Are multi-gpu setups as bad as they used to be?

[LtMatt]

I don't know why people go on about the power profile, my cards run at 1200/1800 and the power limit is set to default and I still don't get any throttling, only time I have seen it was using furmark.

The problem is the Boost clock speed and the stupidly high Boost voltage which is applied is way above what the previous levels were. Stock 7950 voltage is around 1031 mV. Sometimes a bit higher, sometimes a bit lower. Boost voltage on a 7950 is 1.25v. That is some voltage jump for a core clock speed bump of 75mhz. That's even higher than a 7970 ghz which is around 1.218v or something like that. Remember typical stock voltage for a 7970 is 1093v up to 1175mv.

If you don't have boost cards, you will never have this problem.

 

[Drangueos]

I've been using crossfire for more then a year now and eyefinity 3 monitors.

Never had any stuttering or something like that.

I was using 2 HIS reference 7950 for 1 year, the only problem I had with them was when the ULPS was on, when the 2nd card was kicking in the system would crash sometimes, with UPLS off = no problems at all.

Now I have 2 MSI 7950 TF3 for 2 months and no problems at all (didn't check about the ULPS because I leave it always off for overclock)

*Power limit always at +20%

 

[Gregster]

@ the OP, I would go for another 680 if you can get one cheap enough. As you have a 680 already, this would be a sensible way to go (price for the other 680 dependant of course).

 

[oldestgregg]

I think RadeonPro fixes something else other than the reported issues with AMD CF but to be honest I'm not an expert on it as I don't personally suffer from any issues and... slightly tangentially now: they're massively over exaggerated (not generally on these forums but on others)

Its defo latency Rusty. Flip que size to 1 and its a whole lot better but,yes,you are right the problems were massively overplayed in the first place wernt they.

 

[Rusty0611]

Its defo latency Rusty. Flip que size to 1 and its a whole lot better but,yes,you are right the problems were massively overplayed in the first place wernt they.

RadeonPro doesn't solve runt/dropped frames.

 

[LtMatt]

RadeonPro doesn't solve runt/dropped frames.

Didn't pcper grudgingly admit that vsync did for the most part though?

 

[Rusty0611]

I think they were referring to the fact that vsync solves the effect to some degree but not the underlying problem.

 

[Davepen]

I use 680 sli running at 2560x1440 and it's really nice.

One card just wasn't quite enough at that res, and aside from new games not always being supported right away, and remembering to reenable sli each time I install a driver update, I'm really pleased with it.

Have used crossfire before and sli is 100% better

 

[humbug]

The boost throttling thing does not seem to happen on AMD's reference GPU's.

The problem seems to sit with none reference GPU's, or with their BIOS, mine is a classic example, the volts set in its BIOS is some 1.25v, which is very high and far to high, because of that its in and out of its TDP range which is causing it to clock up and down erratically.

I simply down volted it to a reasonable level, 1.19v, which in its self has fix the throttling at stock, for overclocking I simply added 10% to the board power line.
It runs as solid as a rock at 1200 / 1600 (up from 975 / 1500)

They have since released a new BIOS with much less aggressive volts, which has fixed the problem.

Another thing to remember is that with some Boost edition GPU's overclocking applications have what is seemingly unlocked voltage control when infact setting any volts does nothing, so you may think your setting higher or lower volts, but is not actually doing anything, so if you have a none reference Boost GPU / BIOS GPU the problem may persist because undervolting is not actually taking effect.

Its not AMD's fault unless you want to twist it into making it AMD's fault, if AMD learn about whats going on with the mistakes 3'rd parties are making i'm pretty sure AMD will tell them where they are going wrong and tell them to fix it.

With that none reference Boost edition GPU's with this problem have been getting BIOS updates to fix it, its not a problem anymore, though there might be some floating around with an older BIOS, like mine was, in which case just use the app that came with the GPU to lower the volts, or update the BIOS, job done.

BTW, Canucks picked on Nvidia GTX 600 GPU's with exactly the same problem, this is or was a GPU wide issue regardless of colour, don't make it just about AMD.

 

[pgi947]

I'm not a massive fan of boost clocks tbh, give us the classic 3 clock speed steps, its all we need really. I'm not sure what purpose it serves vs plain old 3d static load clocks.

 

[LtMatt]

The boost throttling thing does not seem to happen on AMD's reference GPU's.

The problem seems to sit with none reference GPU's, or with their BIOS, mine is a classic example, the volts set in its BIOS is some 1.25v, which is very high and far to high, because of that its in and out of its TDP range which is causing it to clock up and down erratically.

I simply down volted it to a reasonable level, 1.19v, which in its self has fix the throttling at stock, for overclocking I simply added 10% to the board power line.
It runs as solid as a rock at 1200 / 1600 (up from 975 / 1500)

They have since released a new BIOS with much less aggressive volts, which has fixed the problem.

Another thing to remember is that with some Boost edition GPU's overclocking applications have what is seemingly unlocked voltage control when infact setting any volts does nothing, so you may think your setting higher or lower volts, but is not actually doing anything, so if you have a none reference Boost GPU / BIOS GPU the problem may persist because undervolting is not actually taking effect.

Its not AMD's fault unless you want to twist it into making it AMD's fault, if AMD learn about whats going on with the mistakes 3'rd parties are making i'm pretty sure AMD will tell them where they are going wrong and tell them to fix it.

With that none reference Boost edition GPU's with this problem have been getting BIOS updates to fix it, its not a problem anymore, though there might be some floating around with an older BIOS, like mine was, in which case just use the app that came with the GPU to lower the volts, or update the BIOS, job done.

BTW, Canucks picked on Nvidia GTX 600 GPU's with exactly the same problem, this is or was a GPU wide issue regardless of colour, don't make it just about AMD.

Good info and as i expected AIB bios is the problem here.

 

[tommybhoy]

In answer to the op, other than having to roll up your sleeves from time to time, it's fine-but it's never going to be as simple as a single core card.

It's AMD's fault because they let this happen on their cards. "Boost" is a feature of the 7970 GHz cards, they created the feature, they created the problem. Yes AIB's may of exacerbated the issue, but nevertheless it's a great disappointment on AMDs side to not of thought the feature through.

Good info and as i expected AIB bios is the problem here.

^
This.

It isn't AMD's 'fault' per say, yes they designed Boost and GHz bios but my 7970 OC didn't suffer spiking on GHz bios, the TFIII Boost edition, doesn't have the problem either, if the fault was AMD's, every GHz/Boost edition would suffer and it would have been picked up and widely publicised.

The first batch of His Ice-Q boosts don't suffer from the problem(there has been a change in the Ice-Q's somewhere as the first batch can flash non boost Ice-Q bios-the newer batches can't), so His have changed something, not AMD and Asus definitely changed a lot over the reference 7970.

As stated by Humbug, Nvidia has had boosting problems too with some cards causing stutter(some still do), unfortunately it's one of these things which like yourself has resulted in a swap over to the other team.

Most won't know it's happening unless they sit with an OSD displaying clock speed to check for sure as I only found out with a comparison with Matt threw up the wrong numbers, otherwise I wouldn't have noticed, but that will be partly down to using CrossFire, I might have noticed sooner if I was just using a single gpu.

I can appreciate the bad taste in your mouth Rossi though as it shouldn't happen in an ideal world, but with the nature of PC's being the way it is, there you go.

The fix I posted had the warnings due to the fact, I can't hold every ones hands and do it for them and not willing to take the stick for PEBCAK when they do something wrong and muck up the OS or whatever.

It DOES eradicate the spiking btw.

 

[Final8y]

Quadfire 5970 and still smiling after 3 years, but i am a Vsync user.

 

[Rusty0611]

Well my second crossfire card is a boost card (MSI TFIII BE) and I don't get these problems. It just seems massively presumptious to declare that THIS was Rossi's issue without really investigating it and proxy solving...

And anyway, even taking it as true - which I don't necessarily agree - it doesn't resolve the point that at the least games should be in a form of playability out the box. If you're going to say that by going CF/SLI you're sacrificing playability out of the box then I would strongly disagree.

You can't just rule out the drivers being the issue based on a couple of loose assumptions. I'm not ruling it out as being the underlying cause, I'm ruling out the assumption that this would definitely resolve it which is the undertones I'm reading here and I am disagreeing that it's an acceptable solution to the problem should it indeed be the problem in the first place. Confusing paragraph? Yes .

 

[Rroff]

Didn't pcper grudgingly admit that vsync did for the most part though?

VSync or artificially capping the framerate at a low level solves it aslong as your maintaining that framerate but I don't see that as a proper solution tho - not everyone wants to use vsync or cap their game to 40-50fps.

 

[Gregster]

It isn't AMD's 'fault' per say, yes they designed Boost and GHz bios but my 7970 OC didn't suffer spiking on GHz bios, the TFIII Boost edition, doesn't have the problem either, if the fault was AMD's, every GHz/Boost edition would suffer and it would have been picked up and widely publicised.

The first batch of His Ice-Q boosts don't suffer from the problem(there has been a change in the Ice-Q's somewhere as the first batch can flash non boost Ice-Q bios-the newer batches can't), so His have changed something, not AMD and Asus definitely changed a lot over the reference 7970.

As stated by Humbug, Nvidia has had boosting problems too with some cards causing stutter(some still do), unfortunately it's one of these things which like yourself has resulted in a swap over to the other team.

Most won't know it's happening unless they sit with an OSD displaying clock speed to check for sure as I only found out with a comparison with Matt threw up the wrong numbers, otherwise I wouldn't have noticed, but that will be partly down to using CrossFire, I might have noticed sooner if I was just using a single gpu.

I can appreciate the bad taste in your mouth Rossi though as it shouldn't happen in an ideal world, but with the nature of PC's being the way it is, there you go.

The fix I posted had the warnings due to the fact, I can't hold every ones hands and do it for them and not willing to take the stick for PEBCAK when they do something wrong and muck up the OS or whatever.

It DOES eradicate the spiking btw.

A little condescending.

Rossi had no end of trouble and tried many a solution to no avail. Why now after he got a 690 and has no more issues is it PEBKAC? It is easy to say "That would have fixed the issue" but you can't guarantee that.

 

[LtMatt]

Well my second crossfire card is a boost card (MSI TFIII BE) and I don't get these problems. It just seems massively presumptious to declare that THIS was Rossi's issue without really investigating it and proxy solving...

And anyway, even taking it as true - which I don't necessarily agree - it doesn't resolve the point that at the least games should be in a form of playability out the box. If you're going to say that by going CF/SLI you're sacrificing playability out of the box then I would strongly disagree.

You can't just rule out the drivers being the issue based on a couple of loose assumptions. I'm not ruling it out as being the underlying cause, I'm ruling out the assumption that this would definitely resolve it which is the undertones I'm reading here and I am disagreeing that it's an acceptable solution to the problem should it indeed be the problem in the first place. Confusing paragraph? Yes .

Well no one knows for sure but i see the same behaviour and problems Rossi saw with this 7950 boost card and this is at stock. As before im fairly confident this was the issue, as he had a boost card but who knows for sure. At the end of the day, who cares.

In other news i just tried out a bit of poor mans crossfire x16 x4. I was disappointed to find zero micro stutter as well. Pcper had me expecting something terrible, but it was not to be the case. What i did find was very high fps and fluid gameplay, in Battlefield 3 and Tomb Raider at least. Didn't try any others but will over the coming days.

 

[humbug]

Well no one knows for sure but i see the same behaviour and problems Rossi saw with this 7950 boost card and this is at stock. As before im fairly confident this was the issue, as he had a boost card but who knows for sure. At the end of the day, who cares.

In other news i just tried out a bit of poor mans crossfire x16 x4. I was disappointed to find zero micro stutter as well. Pcper had me expecting something terrible, but it was not to be the case. What i did find was very high fps and fluid gameplay, in Battlefield 3 and Tomb Raider at least. Didn't try any others but will over the coming days.

Matt you should get a Latency xFire thread going.

 

[Kaapstad]

A little condescending.

Rossi had no end of trouble and tried many a solution to no avail. Why now after he got a 690 and has no more issues is it PEBKAC? It is easy to say "That would have fixed the issue" but you can't guarantee that.

+1

The GTX 690 is a fantastic card and a far better solution than anything AMD has to offer including the HD 7990. The one thing a GTX 690 does very well is it works straight out of the box, for someone who wants to game this is the most important thing.

When I have dared recommend a GTX 690 in the past, it has attracted a lot of negative comments. The comical thing about it is most of those same people have never been near a GTX 690.

All multi GPU solutions should be like the GTX 690, you take them out the box, plug them in, load drivers and game.

 

[Gregster]

+1

The GTX 690 is a fantastic card and a far better solution than anything AMD has to offer including the HD 7990. The one thing a GTX 690 does very well is it works straight out of the box, for someone who wants to game this is the most important thing.

When I have dared recommend a GTX 690 in the past, it has attracted a lot of negative comments. The comical thing about it is most of those same people have never been near a GTX 690.

All multi GPU solutions should be like the GTX 690, you take them out the box, plug them in, load drivers and game.

From someone who has dabbled in lots of different AMD and Nvidia setups, I feel you have more of a say than the likes of me, Tommy, Matt, Humbug etc. My only experience is with Nvidia in SLI and what I have read on the subject.