The Official PC building thread - 4th Edition

It's been a long time since I've played Battlefield despite owning both BC2 and BF3, although I have to say I think I actually prefer the graphical style of BF3.
I'd also add that I never really dabble with graphics overclocking, the gains are always fairly minimal compared to the hassle and aggravation of it all, far from what you experience overclocking a CPU.
I realise that the HD7900s overclock much better than most high-end cards before them, but since you have so many stability problems I'd just stay well away from them.

I have been playing a fair bit of Global offensive lately and it's great fun (if you object to buying a weapon every round Rich, play demolition or arms race, as they are gungame-type modes where you cycle through the weapons list by getting kills).

I still maintain that given how your stability problems vary depending on the slot the cards are installed in, I would have replaced the motherboard months ago - I spent months tearing my hair out trying to diagnose stability problems once I started using two graphics cards, and on three separate occasions it was the motherboard at fault.
Since I don't overclock my cards nor do I attach multiple displays to my games PC any more, I've left ULPS enabled. It's handy for curtailing noise and power consumption when at idle.
Not sure why you're advocating me HD7950s given that I already run two HD6970s - the difference is typically 15-20% at best with stock clocks, certainly not worth the outlay! Even HD7970s can't be justified for their performance gains. More than that, it's been a long time since I've actually played a game that stresses my PC that much, I haven't really gone for the demanding games recently!

FPS limiters in games annoy me, considerably. Anything that artificially tries to limit a frame rate will by function reduce the fluidity of the game, which is noticeable. It's this reason why I never bother with VSync, tearing isn't that often an issue in modern games outside of OpenGL.
Crossfire took me a while to get working properly in Bad Company 2, about 7 or 8 months after release, but I got there in the end. BF3 has worked flawlessly with crossfire since shortly after its release when crossfire support was added, no more black screen flashing and random corruption etc. other than the occasional green flash which was a game bug and nothing to do with graphics drivers/crossfire.

Not true actually, although advertised to do such, the Toughpower 750W is a single-rail unit, as all 4 rails are soldered together and there are no splitting current limiters. It's a single rail power supply, pure and simple.

The VDDC reporting issues may have something to do with disabling ULPS, that does sound like a possibility, as I know when ULPS is active, sensor information being fed from the cards is disabled - that's by design, not a fault, so perhaps ULPS is being falsely triggered (or one card is dropping out of crossfire due to poor game support, hence entering ULPS and losing you the sensor info).

As said Rich, VSync has multiple frame rate caps, 60, 45 and 30 I've seen, I've not ever tested lower, but I believe there's a 15. That's how VSync behaves - if you can't keep up with a certain frame rate level, the limiter is dropped to the lower, unless sufficiently many frames are being discarded to warrant the limit raised - this continuous switching between levels is very noticeable in game, and is another reason I don't use VSync.

Fortunately, the part failure on my 3008WFP is a common flaw that affects only the 3008, your older 3007WFP is immune to it, as the power supply used is different, due to the lower power draw of the earlier 30" displays.

Card 1 in your system will consume more power than Card 2 as it's the master card - it not only has to render half the scene as the second card does, but it also has to assemble the split scene and render it to output over the DVI, which the slave card does not need to do.
When overvolting anything, remember the effect is exponential - overvolting graphics cards makes them get very hot, very quick, hence why it is to be avoided, especially when using stock cooling. A stock GPU will be happy until 95ºC+, and safe (i.e. not permanently damaged) up to 130ºC should this happen by accident.
Once overvolted this drops way down to about 80ºC max before you risk damages, and the further you go the lower the temperatures can be before you risk damaging the GPU.
Two lessons to learn from that - one, if you're running stock and seeing 85ºC or so load, think nothing of it, that's normal behaviour. Two - if you're overclocking, be damn careful about the temperatures, previous known temperatures against certain loads will not apply and you'll have to pay close attention to the temperature sensors.
In all honesty, just don't do it. The sort of graphics overclock that needs more volts is the sort of overclock that nets almost no benefit, and is the sort of overclock to avoid.

I dread to think how many sheets of paper's worth of posts I've just had to read, but I daresay it would fill a book - fullscreen at 2560x1600 this thread occupies just over 30 full screen widths (so given the taskbar and tabs menu, about 40,000 pixels!)
Also, as Jeff said, this is not the PC gaming/graphics card thread!

 

Wow, amazing information.

This will be the first of the new much shorter posts, lol.

Sam thanks for sharing your deep knowledge, and for explaining why the first card will always be working harder and pulling more power, and also for cluing me in, that the toughpower rails are soldered together - that is actually great news - I had no idea.

Your overvolting discussion left me very concerned, and resonates with what Jeff has been saying, so I'll drop the 1087 on the Trixx settings down to 981 tonight and see if that is stable, if I ever run the 975 clock again.

Wow, that is helpful. Knowing that there are instances where sensor information is no longer furnished, as in ULPS, DOES start explaining things a bit.

Regarding support, would you recommend that I upgrade to the brand new catalyst? I always keep the older versions, for downgrading back if I have to.

Yes, I would like to try those other gaming modes, which in addition gives one more experience with each weapon. I know that they have tried to simplify a bit to introduce a whole new generation of gamers to counter strike.

The buying is slow at first, but it is a little easier if you can hit F2 and buy the same stuff each time, or hit F1 and do the auto-buy. It's nice on casual to not have to buy armor or ammo. It's also nice that they allow each type of grenade on a different key, but to then have to throw it, with the fire button, makes it slow. I love the middle mouse throw grenade of BC2 and most of COD - hit one button and the grenade is launched.

If I ever get around to it, I need to enable the two extra buttons on the SteelSeries laser mouse, which are totally programmable with as complex a key sequence including pauses as is needed, for a quick flashbang pullup and throw all in one button, and maybe the other button for the same thing with a frag grenade. Including middle mouse I could add smoke to that. It is kind of neat that they added in molotov - that makes for a graphically more exciting game - all that fire! But I don't know if I'll ever get around to that!

Jeff, regarding temperatures - again this goes back to Sam's discussion about

higher temps might not be too bad, but shouldn't be mixed with higher-than-stock voltages, to avoid long-term chip damage.

I likely am over-panicking, but somewhere along the line I got the idea that the new 7000 family is not our father's AMD card, and cannot take those higher temps that didn't phase my 8800gtx or my 3850. I got the feeling that the 7950 should be kept under 80, and even under 70 if I can. The XFX that I returned ran at 82 consistently, even at only 900 core clock, and hung every 10 minutes in crysis. At stock clocks it was still hot, and I could only get a little more play time out of it. That made me think the 7950 doesn't like to be in the 80s.

The gigabyte 3-fan card ran cooler, but threw so much heat into the case that I was getting blue screens every hour - telling me that my 9450 overclock was no longer stable with all that extra case heat (back in the small antec sonata case.) The compelling thing that first attracted me to the 7950 HIS IceQ was the traditional single turbine fan design blowing most of the heat out the back - all the others had two fans. Coupled with the very good cooler, I am used to this thing never even hitting 60 in normal play.

But you're probably right - I am likely just over-reacting to the artifacts of the Power Color card developed right after I let it hit 90 on that one Furmark test when I didn't think 90 was an excessive temp. While Sam has said different, the belief that I burned up the card has lingered, so I am haunted by anything that reminds me of high temps, kind of like how Kevin is haunted by leaky water cooling.


As an aside, I know there are only a few gamers here, but I thought it was interesting that Russ recently commented on the Far Cry 2 screenshots, calling it a cinematic title, and he implied that he has tried out the game a bit and was glad that his hardware had the power to manage it. Russ have you actually ever tried boating down the river?

Good point. You know, I guess I was only trying to save a bit of energy, haha. But at the very end of the evening, when I got the 30 fps working again, and ran Wei Shen around the map a bit - I asked myself "Is it totally fluid, yes or no." On a 3rd person I have trouble deciding, since controlling the character, especially when he is running, is a little awkward compared to first person. I found the same on Spec Ops: The Line. So you may be right about loss of fluidity. I'll probably disable the 30 cap tonight.

[After all that work - but the pseudo science was worth it!]

Rich

 

Only ever really use Vsync when my frame rates are ridiculously high ie Half Life, Left 4 Dead, etc. When you have enough overhead it's very nice. No screen tearing

 

Half-Life, Left 4 Dead etc. are good examples of games I do actively play and get huge frame rates, but really, I hardly see any tearing in those titles, if at all.

With drivers, never update 'for the sake of it' if stuff works fine, but if things don't work fine, as in your case, you have nothing to lose by trying an update.

If ambient heat from a graphics card is enough to upset your CPU overclock, it probably isn't wholly stable to start with, when my CPU heatsink was full of dust, my CPU was reaching well into the 70s celsius at load, which I wasn't happy with, but this brought no stability concerns. If your CPU is that heat-sensitive, I think you need to tone down the overclock there a little too.

 

Do notice tearing quite often without Vsync. Might be a matter of tolerances. I generally don't use Vsync at all save for those few examples.

Agreed that ambient heat from a video card should not be affecting the stability of other components. That seems to suggest other issues.

 

Thanks for the interesting comments.

Well, regarding the new catalyst, Sam, I guess I could try it - you're right - I can't say everything is optimal - certainly in regard to BC2 that one card doesn't work at all.

Regarding my cpu, on the other hand, Sam and Jeff, that no longer concerns me in the slightest. My 9450 overclock has been rock solid with the HIS IceQ cards - the blue screens are a thing of the past. Besides running only the HIS IceQ cards which have the single turbine rear-exhausting fan, I am now in the new gaming case with terrific ventilation. Last night's OSD showed cpu temps all under 60 most of the time (at 71 I get linpack errors) with cpu loads of around 80 percent. Prior I had the two kazes as my only intake fans in the beautiful mid-tower Antec Sonata non-gaming case, and I was fighting a heat problem the whole time.

If the second HIS card hadn't bottomed out on that case, I would never have pulled the thermaltake spedo case out of its box. But I am so glad I did - I learned so much. In fact I have a lot of material for future (short) builder posts - positive case pressure - research proving demciflex filters are far superior to anything else - a custom kama bay mod since you can't buy the kama bay fan mounts anymore, etc.

So I am finally in a well-breathing case like the HAFs that you guys have, you Sam and Kevin, and like your CoolerMaster, Jeff, with those great pictures you recently posted. I even have the same kind of terrific cable management.

Regarding tearing, I have always been curious as to what it is - I have tried to see it - but more like you, Sam, I don't seem to experience it. I got ridiculous frame rates of 150, or was it 220, on Left 4 Dead 2, which I played for about 15 hours, a week or so ago - not on crossfire. I just let it rip at max everything - again vSync does not seem to increase fluidity for me, rather the opposite. (When I go back to L4D2 now on crossfire, who knows what frame rates I'll get?)

Last night's Sleeping Dogs crossfire testing:

RATE CAP:
I ran crossfire, stock clocks. I disabled the 30 fps cap, and thanks for explaining that it is basically just vSync, and it seems you were quite right, Sam, it's more fluid without the cap. My frame rates were in the mid 30s to low 40s - as I say the play seemed very fluid other than the general difficulty of controlling the character movement, like when he is chasing somebody and has to hit the center gate part of the fence to jump over. Another difficult time: I had to turn on a walk-through video on this computer while gaming on the other, to see how to do the jump-vehicle-takeover, on the cake van, after replaying that sequence 6-8 times and have him ultimately fall off the van on the successful jumps, or have the van catch on fire due to my ramming it repeatedly, which the game didn't accept, since that ruined the cake, lol.

The trick: you actually have to guide your character (with the normal A or D key) who is hanging with his fingers onto the van roof railing, to get him to slide over to either front door, at which point the usual vault heist animation takes over and he commandeers the vehicle. That seems quite obvious now, but you don't have to do anything for heisting a regular car, once you make the jump. My lack of creativity blatantly reveals itself in those types of situations - thank goodness for walk-throughs.

TEMPS:
I also let the fans run on auto, Jeff, just to see. Both gpus were showing 98 or 99 load. My temps were 75 for the hottest card. The fan was running at about 2400 rpm, on the hottest card, which is near-silent, and only 1400 rpm on the cooler card running at about 67 degrees. (Max rpm is about 3200 with a noticeable high-pitched whine. You start to hear the fans at about 70% tach rate, somewhere around 2950 rpm.)

I played for 6 hours or so, stock clocks, and all was good but the game hung once at hour 4. I was able to get to task manager which informed me the game exe had stopped working, and I just went right back in it.

I actually didn't think much about it until earlier today. However, on reflection, I realize that in 30-40 hours of gameplay (but only 15 of those in crossfire including last night) that is my very first hang.

If you are thinking, Jeff, that one hang doesn't prove anything, I quite agree with you. However, from now on, I'll set fans to fixed, 100%, which I can't hear with the headphones, and my guess is that temps will probably be down to 70 or less, and we'll see if I ever have another Sleeping Dogs hang. If I do, I'll report it.

Just MAYBE, the 7950 card does not like 75 degrees. LOL

Rich

 

Update on Sleeping Dogs - the card no longer likes the game. Period.

I just had two blue screens, the first one 10 minutes into the game, during a long in-game video.

I had neglected to add that the hang yesterday was at the start of a video. I didn't think it was relevant, but I see that there is a completely different load when the video is playing, reduced cpu load, and reduced gpu load. So the game has to switch gears.

Today the first blue screen came at the end of a longish two-minute video in the hospital with the gang leader. So I reduced my cpu overclock by just a bit.

That caused me to drop the fsb, but then on the way to the hospital, I noticed my cpu usage at about 100% on every core! So I got through the video, and then decided to increase back to 418 fsb.

I had the second blue screen in 10 mins. leaving the parking garage in a car.

All gpu temps had been ok, under 75 for sure, with fans at 100% and the two loud fans going.

I then played around with Heaven and furmark - ensured that the two cards were contributing in crossfire, then I grabbed a car and headed out again. It didn't blue screen, but the radio music became garbled, and the frames which had been at about 42 in the apartment, dropped quickly to about 13 - 6.7 - 12. Terrible frame rates!

What?!

Then I read that they had recently released a patch to the game, patch 1.5, which just came out on Aug 24th, but I usually play the game offline, and only the other day did I notice that it wanted to download something, so I went online to let it do that, thinking it was some dlc good stuff.

The patch was designed to keep the game from "running too fast" by controlling the cpu, and it also addressed issues with the in-game bink videos.

I unfortunately have had the patches applied automatically, and I am not sure how I would go about uninstalling that patch - maybe there is a way to do that.

So my first try is to install the latest catalyst, but I have also had the catalyst updating automatically, and maybe I already have the latest. I am turning off auto updates on steam games, and also on catalyst. Like Sam said recently, he doesn't like to update just because ... I agree. So perhaps the new patch really fowled things up.

I am tempted to disable crossfire, overclock the main card back up to 975, and see how driving and radio music are after I do that. It might just be a crossfire issue. Or maybe I have to find out how to uninstall that game patch.


EDIT - not just a crossfire issue. I disabled crossfire, and came back in-game in the apartment at about 26 fps, with one card at 975 overclock, just like how I used to play before implementing crossfire. Cpu usage was about 70% It was okay until I went outside and ran to the parking attendant - it hung - not a blue screen but game hang. So the game seems to be fowled up for 7950. I am uninstalling ccc and will put the new 12.8 in.

Rich

 

DDP, I hear you loud and clear and I am happy to down-size the gaming discussions.

But "No more talk about games" is so far-reaching.

I am pretty sure you wouldn't ever think of saying "No more talk about encoding" as this is a "How to build a PC thread." We build PCs in order to run software, right?

With all due respect, your post may have been in the works for some time, but timing-wise, it comes right at the point when I am actually no longer talking about a game, per se, Sleeping Dogs being the game, but I am talking about stability issues related to crossfire.

Look at my last 2 posts. Kevin and Steve complained, so I had already decided to minimize the gaming discussion and you no longer see any screenshots.

My last two posts have no screenshots, virtually nothing about the game that does not apply to temperatures, and stability, as it relates to my build, and my overclock, and the fact that I have two graphics cards in the new case. The last posts are all about system hangs, blue screens, crossfire, overclocking, driver revisions, and game patches.

Again, patches, drivers, all relates to hardware.

So maybe you could clarify.

I don't feel that you are you saying that stability issues with multiple gpus is not a relevant topic for "how to build a PC" And I am pretty sure you are not saying that this thread is only for people who are planning to build a non-gaming PC, because I am sure that we can agree that at least 50% of the people who plan to build a PC will be doing some gaming on that PC, right?

So then, may I suggest a slight re-wording of your post, such as:

Appropriate in this building thread would be only a bare minimal talk about gaming and playing games, pretty much only as that discussion relates to build issues, such as stability with crossfire, performance, cooling, compatiblity, SSD versus HDD, etc.

Rich

 

I have to agree there. This may not be the PC Gaming thread, but PC gaming is very performance-intensive and has a heavy sway on the components choices many builders make.

Agreed to keep gaming discussion itself to a minimum though.

 

Beautifully put Jeff. I agree 100%.

This post is about stability, and bottlenecking. It will reflect the new - minimal gaming talk - thread philosophy, and refer to gaming only in the context of performance of the computer build, which as Jeff just mentioned, is stressed by gaming almost more than by any other activity. I am excited by this post, because I think I really learned something new just now that will help me in the years to come.


-- UPDATE --

NO MORE GAME PATCH:
Windows 7 restore rolled back the game patch - I went back to the 4th of this month. It only gave me a few dates to go back to. Now steam shows "update required" next to the game title. And lucky me, I didn't lose my save games, which I had forgotten to back up before the restore. The result is that al least now the patch is out of the picture in terms of what might be wrong.

I didn't really think it would work, and I am impressed that it reversed the patch - very impressed. The reason I say that is because, with Steam handling things, as I thought about it more today, I imagined that Steam had just re-arranged things - and that Windows wouldn't pick up on the re-arranging - for example the Patch was not a restore event. Whereas, if I had kept Steam auto update turned off, and if I had manually downloaded and installed the patch, Windows would have created an Install Patch restore point. See what I mean? So I was expecting to get no results from the restore.

But to my surprise, it appears that the restore program is much more clever than I had thought - it must closely monitor the core program files, looking for significant changes, and backing everything up for that restore point. So yes, I am quite impressed to have reversed the patch. No patch. But also, no ability to set a 30 fps rate cap. Somehow that got removed by the restore. That fps rate cap could be useful - more detail about that below.

I INCREASED VDDC TO 1087 ON ALL CLOCKS, INCLUDING STOCK:
I increased vddc on the stock clocks - the opposite of what I did earlier today. The reset on the HIS IceQ card, per Trixx, reading the card bios I presume, is 981. Several other cards I tested, had a Trixx reset vddc of 1087, presumably from Trixx reading their bios. For certain, you can be assured that I didn't come up with that voltage number on my own. I have always been overclocking the cards, usually to 975, sometimes when some of the non-HIS cards were crashing or overheating, to just 900, and sometimes all the way down to stock 800, but I have always left the vddc of 1087, except for the XFX which was running so hot, I tried 981 to see if that would help and it didn't.

But on the HIS IceQ cards, the 975 overclock, a slight memory overclock of 1350 from 1250, +20% thermal headroom, and as I say the vddc of 1087, allowed me to sail through crysis, crysis 2, and warhead, without one crash, with just one 7950. So 1087 mv is a proven vddc, and 981 is not, not proven by me since I don't usually run stock clocks. So with these hangs, it only makes sense to increase the stock vddc to the proven 1087, no matter what Trixx says is the proper reset, and while I'm at it to also increase the +20% thermal headroom of course.

So just now, I sailed out of that parking garage, crossfire, both cards at stock clocks, but at increased vddc of 1087, up from stock vddc of 981, and while I noticed a bit of a hiccup or lag, at least there was no hang - implying better stability.

I INCREASED 9450 OVERCLOCK BACK TO PREVIOUSLY ROCK SOLID 418 FSB = 3.343 GHZ
Prior to rolling out of that garage just now, I had also gone back to my 418 fsb 3.343 ghz 9450 overclock, up from the 3.2 ghz that I had retreated down to when I got the first BSOD. (I still have no answers for this afternoon's two blue screens. The thought that comes to mind, is remembering when I read Sam's recent post, where he said that three different times the perplexing problem turned out to be the motherboard. That thought comes to mind, but I try to push that thought aside. Maybe, ultimately it is the mobo - but I'm not quite ready to finish the upgrade by throwing another $800 over on the mobo/cpu/memory side - ivy will have to wait for a bit.)

So, I went back to my heretofore totally stable max 9450 overclock, of 418 fsb, 3.343 ghz, up from stock 2.66. And I think it is soooo important that I did that, and here's why.

I believe I was hitting the cpu bottleneck wall.

Yes, cpu bottlenecking. That was what was causing the garbled music from the radio, and that was what was causing the absolutely wild drops in frame rate down to the low teens, as I barreled down the road in my flashy sports car earlier in the day, departing the garage.

I think I mentioned in an earlier post yesterday, that in a different area a mile away, in the business district on the way to the hospital, all of a sudden I had noticed all 4 cores at 100%. If there ever were a cpu bottleneck, that would be it, right? By definition, that would be it. Am I right? In the upper right corner OSD, I have the core load laid out on top of each other, at the end of the row that shows the common cpu frequency. I repeat the core frequency like that to get all four core loads next to each other.

So all of a sudden I glanced up and saw in that section, something like this in white font:

CPU0 3343.2 100%
CPU1 3342.8 100%
CPU2 3350.6 100%
CPU3 3342.2 100%

The little 100% repeated four times on top of each other, it amazed me and caught my eye - I have never seen that ever before.

The first time I saw that, this afternoon, as I say I was in the business district, heading to the hospital, rolling along very fast in the car, in the Victoria Peak Central area, with lots of skyscrapers and scenery, flying past me. So just imagine from a cpu load perspective - take the normal scenery, and now extend it all the way up vertically on the map, and add all the reflections on the glass of those buildings in full daylight, and add all the other cars, the people, the physics, and a lot of cars trying to avoid me.

So the 9450 was calculating like mad trying to keep up with two frisky graphics cards, which had no 30 fps rate cap and which were running 99% load each one. Remember - I had disabled the energy savings rate cap in order to achieve extra fluidity per Sam's remarks. And he was right - more fluid! I just had no idea the cpu would be so strained!

(The daylight seems to be a much higher graphics load, because just a bit ago, I went back to a save from yesterday, nighttime in Hong Kong, and I could not replicate the cpu load no matter how fast I drove.)

So all 4 cores showed 100% That was an unforgettable sight that I have never witnessed on an in-game OSD until today, having just recently created the crossfire configuration.

I never saw it in Crysis, nor in any of the other 2 Crysis sequels, running just one 7950. I haven't seen it in BF3 even with crossfire. I am used to cpu core loads in the 70s, and very highly stressed close in the upper 80s. But all 4 at 100% - never before in-game on the On Screen Display which is white font, in the upper right corner of the big 30" screen.

So that alarmed me, and caused me to reboot and raise my overclock back to where it had been, 3.343, about 4% higher. A very good thing that I did that - in this case, that 4% makes all the difference in the world.

What happened just now, the restore removing the patch, the vddc up to more usual 1087, the gpus not at stock but actually at 900 clock, and the cpu back up to the 3.343 ghz overclock, is that, while I raced down the road, only one core was 100%, the others showed 97, 98, 99. While it was not easy to drive really fast in order to stress the game with lots of changes happening quickly, and watch the OSD, and not have a major crash every second, I saw that only once each block did the frame rate drop below about 32 - just for a tiny brief bit might it plummet, but it immediately bounced back up - and the car radio never skipped a beat. Cores of 100, 99, 98, 99 are playable, while cores of 100, 100, 100, 100 produce garbled radio and wildly fluctuating frame rates.

THOSE CORES ARE KEEPING UP - BUT JUST BARELY! That's how close I am to a full cpu bottleneck!!

So, I wish I had not lost the 30 fps rate cap functionality. But I still don't trust the patch. I will try to work with the game as it is now, the cpu barely keeping up, and just live with whatever frame rate dips that might develop. The other possibility is to disable crossfire, which I know will reduce the cpu core loads down to about 80 to 90%, with maybe the first core at 100%. On one card, I was getting 26 average fps at 975 overclock up from 800, and that had worked for me.

Next year - hello ivy bridge.

Rich

 

It's near-impossible for a game patch to start inducing blue screens by fault of the game, although it may pick up on a system instability that already exists. To me, I don't think the game is at fault, but the stability of the machine in general, which seems obvious given the other issues you've had. To properly diagnose this, ditch both the CPU & GPU overclocks before proceeding further, to isolate them from the equation before you then look at a software or hardware problem.

 

You mean hello Haswell? Come 2013 that should be nearing release (estimated for Mar-Jun). It's only a relatively small upgrade (similar to Sandy Bridge vs. Nehalem) but will further add power savings and require a new socket, LGA1150.

 

Have to agree with Sam. You are overclocking two things at once and neither of them has had their stability fully confirmed. Drop both overclocks entirely to see if the problem repeats then apply one at a time. CPU first then video cards after the CPU is proven rock solid stable.

Intel sure love making new sockets don't they? Regardless of how good the CPUs actually are, one can run 4 full product generations of AMD CPUs on any AM2+ motherboard.

EDIT: That awkward situation when you just get done reformatting and have to wait for Winamp to index 150GB of lossless music...

 

Not any AM2+ motherboard, far from it. Personally I still maintain changing sockets is the best approach as it prevents the confusion of the CPU fitting but not being supported by the board/BIOS etc. It also allows for change, and beyond better clock speeds, AMD have really only had one significant change in their CPU architecture since 2005, which was Bulldozer (for better or for worse). Come Broadwell (14nm shrink of Haswell) even the PCH is supposedly going to disappear - the CPU will handle it all.

 

Changing sockets isn't a problem at all IMO, but maintaining socket compatibility would be helpful. For example, if a company releases a new socket that is almost identical to the previous, then there's nothing stopping them from making it compatible with the older CPUs except them wanting to not let people do it. Intel and/or AMD could have more stringent requirements on the motherboard makers IE requiring them to update all motherboards that have hardware compatibility with newer CPUs to also make compatible BIOSs. Heck, AMD/Intel could even make the CPUs be able to be supported by an older BIOS, just not supporting some newer features.

However, when a socket change includes an extreme change in the chipsets, it makes sense to not have compatible sockets, but that the chipsets are connected using stuff such as QPI, Hyper Transport, DMI, UMI, ect. should mean that even if an older chipset has some features that a newer CPU has built-in, they should be able to work together if Intel or AMD wanted to let them work together, again, simply meaning that the mixed features would disable the CPU's unsupported features and use the ones that are on the chipset instead.

For example, even if an Intel CPU is built that has the entire northbridge and south bridge setup built into the CPU die, it still has PCIe which is basically what DMI is and it should be able to be made compatible withthe older DMI chipsets by disabling the features that both the chipset and the CPU have that are on the CPU and using those in the chipset. Compatibility could be maintained if AMD and Intel cared to allow it. That it isn't always maintained is because they don't want it to be, not because it can't be done and is thus abandoned in the name of progress.

About AM2+... There are a few AM2+/AM3 DDR2+DDR3 boards that could theoretically be compatible with everything from AMD from AM2 all the way up to AM4 CPUs that have DDR3 controllers. Now that would some seriously expansive compatibility if BIOS support is maintained on them.

Also, to be fair, even Ivy Bridge and Core 2 are quite similar architecturally. The biggest differences between them are probably the cache/interconnect improvements and how much of the chipset has (especially memory controllers) has been integrated into the CPU.

 

Man, when Blaze and Sam start talking, I feel like I'm at a chip designer convention and can only just barely try to attempt to understand bits of the discussion. (Why don't you guys write a column for Anandtech - oh I forgot, Sam doesn't like those guys anymore. LOL)

This is going to be an extremely short post about stability. I hear you guys, Jeff and Sam, about backing off my overclocks. While it does make sense, I have the feeling it would cripple performance. But maybe not.

Jeff, you're going to laugh your butt off, but while I was posting last night and eating, etc. I had accidentally left the game running from when I successfully exited the garage and flew down the street without a crash, max cpu overclock at 3.343 ghz, cpu core loads just under max, at about 100, 99, 98, 99, something like that, with radio playing smoothly, and fps mostly stable at 32. The 9750 overclock was 900, 1310 memory, and 1087 mv vddc.

The part you're going to laugh at was - I had left the fans on auto, and I came back after about 3 hours to find the car just sitting there, with passersby strolling around, 32 fps, but temp on card number one showed 85.

Yes, 85. It apparently had been like that for hours. I wiped off my glasses to make sure it was 85, not 65. I didn't smell any smoke. For some inexplicable reason, that didn't bother me that much. Maybe you and Sam have finally cured me of my temperature panic.

I reasoned - "Well, nothing seems overly hot. The temps here in the trailer aren't that hot. Everything seems stable. Fan is on auto, at around 2950 rpm, whereas full speed is closer to 3200 rpm. So apparently HIS doesn't think this is an unreasonably high temp. Jeff mentioned occasionally hitting 78 - this is well above that - but ...."

So believe it or not, I didn't change a thing, and I continued in-game like that for two more hours of complete stability, temp at 85 the whole time, fps hitting low 40s in some sections, but never below 32. I exited, and the 40 mins of charting that I get from Afterburner on the bottom of the screen, was expanded, and yes, it showed a steady 85 degrees (applying windows 7 magnifying glass to verify 85, not 65.)

So then I ran some Furmark. It appears that the new Furmark at 2560x1600 is a quite close approximation to the load from Sleeping Dogs at the same 2560x1600 resolution, DX11, extreme settings, with the high res texture pack.

At graphic card settings of: 900 core clock, 1310 memory, vddc 1087, auto fan, I let Furmark run for about 10 mins, and it stayed steady at 85, then touched 86 for a couple seconds. I closed it. I noted that, still at 86, the fan tach on the hot card never broke the 3000 rpm barrier - HIS prefers to keep that fan whine hidden. The other card was running about 75, the fan at a silent 2400 rpm.

I set:
#1 fan to fixed - maximum,
#2 reduced core to 800 (reduced memory to 1250) This is zero overclock.
#3 and reduced vddc to 1012 (down from the 1087 but not all the way down to the 981.)

Those three changes dropped temps about 9 degrees - fan setting, core clock, and vddc - each one virtually equal in its effect.

I restarted furmark fast enough from high temps at one point, loading in the lowest Trixx profile, that I was able to catch the gpu at 78 before it fully cooled, and then over the next 5 minutes, it remained at 78, but then began touching 77 from time to time. So instead of coming up from bottom, I was coming down from high, which made me reason that it would have settled out at 77. Coming from the bottom, I watched it do the same at 76, then finally start hitting 77 a few times.

So, as I mentioned, for that set of ambient conditions, and the above settings, 77 degrees on the hottest card was a 9 degree reduction from settings of: auto fan, 900 core, and 1087 vddc.

I prefer the cooler settings, and I don't mind the fan whine which disappears once the headphones start playing the game sounds. Additionally, with the 7950s at a slower core clock, that should reduce the cpu load and eliminate the crazy fps drops and garbled radio effects from cpu bottlenecking. At the ratio of 800/900 gpu core clock, if I get minimum 28 fps rather than 32, I probably won't be able to tell the difference.

All of this will depend on ambient temps, of course, which at this very minute are near 86 whereas during last night's testing were closer to 80 - and furthermore the 1012 vddc has not been tested for stability.

For now I'll leave my cpu at 3.343 max overclock, and trust that the patch was the real villain. But if my cores drop to around 96% average load, maybe I could again slightly back off the overclock to 3.2 ghz, if any stability issues arise - any BSODs. I'll advise how things go.

Rich

 

http://forums.afterdawn.com/t.cfm/f-216/recommendation_required-941309/#5762255



This from now on:

http://bit.ly/Ov5uQx

I really don't know how - or why - I'm still replying to threads like that after 6 1/2 years.

 

That is the single most hilarious and unique link I have ever clicked XD

Rich, glad to see you're over the new hardware excitement or getting there. I certainly remember the fun I had with my first Crossfire setup. No 85 is not a dangerous temperature. If it worries you, you can always set up the fans to be more aggressive without having them at 100% all the time.

My fan profile looks like this:

http://imageshack.us/a/img28/1480/fanprofile.png

I can have 100% fan at load while having the cards stay relatively tame for noise at idle. Some noise doesn't bother me much as I already have plenty of white noise in my room.

 

Never used LMGTFY before? It's great