CrossFire Versus SLI Scaling: Does AMD's FX Actually Favor GeForce?

[Crashman]

[citation][nom]hapkido[/nom]This isn't a CPU test and the CPUs were intentionally put into different graphs. It was a test to see if cf7970s scale as well as sli680s on with an 8350. The i7 is there to rule out the possibility that cf doesn't scale well on that CPU. While I would like to see more data with more CPUs and more GPU combinations, the article shows evidence that cf scales worse than sli using the 8350.[/citation]Bravo. I'd be happy if everyone who's making these Intel-v-AMD arguments would simply read the title and the last sentence.

Core i7 performs better, maybe a little maybe a lot, I didn't really pay attention to the size of hte difference because that wasn't the point. The better-performing processor is just the "control group" and the lesser-performing processor is the "experimental group", to see how the CrossFire vs SLI scaling changed using the different processor architecture.

 

[jurassic1024]

Even before reading this article, which is great btw, I knew better than to buy AMD CPU's for a gaming machine. If you have to max out detail or overclock (CPU) to remove CPU and memory bottlenecks, then count me out. Not even an FX 8350 with an MRSP of $100 would change my mind - ever.

 

[Xerpadon Xerilious]

[citation][nom]HKILLER[/nom]An I7 3770K which is 150$ more expensive than FX 8350,Performs 10% better with single GPU and with Dual GPU 20% better.not to mention the 990FX uses PCIe 2.0 while Z77 uses PCIe 3.0 also Z77 is 60$ more expensive than the other one!shouldn't it have done atleast 50% better?the only time it really out shined FX was in Skyrim...also they should do another one with the military grade version of Sabertooth 990FX Gen3/R2.0 which has PCIe 3.0 i can say then the difference would be much less than now![/citation]

Exactly my thoughts, they are comparing 2 completely different classes this seems like a total setup its not more different then putting a light weight champ vs a heavy weight champion boxer in the ring.

Pci-e 2.0 vs Pci-e 3.0, also i believe the HT 3.0 is also Greatly bottlenecking i assume most people know this if not google for more information. anyways this article is simply unfair i'd bet if they redid it Withe the Sabertooth 990FX R2.0 Gen3 the results would be Stunningly opposite, from other reviews on the web they say the Sabertooth 990FX (pci-e 2.0) vs the Sabertooth 990FX Gen3 (pci-e 3.0) the performance varies from as low as 8% up to 20%.

 

[computertech82]

...and the results STILL show intel stomping the amd...AS USUAL. Zero surprise there.

 

[Crashman]

[citation][nom]Xerpadon Xerilious[/nom]Withe the Sabertooth 990FX R2.0 Gen3 the results would be Stunningly opposite, from other reviews on the web they say the Sabertooth 990FX (pci-e 2.0) vs the Sabertooth 990FX Gen3 (pci-e 3.0) the performance varies from as low as 8% up to 20%.[/citation]What reviews?

1.) It's AMD's top CPU on AMD's top chipset.
2.) AMD 990FX does NOT support PCIe 3.0
3.) Adding a PCIe 2.0 to 3.0 bridge does NOT give the 990FX PCIe controller extra bandwidth
4.) The PLX PCIe 3.0 Bridge chips add latency.
5.) The ONLY benefit of a bridge chip is distributing bandwidth to more slots (3-way or 4-way)
6.) The 990FX supports two x16 slot natively, so the distribution benefit doesn't apply to 2-way
7.) The bandwidth doubling effect of PCIe 3.0 has only proven beneficial on x4 and x8 slots
8.) The bandwidth doubling effect of PCIe 3.0 has proven useless to current cards on x16 slots
9.) The bandwidth doubling effect of PCIe 3.0 ENDS at the chipset, which is PCIe 2.0
10.) The new board has been out for two months, yet I haven't seen a review in the U.S.
11.) The most likely reason we don't see tons of U.S. reviews is that the PLX bridge REDUCES performance compared to "native" x16/x16 mode.

Any 8% to 20% gain would only apply to the SLOWER slots, which are those that borrow lanes from the primary slots. You can't speed up the primary slots because they're full x16, tied directly to the 990FX, and push 100% of the 990FX's bandwidth. HyperTransport overclocking could help, but that would be the subject for a separate HyperTransport overclocking article.

 

[razor512]

they should have added an AMD x6 1100t (it has faster single threaded performance than the any of the current

with all 8 threads being properly used, the 8350 can match a core i7, but no game will do it and very few professional aplications will go it also. overall, AMD messed up by moving from 6 full/ proper cores, to 8 crippled cores.

 

[hapkido]

they should have added an AMD x6 1100t (it has faster single threaded performance than the any of the current

with all 8 threads being properly used, the 8350 can match a core i7, but no game will do it and very few professional aplications will go it also. overall, AMD messed up by moving from 6 full/ proper cores, to 8 crippled cores.

This is a valid concern, but Piledriver is AMD's current architecture. Again, it's not a CPU comparison. It's a test that shows crossfire scales worse on AMD's current platform than it does Intel's current platform. We're not interested in the actual FPS or what CPU or even what GPU did better, we're interested in the difference.

(I'm going to make some numbers up for the sake of argument.) If CF and SLI both scale +80% adding a second card on Intel's platform, and SLI scales +80% on AMD's, but CF only scales +60% on AMD's, it gives evidence that CF scales poorly on AMD's platform. That's valuable information to self-builders.

I think as an enthusiast PC gamer your best bet (right now) is to go Intel CPUs and AMD GPUs, but if you bought an 8350 because you do a lot of 8-threaded stuff in addition to gaming and were planning on multi-GPU (using the savings on the CPU versus an i7), that's something you'd want to know. Building PCs isn't about brand loyalty, it's about spending your money wisely.

 

[utroz]

Tom's seriously needs to quit doing benchmarks and then giving the hardware away and then releasing the benchmarks after the drivers are old as heck in the GPU world.... Come on Tom's used to do everything to improve performance. Just lookup the old Super Socket 7 article when they got a K6-3+ 450@600MHZ. The reviewers are slacking lately....

 

[JackNaylorPE]

Until we start to see double digit amounts of games in your test suite, these reviews will remain irrelevant to many of us and the conclusions out of touch with the majority of other sites.

 

[Crashman]

Tom's seriously needs to quit doing benchmarks and then giving the hardware away and then releasing the benchmarks after the drivers are old as heck in the GPU world.... Come on Tom's used to do everything to improve performance. Just lookup the old Super Socket 7 article when they got a K6-3+ 450@600MHZ. The reviewers are slacking lately....

Yeh, those damn lazy reviewers who work only 12 hours a day six days a week, someone should fire the whole lot of 'em.

 

[englandr753]

Wow people, easy on the reviewers. The time it takes to gather this type of data should allow a little margin of error. If the latest drivers add 10% performance in a certain game for your setup, do the math.

Now I don't feel so bad not upgrading to a 7970 setup. My 3x 6950's may just be about right for my 8350 and Skyrim plays perfectly fine at my res. Every other game I've thrown at my rig feels smooth as butter as well so no complaints here. =D

 

[ramcoza]

Why couldn't you do this test with i5 3570K instead of i7 3770K?
If you did so, there is no need to put a quote like
"That is fair enough. The FX is a less expensive processor, so we're completely fine with it not performing as well."

Moreover there could have been a clear answer to which CPU scales better.

This is not a fair test. You should have tested the processors of same price range.

 

[belardo]

Goodbye Tomshardware: Been visiting this site for over 10 years. Have made over 16,000 posts here.
Seen some good, some bad... some stupid things here. Been going down hill for quite a bit lately and the latest change with their end-user blogging system pretty much ends it. Why bother? If you don't want people to use your blogs with responses, just rip it out all together. When the system fails more than it works, its bad. When you can no longer QUOTE people, its bad.

Other websites know how to do this: theregister.co.uk is among the very best. There is of course Anandtech and Dailytech that are better than the old and new system used at TomsHardware. For "the authority of tech", it sure doesn't seem like it.

Last month, when you (Toms) censored and blocked people from responding to the article in which AMD changed their logo and such - because every one of us was MOCKING AMD.... as silly logos do not make a CPU faster.... that was a first. I considered leaving the site then... I'm sure some have left.

But this new blogsoftware is the last straw. Its bad when you cannot posts or see posts. Its bad when what used to take 1-2 clicks now takes 10+ to do the same thing... its just not worth it. Not worth my time to read either.

I'm typing this HERE, because you sure as hell can't get most postings to work on the other articles. I've tried Chrome and Firefox... My Windows7, my iPad and my Android phone.

Thanks for the memories Tom (The original Tom) and many of the people who bothered to read and share their opinions on this website.

 

[smeezekitty]

Even before reading this article, which is great btw, I knew better than to buy AMD CPU's for a gaming machine. If you have to max out detail or overclock (CPU) to remove CPU and memory bottlenecks, then count me out. Not even an FX 8350 with an MRSP of $100 would change my mind - ever.

Goodbye Tomshardware: Been visiting this site for over 10 years. Have made over 16,000 posts here.
Seen some good, some bad... some stupid things here. Been going down hill for quite a bit lately and the latest change with their end-user blogging system pretty much ends it. Why bother? If you don't want people to use your blogs with responses, just rip it out all together. When the system fails more than it works, its bad. When you can no longer QUOTE people, its bad.

Other websites know how to do this: theregister.co.uk is among the very best. There is of course Anandtech and Dailytech that are better than the old and new system used at TomsHardware. For "the authority of tech", it sure doesn't seem like it.

Last month, when you (Toms) censored and blocked people from responding to the article in which AMD changed their logo and such - because every one of us was MOCKING AMD.... as silly logos do not make a CPU faster.... that was a first. I considered leaving the site then... I'm sure some have left.

But this new blogsoftware is the last straw. Its bad when you cannot posts or see posts. Its bad when what used to take 1-2 clicks now takes 10+ to do the same thing... its just not worth it. Not worth my time to read either.

I'm typing this HERE, because you sure as hell can't get most postings to work on the other articles. I've tried Chrome and Firefox... My Windows7, my iPad and my Android phone.

Thanks for the memories Tom (The original Tom) and many of the people who bothered to read and share their opinions on this website.

The new comments system really does SUCK.
It is very very buggy, puts posts needlessly out of order and sometimes up or down voting causes "error" with no other details.

You can quote people if you go to the corresponding forum post and it is less buggy but still not very good.

 

[belardo]

I'm posting here because 1 out of 10 posts actually *WORKS*... so I've wasted my time trying to use it.

Its a pain in the ass useless garbage in that you have to click MORE MESSAGES 10 times... and still its a confusing mess. Its the Windows8 version of Comments software.

I don't use IE... it doesnt work in Firefox or Chrome so I suspect its useless for Opera. On my mobile devices, there are no comments to read most of the time.

Again, whoever vetted this crap... screwed up royally. And censoring people is also a no-no.

 

[utroz]

Tom's seriously needs to quit doing benchmarks and then giving the hardware away and then releasing the benchmarks after the drivers are old as heck in the GPU world.... Come on Tom's used to do everything to improve performance. Just lookup the old Super Socket 7 article when they got a K6-3+ 450@600MHZ. The reviewers are slacking lately....

Yeh, those damn lazy reviewers who work only 12 hours a day six days a week, someone should fire the whole lot of 'em.

Thats not what I meant. I just want to get benchmarks with the drivers that are from within the last month or so. I do realize that benchmarking takes a lot of time and effort and was not trying to say you guy's are lazy. It just seems that tom's is doing more "cut and paste" articles lately thats more my point.

 

[Crashman]

[citation][nom]utroz[/nom]Yeh, those damn lazy reviewers who work only 12 hours a day six days a week, someone should fire the whole lot of 'em.[/quotemsg]Thats not what I meant. I just want to get benchmarks with the drivers that are from within the last month or so. I do realize that benchmarking takes a lot of time and effort and was not trying to say you guy's are lazy. It just seems that tom's is doing more "cut and paste" articles lately thats more my point.[/citation]Cut and paste? This is a follow-up to an article written at the end of January. It got delayed with many appologies. Two of the cards went away. And to make those two things as right as possible, the new Nvidia data came from drivers of the same age.

If you would like to see one of the reasons a few of the articles got backlogged, please read the last System Builder Marathon article. It's a laugh a minute!

 

[oblivionlord]

Intel is farrrrrrrr from reaching a point of diminishing returns.

The framerates displayed here are not minimum framerates therefore there is much to be improved. A 120hz monitor plays games best at a 120fps 'consistent' framerate without any dips. Since none of the modern games released today at 1920x1080 can achieve 120fps with even a Titan then I say there's quite alot to be improved on. Crysis 3 is the future which will reflect the same performance that next gen console ports to pc will display. It will only get more demanding at time goes on.

 

[JonnyDough]

[citation][nom]sarinaide[/nom]It is true my A10 5800K runs slightly faster with a 680 opposed to a 7970, needless to say I just threw in some 6850's made life a lot simpler and enjoyable, had much better CFX scaling out of the 6850's than I did 7770's and 7850's, ol faithfuls.[/citation]

Drivers. Give 7000 series a bit of catch up time and they will likely do better with scaling.

 

[bgunner]

To me it is not all that surprising that Nvidia was able to work better with AMD's CPU as Nivida has had some decent success with scaling in the past. It is obvious though that it is time for AMD to have both sections work a little closer when it comes to GPU - CPU performance.

For those like me that need to build budget machines AMD IMHO has the market share.

 

[veyron1001]

This is an invalid article. Not only are the Nvidia cards overclocked a lot more, the intel processor is as well.

 

[Mac_McMan]

[citation][nom]veyron1001[/nom]This is an invalid article. Not only are the Nvidia cards overclocked a lot more, the intel processor is as well.[/citation]That is an invalid comment. Both processors were overclocked to the same frequency, and you've missed the entire point anyway. There was no CPU comparison intended. There was no GPU comparison intended. Like Crashman said, read the title, then the last sentence, and shut up.

 

[c911darkwolf]

Did you also know that the new Corvette runs faster then the Ford Focus?

Come on... AMD's spot in the industry has always been selling good products at a decent price.

You want preformance?
Intel I7 3770 IVy $329

You want affordable?
AMD FX8350 $199

Also your comparing a FASTER processor vs a slower one. You should scale back the intel one and compare for different results.

AMD sells to the common person in ranges we can afford. Intel has always been too pricey and Bleeding Edge. Don't compare Corvette's to Geo Metro's and tell me Corvette's are faster, because the point is mute there not in the same market.

 

[hapkido]

Sigh... Do you guys still not understand this is NOT an AMD vs Intel CPU review and NOT an AMD vs Nvidia GPU review?

The actual performance numbers are irrelevant and there are plenty of reviews and benchmarks that cover what's better. Look at percentage difference in how Crossfire scales on AMD vs how SLI scales on AMD and how Crossfire and SLI scale on Intel. If you just tested CF and SLI on AMD without also testing on Intel, the conclusion would be that CF simply scales poorly. By also testing on Intel, you know it's not an issue with CF because it scales as well as SLI on that platform. By comparing SLI scaling on both platforms, you see that's also similar, so you know it's not an issue with using two GPUs on AMD.

The only outlier is CF on AMD. THAT is what they were testing. Further testing with more hardware combinations and more games would be needed to draw a conclusion, but this test points to CF scaling poorly on AMD.

 

[sammiej230oo]

I'm looking for something to do gaming with my 7xxx series, some video encoding etc..

http://www.amazon.com/dp/B007SZ0E1K/?tag=b0009-20

Will this suffice for the next two years or so til' a new batch is out?