isolate the bottleneck before you can draw a conclusion.

 

[Idontcare]

Test
1. Linx
2. Furmark
3. Linx+Furmark

i7-3770K: 30Gflops, 351fps, 19Gflops(?)+ 265fps with separate GPU: 30Gflops, 491fps, 28Gflops + 491fps
A10-5700: 15Gflops, 421fps, 12Gflops + 365fps
FX 8350: 30Gflops, 541fps, 29Gflops + 516fps (separate GPU)

if you look at the numbers it shows the Intel IGP is very CPU dependent. Not nice for a laptop. AMD's IGP is quite indepent of the CPU(downside=more powerdraw).

Very interesting results :thumbsup:

The separate GPU result for 3770K vs using its IGP is very telling.

interesting results, perhaps it's more related to the shared l3? but your concern about the laptops with Intel IGP, I don't know... it's hard to imagine this sort of load in any regular software... also the IGP and laptop users are hardly the ones loading CPU and GPU 100% at the same time!?

IMO this is getting to the "snappy" aspect that people subjectively speak to when describing their real-world experiential observations with AMD systems running an iGPU.

In the test beds of hardware reviewers you rarely find anything resembling a real-world setup. Hardware reviewers don't load their system with all the standard background apps that the end-users do. No anti-virus loaded up and running in the background, no email app in the background that is periodically updating, twitter, chat, facebook. No realtime anti-spyware app loaded and running in the background.

But what we are seeing here is the beginning of some tantalizing data that provides us some insight into how or why an AMD platform might deliver a better experience over that of an Intel system once all the clinical laboratory conditions are no longer in play, once the rig is in real-world use, bogged down from 6 months of background auto-updating and the loading of intermittent cpu resources.

I distinctly remember the nascent days of SSDs, before light was shed by Anand's magnum opus which tore asunder the prevailing notion that unproven observations of SSD's performance "sucking ballz and stuttering" were scientifically confirmed.

The difference before and after the publication of Anand's SSD Anthology was that prior to its publication anyone who commented on their observations of quirky behavior with SSDs were immediately dismissed as anti-SSD FUDsters, and after Anand's publication anyone who didn't acknowledge and accept the stuttering reality of jmicron controller based SSDs were viewed as ignorant noobs.

Perception was changed overnight but the reality of the situation was unchanged.

And I personally believe that we have a similar situation today with iGPU and CPUs. We have reality, we have perception, and we have the public's popular opinion.

And right now anyone who has experience with reality that would leave them with the perception that one system is snappier than the other will face stigmatization in the face of the public's popular opinion which carries with it the perception that until someone like Anand cracks this nut and publishes an "iGPU Anthology" article any references to an AMD system feeling faster and snappier under real-world use scenarios is just making it up or biased.

Until Anand decides to pursue such an effort, the door of opportunity is still available to crack this wide open and shed some light shed on it within the enthusiast community.

I don't know if AMD's APU is really driving an experiential difference in real-world situations but I don't think anyone has really looked into it yet.

 

[NTMBK]

Very interesting results :thumbsup:

The separate GPU result for 3770K vs using its IGP is very telling.



IMO this is getting to the "snappy" aspect that people subjectively speak to when describing their real-world experiential observations with AMD systems running an iGPU.

In the test beds of hardware reviewers you rarely find anything resembling a real-world setup. Hardware reviewers don't load their system with all the standard background apps that the end-users do. No anti-virus loaded up and running in the background, no email app in the background that is periodically updating, twitter, chat, facebook. No realtime anti-spyware app loaded and running in the background.

But what we are seeing here is the beginning of some tantalizing data that provides us some insight into how or why an AMD platform might deliver a better experience over that of an Intel system once all the clinical laboratory conditions are no longer in play, once the rig is in real-world use, bogged down from 6 months of background auto-updating and the loading of intermittent cpu resources.

I distinctly remember the nascent days of SSDs, before light was shed by Anand's magnum opus which tore asunder the prevailing notion that unproven observations of SSD's performance "sucking ballz and stuttering" were scientifically confirmed.

The difference before and after the publication of Anand's SSD Anthology was that prior to its publication anyone who commented on their observations of quirky behavior with SSDs were immediately dismissed as anti-SSD FUDsters, and after Anand's publication anyone who didn't acknowledge and accept the stuttering reality of jmicron controller based SSDs were viewed as ignorant noobs.

Perception was changed overnight but the reality of the situation was unchanged.

And I personally believe that we have a similar situation today with iGPU and CPUs. We have reality, we have perception, and we have the public's popular opinion.

And right now anyone who has experience with reality that would leave them with the perception that one system is snappier than the other will face stigmatization in the face of the public's popular opinion which carries with it the perception that until someone like Anand cracks this nut and publishes an "iGPU Anthology" article any references to an AMD system feeling faster and snappier under real-world use scenarios is just making it up or biased.

Until Anand decides to pursue such an effort, the door of opportunity is still available to crack this wide open and shed some light shed on it within the enthusiast community.

I don't know if AMD's APU is really driving an experiential difference in real-world situations but I don't think anyone has really looked into it yet.

It also feels very similar to the recent situation with TechReport and stuttering in graphics cards. By altering how they test, what metric they look at and how they focus their testing, they were able to test and demonstrate an entire side of graphics performance which had been overlooked until recently.

Of course, replicating a "real world workflow" is something notoriously difficult to do, and which no-one has found a good solution to yet. I remember there were benchmarks back in the day like Winstone which claimed to, but I don't think they did a very good job of it.

 

[Idontcare]

It also feels very similar to the recent situation with TechReport and stuttering in graphics cards. By altering how they test, what metric they look at and how they focus their testing, they were able to test and demonstrate an entire side of graphics performance which had been overlooked until recently.

Of course, replicating a "real world workflow" is something notoriously difficult to do, and which no-one has found a good solution to yet. I remember there were benchmarks back in the day like Winstone which claimed to, but I don't think they did a very good job of it.

The bursty nature of background activity makes it extremely difficult to capture and interrogate in a lab-like environment...but the fact that hardware reviewers know they must setup their test rigs with pristine OS installations and zero background apps if they want to produce repeatible results is proof that their test conditions are not generating results that are going to be reflected by the end-user running the same app under real-world usage scenarios.

Hardware reviewers know their performance consistency falls apart once the real-world imposes itself onto their test platforms, so they do everything they can to keep their rigs from falling prey to the real-world. End-users don't have the luxury of following suit.

 

[bgt]

You need to isolate the bottleneck before you can draw a conclusion.

I will

 

[anikhtos]

The bursty nature of background activity makes it extremely difficult to capture and interrogate in a lab-like environment...but the fact that hardware reviewers know they must setup their test rigs with pristine OS installations and zero background apps if they want to produce repeatible results is proof that their test conditions are not generating results that are going to be reflected by the end-user running the same app under real-world usage scenarios.

Hardware reviewers know their performance consistency falls apart once the real-world imposes itself onto their test platforms, so they do everything they can to keep their rigs from falling prey to the real-world. End-users don't have the luxury of following suit.

just to drop a line about my findings
i have a laptop with i3-2330 cpu which is at 2.2ghz and has the hd3000

after playing a few games (old ones low settings) the laptop was hot and the fan was so noisy so i went to the control panel power options and made the cpu to throttle all the time at 800hz.

so i am now with 1/3 of the cpu power and i was afraid that the laptop would be not that responsive so i start doing the everyday things i do with it. Apart that the browser may take 1-2 sec on same pages more to load more or less the laptop felt the same. So next was playing again the games. The feeling inside the game was the same as before only this time laptop run much cooler and the fan did not work. Wonderful.

But a feeling is a feeling i run passmark (i know it is not the best) and run it again to compare. Cpu scores took a dive. But 3d performance stayed not only the same it gained 3-4%!!! Next round run the 3d test but with cpu z open. At amazingly only at 3d load the cpu turbo up until 2ghz!!! Not at any other load i see the cpu goes beyond the 800hz.

I imagine it is more to do with the tdp of the cpu.
When running the 3d the igpu turbos and manage to stay turbo longer but there is also room to turbo the cpu to profit of the igpu turbo.

okay maybe 2-3% is at the statistical boundary of error nevertheless it points out something. maybe intel was a bit too aggressive with the tdp at least with the laptop variants.

well that are may personally findings. Rerun passamark just now running also cpu z , hardware monitor, torrent open and this browser.
cpuz report again cpu at 2gz hardware monitor report tdp at 28watt for the cpu and passmark it still scored 1% better.

I am open to discussion about this findings

 

[bgt]

Test i7-3770K@stock without external GPU:

1. Linx
2. Furmark
3. Linx+Furmark

1333mc: 30Gflops, 351fps, 19Gflops + 265fps
1600mc: 30Gflops, 331fps, 21Gflops + 310fps
1833mc: 30Gflops, 335fps, 24Gflops + 305fps


1. Winrar
2. furmark
3. winrar + furmark

1333mc: 8300, 335fps, 3600 + 293fps
1600mc: 8800, 331fps, 4950 + 310fps
1833mc: 9300, 335fps, 5700 + 318fps

 

[ShintaiDK]

So its not CPU related, its purely memory bandwidth related :thumbsup:

 

[bgt]

So its not CPU related, its purely memory bandwidth related :thumbsup:

ShintaiDK, looks like it. In contrast to the A10-5700 which does have a very different hardware system. The GPU does not interfere much with the CPU and here the GPU speed is more memory speed dependent.

 

[ShintaiDK]

ShintaiDK, looks like it. In contrast to the A10-5700 which does have a very different hardware system. The GPU does not interfere much with the CPU and here the GPU speed is more memory speed dependent.

Well you do use 1866Mhz on the 5700. Plus CPU wise, the 5700 is pretty darn slow with half or less the performance. Again meaning the CPU part needs half or less the memory bandwidth.

 

[sm625]

Test i7-3770K@stock without external GPU:

1. Linx
2. Furmark
3. Linx+Furmark

1333mc: 30Gflops, 351fps, 19Gflops + 265fps
1600mc: 30Gflops, 331fps, 21Gflops + 310fps
1833mc: 30Gflops, 335fps, 24Gflops + 305fps


1. Winrar
2. furmark
3. winrar + furmark

1333mc: 8300, 335fps, 3600 + 293fps
1600mc: 8800, 331fps, 4950 + 310fps
1833mc: 9300, 335fps, 5700 + 318fps

Very good data. It tells us that the chip basically needs a third memory channel in order to make the IGP truly viable. Either that or 2600 mhz memory.

I wonder how the scores would look if the cpu clock speed was capped at around 2.0 to 2.5GHz and 1833mc.

 

[ShintaiDK]

Very good data. It tells us that the chip basically needs a third memory channel in order to make the IGP truly viable.

Hell no, lets not go back in time. I already shievers when I see LGA2011 and get bad flashbacks from the FP/EDO times.

Its way overdue for 128 or 256bit DIMMs

 

[Lorne]

BGT:, Was the I7 Lin-X test w or wo HT?.

 

[bgt]

BGT:, Was the I7 Lin-X test w or wo HT?.

with

 

[bgt]

Well you do use 1866Mhz on the 5700. Plus CPU wise, the 5700 is pretty darn slow with half or less the performance.

Be real now........it has only 2/4 threads i.o 4/8 from the i7. It shows how well the 5700 performs.

 

[Lorne]

Scores will be lower with Lin-X with HT on as it sucks up all memory bandwidth and starves CPUs, But I believe its other reasons as well.

In my opinion Lin-X has an averaging system problem that seems kind of skewed across all cores insted of total work done.