Hi all It seems that a lot of people here use Sandy Bridge processors and I can see why, they have excellent performance. However I was late to the party and decided to wait for Ivy Bridge because it promised lower power consumption. So how did it turn out. Well so far very well. The stand out feature of Ivy Bridge is low voltage, much lower than Sandy Bridge provided you don`t push the frequency too far. I am currently running my processor at 4.4GHz with Vcore of 1.100v. I have not found the limits yet, more testing is required. I`m getting a steady stream of P7903 on the CPU with TPF of 4min 6sec for about 21,000 PPD and P8020 on the GPU with TPF of 4min 13sec for about 19,600 PPD. This gives me about 40K PPD overall for power consumption from the wall of 277W. That`s 144PPD/W Using the computer for surfing the web slows down the SMP folding a little bit but seems to have no effect on the GPU. If I get a P8007 or P8010 on the GPU then output from the graphics card goes down to about 17,000 PPD and these work units use more CPU time so output from SMP goes down to about 18,000 PPD, the overall hit is about 5K I did get one P7809 on the CPU and it was running at about 30,000 PPD I have been mostly doing GPU folding up to now so I don`t know what to expect from these SMP projects. How do these numbers compare with other processors? Do not try to run Windows Vista on Z77 boards with Ivy Bridge. Asus states that this board does not support Vista but I completely failed to read the small print It seems the main problem is the Intel graphics drivers for HD4000/HD2500 are not available for Vista so I don`t get any of the fun features that the on board GPU has to offer and I can`t run LucidLogix Virtu. I am shutting down the system today to install Windows 7. See you soon.
Update I have got Windows 7 installed. The USB3 drivers have installed now, that`s something else that did not work with Vista. Still using the display from the graphics card at the moment, I will try out the on board graphics when I have time. In the words of the famous cyborg "I`ll be back"
Very useful information, thanks for posting. The Vista warning is a good tip. Are you using v7 folding and an AMD graphics card? From the big CPU hit I think this must be the case. Otherwise I think your CPU numbers fall in the right ballpark... I've lost track of the Sandy Bridge numbers at all sorts of overclocks. OK scratch that remark about the GPU - I've just read your sig.
Projects, Points and stability. I have had this machine up and running for about 10 days now and for most of that time the SMP slot has been running project 7903. This has given me a distorted view of the processors performance. This project seems to stress the cpu less than other work units which has enabled me to get away with a dodgy overclock for about a week. Now with different work units I am getting random crashes causing the computer to reboot. So far I have increased the Vcore by 20mV to 1.120V but I need to test it long term before I can say its stable. P7903 has been getting me 20,000 to 21,000 PPD, this was a bit less than I hoped for but still worth the effort. Now it appears that P7903 is one of the lowest value SMP projects. So far I have had P7809 at about 30,000 PPD, P7012 at about 40,000 and P7610 at about 25,000. These numbers are much more impressive, the 3770K is starting to look like a very good folder indeed Tweaks to try and improve the TPF of work units have not been successful. I have ended up going back to my original set up (Windows 7 64bit & FAH V7.1.52 with all default settings). If anyone has any tips for better performance then please let me know. It occurred to me that the best way to get more points was to get better work units So I changed the SMP slot type to advanced to see if it had any effect. Since then I have had a different project every time, hopefully this will be the key that unlocks the true potential of the 3770K. The quest continues.
Update Hi All My overclock seems to be stable now with no crashes for a week. I`m still running 4.4GHz at 1.120v, CPU temp is 39deg C but Core Temp reports the hottest core at 60deg C. I should own up at this point, I`m using water cooling (XSPC rasa750 RS240 kit). I am still getting a wide variety of work units on the CPU, the PPD is up and down all over the place. If I stop fiddling with settings and just let it fold I get a lot more points. More Power I have now added a second GTX560ti to this box. To reduce power consumption I have undervolted both graphics cards, now at 0.950v. To make this stable I had to reduce clock speed from 900MHz to 840MHz but this is still higher than a reference card. Power consumption depends on what work units the GPUs are running. Currently that`s 2 P8009s, these work units use less power than P8020s but they also give less PPD. Current power consumption is 366W, with the hottest work units I estimate about 406W. More Points The actual PPD depends on the mix of work units being processed at any one time, This is impossible to predict. The best combination would give me about 80,000 PPD, the worst combination would be about 45,000 PPD. Make room at the top
The SMP slowdown effect Hi again For the last 2 days my GPUs have been getting a steady diet of P8008 work units. I know from experience running this graphics card in my old rig that these work units slow down the SMP client a bit. Loosing 300 PPD from the Core 2 Duo was not much of a problem but how much was I loosing on the 3770K with 2 GPUs fighting for CPU cycles I had to find out. At this time there was a P7015 running on the SMP slot with TPF of 2m 6s giving about 31,000 PPD. First I paused both GPUs to let the SMP slot have maximum resources, The TPF went down to 1m 35s a massive difference. Next I paused the SMP slot and started up both GPUs to see how much CPU time the GPU slots were stealing. To my surprise CPU usage was very low, less than 5%. How is it possible that stealing such a small amount of CPU cycles can have such a big effect? I do not know the answer but here are some thoughts. 1. It must have something to do with how task scheduling works in multi-threaded applications. 2. The application can only run at the speed of the slowest thread. 3. If you run SMP 8 on an i7 processor, it can only work correctly if it gets 100% of CPU resources. More Testing I started up both GPU slots and set the CPU to run SMP 6. This gave a TPF of 1m 49s and a gain of about 7,500 PPD over SMP 8 However CPU usage was only about 80%, is there more performance to be had? Now I know you are not supposed to run odd numbers of threads but I just can`t resist it I set the CPU to SMP 7 and re-started the slot. Nothing exploded, so far it`s running normally and the TPF has gone down to 1m 40s, a gain of about 12,800 PPD over SMP 8 CPU usage is about 90% The computer is now much more responsive for other tasks like surfing the web. So far it`s a win / win situation. I`ll keep you posted.
I believe it is only advised not to run on uneven number of threads, but doing a systematic survey such as you are can give a better picture. Oh, and I believe it is prime numbers, not odd numbers, that can give poor results.
I think the reason there is such a big difference is, even though its only using a small amount of cpu load your graphics cards take so many percent from 1 core to run, and the way that the client runs is that all cores have to run at even speeds or process duties. If one cores performance drops it knocks back all the others, so by dropping off the 1 or 2 cores to give the graphics cards cpu power to use, the remaining cores can run at 100% with little to no fluctuations. Something like that anyway. I run mine on SMP 7 to alllow for other operations the computer does, which makes it more responsive to use and really no noticeable drop in ppd (as it is always doing other things as well as folding).
Currently running my server on an i7-2600k oc'd to 4GHz and get a TPF of 07:42 on 7 cores out of 8. Gives me almost 25k ppd. Running on Ubuntu 12.04. Temps just under 80°C. Thinking on upgrading my Q9450 to i7-3770k, so interested in your post quite a bit.
The 3770K Hi Scorpuk I am very impressed with the 3770K and found it very easy to overclock. I don`t have much to compare it to, the last system I built was a Core 2 Duo E6600. Check out this guide, this article and this article. The main problem with Ivy Bridge seems to be heat, the hotter it gets the less stable it is. If you increase the voltage it gets even hotter. Do not rely on the CPU temperature reported by hardware monitors, the core temps will be much higher. I use Core Temp and the sidebar gadget add on. I decided to go the whole hog and fit a proper water cooling loop. This may be overkill but it works.
I don’t think water cooling is overkill as I am thinking about it now for my i7-2600k server. 80°C is a tad high to be running 24/7. Unfortunately my case, compucase S411, isn’t designed with 120mm fan fixings. So might need to change the case. Went for this case as I was originally going to rack mount my server in the garage. Mind you that was over 4 years ago and its still in the house. lol I guess thats still the plan, so looking for a new rackmount that can take an H80 or equivalent.
Long term results Hi All The actual points this machine gets per day depends on the mix of work units it gets from the servers. The variation in performance of SMP work units is massive, this makes it very difficult to judge performance based on timing individual work units. A better way to assess performance is to take an average over a longer time period. The stats on extreme overclocking provide a rolling 7 day average PPD, this is a much more useful number but to be valid you need to let the computer run for 7 days without interruption or changes. Stats My average PPD is now 56,268. The graphics cards have been producing a steady 16,100 each so the SMP slot has averaged 24,068 PPD Not bad for running 7 threads in windows. The power consumption is 366W so overall efficiency is 153.7 PPD/W This is better than I expected, the 850W power supply does not even get warm. I think that a 3770K run as a dedicated folder on SMP should get about 30,000 PPD and over 200 PPD/W. Now where`s that empty case
Update Hi All I have been getting 8019 work units on the GPUs lately, this is good for points but these units cause a lot of screen lag which is very annoying. So I have finally switched the display output over to the integrated GPU. This has caused no problems so far. I had to install the Intel graphics drivers from the motherboard CD and reboot the PC then everything started up as normal. I have FAH running on 2 GTX560ti cards with Nvidia drivers and the display running off the IGPU with Intel drivers. I had thought that I would need to run LucidLogix Virtu in order to make this work, but it all runs fine without it. Power consumption is up by about 9W, but it is well worth it to have a more responsive PC. Before I would sometimes pause folding on one GPU to watch videos or TV, now I can leave everything folding all the time. See you soon.
Great work on the write up for IvyBridge bud, some great info there With your GPU/IGPU set up, where do you physically have your monitors plugged in to? I'm hoping you are going to say one of the 560's!? I am just starting to build my new Ivybridge rig now using an Asus Gene V, 3770k and Samsung 30nm RAM, it will of course be watercooled and running Windows 7 (as will be my main rig) with an Ubuntu VM for folding on, will let you all know what I get out of Ubuntu on it
Display and VM Hi Coolamasta I Use a single monitor with a DVI cable, I have this plugged into the motherboard and I set IGP as the primary output in the BIOS. If you want to play games without swapping the cable or changing the BIOS settings you could try installing the LucidLogix software that comes with Asus motherboards. Let me know how you get on with the VM. I did give it a try when I first built this machine but it wasn`t very successful. I still have the Virtualbox software installed and an Ubuntu iso on the hard drive so I could easily build another VM.
Cheers for the Info bud, fired up my 3770K last night on the Gene V and straight away clocked it up to 4.5ghz with a 1.215 vcore and CPU PLL fixed at 1.8v, RAM is at 1866mhz at the moment, seems stable so far but when I dropped vcore down to 1.200v strange things in Windows started happening so I think I am on the edge of stability at 1.200v Only played with it for a few hours last night and that was to build it as well, using an EK Supreme Waterblock, 18w DDC pump, EK pump top Res and a huge thick Phobya dual 120 radiator and temps @ 4.5ghz seem to peak around 72 degrees on the hottest core running LinX. More playing to be done this week and im going to try for 4.7ghz next and maybe even give 5ghz a crack too although I wont be leaving it at 5ghz, just want to see if "I can" lol Once settled on speed (prob will be 4.5ghz) then I will get the V7 FAH windows client on there and see what its putting out and then try the Ubuntu VM and see what that is doing as well
Update Hi All The 3770K is getting a slightly different mix of work units now and much more consistent PPD. With just SMP on 8 threads and no GPU folding its getting 32,000 PPD for 150W. That`s 213 PPD/W but this includes 2 GTX560Ti at idle. If I removed the graphics cards I could reduce the power consumption quite a bit. With both GPUs folding and SMP on 7 threads its getting 62,000 PPD for 400W. That`s 155 PPD/W, not bad at all. The SMP scores include the current 10% bonus for a4 work units. Overall I am very pleased with the performance. Coolamasta how is your Ivy Bridge project going? Any updates on folding or temps?
Your findings sound good to me bud, I've not folded on mine for a week or so as my SR-2 does most of my folding and the Ivybridge rig is my main machine I use for everything. After playing with a lot of multipliers and voltages then I've decided to stick with a 4.5ghz clock for everyday, (max temp is around 70 degrees while stressing) the heat just flys up after 4.5ghz, to hit a stable 4.6ghz even though I didn't need much more voltage the temps went up 12 degrees, I played with 4.8ghz but aborted the stress testing when the CPU started hitting over 90 degrees, that's just too hot for my liking and I was only around the 1.3v on the core The custom water loop I have is more than capable of cooling the 3770K and monitoring the water temp it hardly changes in temps from 4.5ghz to 4.8ghz despite more voltage so im convinced its purely just the limitation of the crappy TIM Intel used between the die and IHS In a few months I might get brave and de-lid it, de-socket clamp the motherboard and put the waterblock directly on the die core and push 5ghz+ out of it
