My i7-3770S Review

sm625

Diamond Member
May 6, 2011
8,172
137
106
I just put together these parts:

GIGABYTE GA-Z77-D3H
i7-3770S
16GB RAM
256GB SSD

Needless to say this thing is very fast. 7.8 processor and memory score, 7.9 hard disk score. It took less than an hour to install windows, the gigabyte driver disk, and fully update it.

I moved one slider in the bios and it runs pretty much constantly at 3.94 GHz, unless idle. It is supposed to be 4.02GHz but I never saw that. Even with 8 threads of prime95 it is supposedly running at 3.94GHz.

Temperature under 8 thread prime95 load is 95C, after 10 minutes. Who knows where it would top out...

Super pi 1M time is 9.34 seconds.

With 6 prime 95 worker threads set to cores 0-5, and super-pi set to core 6-7, super pi time drops to 10.99 seconds.

Both times are very consistent. I repeated this test several times.

I assume there is some throttling going on though it is nothing too terrible. CPU-Z still insists it is running at 3.94GHz.

Now here is where it gets interesting imo.

In addition to the previous test I added Furmark into the mess. So I had the following test configuration:

Furmark 800x600, no MSAA, windowed mode. Burn-in running, affinity set to cores 0-5
Prime 95 6 worker threads running, large FFT, affinity set to cores 0-5
Super-pi set to cores 6-7
I set everything else (dwm, explorer, etc) to cores 0-5
Aero enabled


And with that test configuration, super pi 1M time drops to 17 seconds. Very repeatable. Temperature is around 85-88C when running these tests.

The system is very unresponsive when furmark is running. It feels like an atom netbook even when you're just browsing sites like MSN. Even with no cpu load at all.

This HD4000 gpu runs furmark 800x600 windowed at 12 fps, which is 3 times faster than the Nvidia 9xxx series discrete gpu I have in my other machine. Yet my other machine does not slow down this badly when furmark is running. Furmark running has virtually no effect on my general usage experience on my machine that has the discrete gpu. But on these integrated graphics systems, the effect is horrific. I noticed the same thing on my llano system. If you load the gpu it inexplicably slows down the whole system to a crawl. This does not happen the same way with discrete cards, even a supposedly far less capable discrete card! To my knowledge, not one tech review site has revealed this problem. And it definitely is a problem.

I would very much like to see someone with a properly cooled i7-3xxx set their cpu to 4.0GHz and repeat the super pi test in the above test configuration.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Thats the downside of the shared memory bandwidth by the iGPU. Basicly all we need is some kind of bandwidth control system or load balancer. That applies to both AMD and Intel.

DDR4 wont help this issue, because iGPU will just use more bandwidth and the same happens again.

Your Prime95 vs SuperPi also seems normal due to memory load and lower turbo.

Furmark is singlethreaded btw. Atleast the version I use.
 
Last edited:

sm625

Diamond Member
May 6, 2011
8,172
137
106
I played a game of League of Legends on it. It can run 1440x900 windowed mode on max settings, with one exception. You have to turn shadows just one notch down. It gets like 60-90 fps with shadows turned down one notch. But with shadows at max it gets 22 fps. Its hard to even tell the difference between the two settings.
 

T_Yamamoto

Lifer
Jul 6, 2011
15,007
795
126
I played a game of League of Legends on it. It can run 1440x900 windowed mode on max settings, with one exception. You have to turn shadows just one notch down. It gets like 60-90 fps with shadows turned down one notch. But with shadows at max it gets 22 fps. Its hard to even tell the difference between the two settings.

On the igp?
 

sm625

Diamond Member
May 6, 2011
8,172
137
106
Why not? Newegg was out of all the other ib 8 thread chips. :) But even so it is priced fairly. It is the cheapest way to get 8 threads and a super pi time of 9 seconds. A 2600 is $20 less (~7%) but the IPC is 7% less too. K version pointless because of stock cooling.

I didnt build this for myself, of course. For me I would have dumped the HT and spent the money on a K version and a cooler. (And new TIM lol)
 
Last edited:

lakedude

Platinum Member
Mar 14, 2009
2,778
528
126
1) Got a 3770s and a CM 212 pro or plus or something. It keeps Ivy much cooler. Would be happy to duplicate except have discrete graphics with the 4000 shut off. Would take some monkeying around...

2) When you loading up a computer with every torturous workload you can find it isn't all that surprising that the system acts sluggish. I'm guessing you could turn the priority of the benchmarks down in software to restore some user control. I load my system up with BOINC (both GPU and CPU tasks) yet the system is fully usable because BOINC runs in a low priority (and I have a discrete GPU).
 

sm625

Diamond Member
May 6, 2011
8,172
137
106
Super pi 1M time is 9.34 seconds. With nothing else running.

With 6 prime 95 worker threads set to cores 0-5, and super-pi set to core 6-7, super pi time drops to 10.99 seconds.

Can you try this test with your discrete gpu? If your scores dont shift like mine did then it would be fair to say that it is the igp that is causing the slowdown, even when it is only handling aero.
 

lakedude

Platinum Member
Mar 14, 2009
2,778
528
126
I'll try running those tests late tonight. Will try various combinations...

FYI Cooler is a CM Hyper 212 plus, temps running BOINC on 6 out of 8 cores plus 4 GPU tasks (80% CPU load, 50% GPU load) mid forties to mid fifties. Highest I've seen has been 56C. The CPU is drawing about 41 watts, total system power is about 165 watts according to the Kill-a-watt.

Running Kombustor (like Furmark, loads up HD-7850 GPU) and LinBench (loads up CPU) Total power was around 260 watts. I think that is amazing for a system this powerful...

System idle power is around 55 watts...
 

lakedude

Platinum Member
Mar 14, 2009
2,778
528
126
Interesting...

Been monkeying around with Super Pi and overclocking in the BIOS.

Edit: got it figured out better...

My CPU/BIOS allows for a 4 "bin" increase on all 4 turbo settings so I cranked em all up. Max single core went from 39x to 43x. Max 4 core went from 35x to 39x. When all 4 cores were loaded before the CPU ran at 3.5GHz (in spite of being called a 3.1GHz part). Now the system will run 3.9Ghz all the time and will turbo up to 43x in lightly threaded apps. Power use went up about 15 watts running BOINC. Temps slightly higher with the Hyper 212 plus keeping em under 60C.

Super Pi is coming in around the high eights. Scored 8.5 with a 103% clock speed setting...
 
Last edited: