Mixfire vs Regular Crossfire

[IllogicalGlory]

So when I bought my GPUs I wasn't very smart about it. I ended up with a standard Radeon 7850 working with a 7870 GHz edition I didn't realize that the weaker one would cause a bottleneck. I OCd my 7850 enough to get 100%/90% GPU usage in benchmarks, but in games I find it's more like 100%/80%.

My question is: would be better if I sold my 7870 GHz on Ebay and bought another 7850? I could possibly make a bit of my money back. Doing the opposite is not an option since I can't fit two of those XFX 7870s in my case.

It's kind of pain in the a*s, but I'm thinking it would be probably be a smart idea.

I could be wrong though; what do you think?

 

[nurturedhate]

It doesn't have to be an xfx brand 7870 to work.

 

[tential]

Wow, I must be very behind. I didn't even know mixfire worked lol.

 

[Termie]

It's not like you're going to get better performance with two 7850s, and given the transaction costs on ebay, I'm going to bet you lose money going from a 7870 to a 7850. Don't bother.

Here's a better plan - sell the 7850 and buy a Sapphire HD7870. It's a lot shorter than any 7870, and most 7850s.

Link: http://www.amazon.com/Sapphire-DL-D...id=1372804581&sr=1-1&keywords=sapphire+hd7870

 

[BigChickenJim]

I'm not sure mixfire always causes a bottleneck. My CFX setup (dual 7870s) involves two identical cards for aesthetic reasons, but during my research I came across several articles/posts claiming that there is no down scaling when the cards are in the same performance series (i.e. 78xx, 79xx, etc.). The downscaling issues were supposedly eliminated by modern Crossfire tech. In other words, your setup will perform better than a dual 7850 config by a wide margin, but not quite as well as a dual 7870 config.

So I guess my answer is no, you shouldn't sell your 7870.

 

[Termie]

I'm not sure it always causes a bottleneck. My CFX setup (dual 7870s) involves two identical cards for aesthetic reasons, but during my research I came across several articles/posts claiming that there is no down scaling when the cards are in the same performance series (i.e. 78xx, 79xx, etc.). In other words, your setup will perform better than a dual 7850 config by a wide margin, but not quite as well as a dual 7870 config.

....

I'd like to see a link for those claims. I'm pretty confident that's entirely wrong.

Based on the OP's post, he's not getting anywhere near 7870CFX performance. The 7870 is 20% faster than the 7850 at reference speeds. The GPU usage delta of 80% on the 7870 and 100% on the 7850 clearly indicates he's operating at 7850CFX levels of performance.

 

[SickBeast]

I had a horrible experience with this exact setup with lots of microstutter.

 

[bystander36]

When pairing up two different cards in crossfire, they just render the frames given to them at the speeds their cards are clocked at. There is nothing special that happens until the new crossfire drivers are released.

The problem with having two cards that operate at different speeds is that they will naturally find themselves getting out of sync and cause runt frames.

Once the frame pacing patch goes in, it will then end up causing the faster card to hold back frames to deliver them similarly fast as the slower cards pace, as the drivers should be forcing them to alternate frames fairly evenly spaced a part.

 

[BigChickenJim]

I'd like to see a link for those claims. I'm pretty confident that's entirely wrong.

Based on the OP's post, he's not getting anywhere near 7870CFX performance. The 7870 is 20% faster than the 7850 at reference speeds. The GPU usage delta of 80% on the 7870 and 100% on the 7850 clearly indicates he's operating at 7850CFX levels of performance.

I've already said that my information comes from multiple sources that I ran across while doing my own research. Nobody seems to have a definitive answer or article to point to, and I'm not going to post links to 30 different threads here. Also, I never held those claims out to be the absolute truth. I simply relayed the information that I've seen on the subject. You're free to use Google if you'd like to make your own judgement about the veracity of the claim.

It seems to me that the best way to figure out if the 7870 is having its clocks throttled is to simply have the OP run a benchmark and compare its operating clock speeds in CFX with the 7850 to its reference clock speeds as a single card. GPU-Z would be the best program that I know of to do that with.

Edit: Bystander kindly provided an excellent summation of the prevailing wisdom when it comes to mixfire and its drawbacks. I hadn't stopped to consider the effect of this month's frame-pacing driver, though... That's a good point.

 

[Termie]

I've already said that my information comes from multiple sources that I ran across while doing my own research. Nobody seems to have a definitive answer, and I'm not going to post links to 30 different threads here. You're free to use Google if you'd like to make your own judgement about the veracity of the claim.

It seems to me that the best way to figure out if the 7870 is having its clocks throttled is to simply have the OP run a benchmark and compare its operating clock speeds in CFX with the 7850 to its reference clock speeds as a single card. GPU-Z would be the best program that I know of to do that.

Edit: Bystander kindly provided an excellent summation of the prevailing wisdom when it comes to mixfire and its drawbacks. I hadn't stopped to consider the effect of this month's frame-pacing driver, though... That's a good point.

I already did a search. In fact, I've researched this many times over the past 3 or 4 years. I've never found anything that supports your theory. I asked because I was curious if new evidence had come to light.

And just to clarify, you might have a slight misunderstanding of how Crossfire works. The clocks on the 7870 will not throttle. They will always operate at full speed, but the GPU usage will be below 100%, just as if there were a CPU bottleneck. The OP has indicated the GPU usage is at 80%. He already has his answer.

 

[bystander36]

I already did a search. In fact, I've researched this many times over the past 3 or 4 years. I've never found anything that supports your theory. I asked because I was curious if new evidence had come to light.

And just to clarify, you might have a slight misunderstanding of how Crossfire works. The clocks on the 7870 will not throttle. They will always operate at full speed, but the GPU usage will be below 100%, just as if there were a CPU bottleneck. The OP has indicated the GPU usage is at 80%. He already has his answer.

As you have seen with all the FCAT data, there is nothing that speeds up or slows down either card in Crossfire. They are constantly delivering frames at different rates until you put in something that limits the speed of the GPU's, such as a FPS cap, Vsync or a bottlenecked CPU.

So if there is something limiting the frame production, you'd be right, but FCAT also shows quite clearly, that if there is nothing limiting frame output, both cards do not perform at the same rate, even when they are the same card.

The reason you see 80% usage is due to some form of bottleneck.

Here was an interesting article talking about Crossfire from one of the driver dev's. I'm not sure what happened to the actual quote, as it seems to have been edited to only include the overview: http://www.anandtech.com/show/6857/amd-stuttering-issue

He went on to say that AMD has never considered anything but having their cards to deliver frames as fast as possible to limit latency.

AMD’s current position on micro-stuttering is that they are favoring latency (and not frame intervals) above all else. By keeping their latency low and as even as possible the resulting input lag from multi-GPU setups is reduced, making the experience more responsive for the user. This is a position that’s essentially in alignment with how they’re handling single-GPU stuttering too, but in the single-GPU world there isn’t a deliberate frame pacing aspect to take into consideration since they merely need to render frames as fast as they receive them.

It wasn't talking about different GPU's, but it wouldn't make any sense to attempt to frame pace on mixed GPU's, and not on identical GPU's.

 

[tential]

I already did a search. In fact, I've researched this many times over the past 3 or 4 years. I've never found anything that supports your theory. I asked because I was curious if new evidence had come to light.

And just to clarify, you might have a slight misunderstanding of how Crossfire works. The clocks on the 7870 will not throttle. They will always operate at full speed, but the GPU usage will be below 100%, just as if there were a CPU bottleneck. The OP has indicated the GPU usage is at 80%. He already has his answer.

This was the original allure of Crossfire. When it first came out, they had advertised Mixfire saying that "No longer do you have to give up your old card. You can add a new card for more performance!" Then we just never saw it, and no one really seemed to care too much. Now though, I kind of want to see a review now that it apparently works again.

 

[bystander36]

This was the original allure of Crossfire. When it first came out, they had advertised Mixfire saying that "No longer do you have to give up your old card. You can add a new card for more performance!" Then we just never saw it, and no one really seemed to care too much. Now though, I kind of want to see a review now that it apparently works again.

I know it worked with the 6950 and 6970. At least I had tested it out with one of the cards BIOS flashed to a 6970, and the other was a 6950. They used their given clocks and just did their thing. Of course I went on to unlock both cards as well.

 

[BigChickenJim]

I already did a search. In fact, I've researched this many times over the past 3 or 4 years. I've never found anything that supports your theory. I asked because I was curious if new evidence had come to light.

And just to clarify, you might have a slight misunderstanding of how Crossfire works. The clocks on the 7870 will not throttle. They will always operate at full speed, but the GPU usage will be below 100%, just as if there were a CPU bottleneck. The OP has indicated the GPU usage is at 80%. He already has his answer.

Again, this is not my theory.

And I suspect that if the situation was as simple as "the faster card limits its usage to a percentage equivalent of the slower card" we'd have seen that information a long time ago. I'm not going to pretend to be an expert in modern CFX architecture and design, but I don't think the data that the OP provided is enough to make a broad judgement about mixfire downscaling.

 

[bystander36]

I'm sure there are times when the pipeline and CPU usage will cause some natural down scaling, but it is not something built in, and it will depend on the game engine and CPU capabilities. It is not a designed feature, just a side effect of how some engines react (Edit: all game engines I tested to some degree).

 

[tential]

Why does this have to be a theory? I'm sure at least 100 people on here have 3-4 generations of AMD products sitting somewhere around. This is an enthusiast forum after all. Shouldn't be too hard to test some mixfire and figure it out lol. This is my first Desktop GPU in 8 years so I don't have any but I bet a lot of people do. People love benching on here for fun, it shouldn't be hard to figure it out and get some real life data.

 

[bystander36]

I was thinking about reloading them. I had tested it before, though I did not take any screen shots to see usage, but I had tested for most all this stuff before. I was considering tossing my 6950's in with one BIOS flashed to a 6970 and see how it reacts, but I'm fixing dinner first.

 

[Sohaltang]

Wonder what would happen if I tossed in my 5770 beside my 7970?

 

[bystander36]

It won't work. The cards must be based on the same chip, which normally means the first two numbers of the cards must be the same. 5970 (has dual 5870's), 5870, 5850, and 5830's can crossfire. 7970, 7950 and 7990's can crossfire. I have a feeling the 7870 XT can also, but I don't know on that one for sure.

 

[bystander36]

Anyways, I loaded up my two 6950's, one with the 6970 bios and clocked higher.

What I found was they operated at their original clocks, as most of us agreed on.

As far as GPU usage, that was a bit odd. The differences of the GPU usage bounced around a lot. They differences in clocks varied from 15% down to 8% depending on what game or benchmark I was playing.

This seems to go along with what I was saying, and what Termie had said.

There is no hard code for it to throttle, but due to CPU limitations and or game engine limitations, the faster card will get slowed down.

I also have noticed AMD's usage is very wonky. In some tests that were definitely GPU bound, like Metro 2033, it was showing usage in the 20% and I had similar results in other games. Games which on 680's show near 99%. There is no way the 6950/70's were being held back by the CPU, yet their usage was showing very low in a few different games.

 

[Termie]

Anyways, I loaded up my two 6950's, one with the 6970 bios and clocked higher.

What I found was they operated at their original clocks, as most of us agreed on.

As far as GPU usage, that was a bit odd. The differences of the GPU usage bounced around a lot. They differences in clocks varied from 15% down to 8% depending on what game or benchmark I was playing.

This seems to go along with what I was saying, and what Termie had said.

There is no hard code for it to throttle, but due to CPU limitations and or game engine limitations, the faster card will get slowed down.

I also have noticed AMD's usage is very wonky. In some tests that were definitely GPU bound, like Metro 2033, it was showing usage in the 20% and I had similar results in other games. Games which on 680's show near 99%. There is no way the 6950/70's were being held back by the CPU, yet their usage was showing very low in a few different games.

Thanks for doing this. If you are able to, it would be interesting to do a few benchmarks in Metro2033 that use both 6950s at stock clocks, and then one on the 6970 bios, and see if the mixed duo results in a higher benchmarked FPS (setting aside frame pacing issues for now). That would at least answer the question of whether mixed setups perform similarly to matched slower cards, or somewhere in between.

 

[bystander36]

I took them out, but I had run the Metro 2033 benchmark. It was not giving very good readings. It was showing usage from 20% to 50%, yet with 680's, I'm getting from 97-99%.

There may be something wrong with the way the AMD cards register their usage. They were performing as they should, but the percentages were way low for what they should have been. Only the Unigine benchmarks showed usage like you'd expect. Well, closer to what you'd expect (80-97%), even there they never hit 100%.

Although before I rebooted after installing the CAP profiles, they showed usage more normally. I wonder if the CAP messed things up on usage results.

 

[bradly1101]

This may be totally ignorant to say, but what about underclocking the 70, or giving it a 50's bios? Would that help?

 

[UaVaj]

I had a horrible experience with this exact setup with lots of microstutter.

as piss poor as crossfire is. can only imagine mixfire.

however if it does works for you - definitely more power to you.

 

[bystander36]

This may be totally ignorant to say, but what about underclocking the 70, or giving it a 50's bios? Would that help?

Probably.