FX chips better than i5 for video editing?

[AtenRa]

http://www.guru3d.com/articles_pages/amd_a10_7850k_apu_review,12.html

No 7700K, but unless it's a problem specific to the 7700K, that looks fine and dandy, and 3.6 v. 3.4 would be about right.

guru3d used the Gigabyte G1-sniper A88X-CF in that review, not the Asrock.

 

[Cerb]

guru3d used the Gigabyte G1-sniper A88X-CF in that review, not the Asrock.

Got my tabs mixed up.

http://www.tomshardware.com/reviews/a10-7850k-a8-7600-kaveri,3725-10.html

There's another, definitely w/ the ASRock. Scores look right compared to Richland.

 

[Abwx]

Whatever else you want to say about the i3, there is no way for AMD to catch the faster ones like the 4340, 4360 and 4370 in ST performance, not even an OCed FX-9590. The deficit won't be made up completely, not even close. I don't even really want to post benchmarks proving this because fans gonna believe what fans gonna believe, but here is one:

http://www.anandtech.com/bench/CPU/1030

Sysmark is basicaly an Intel Bench, ask Bapco, nicely choosen, anyway.

And another:

http://www.anandtech.com/bench/CPU/344

You know that CB, although good to compare CPUs within a same brand, favour Intel, benches made by Anand point to a 18% advantage, 10% advantage with CB R10 and miraculously 28% with the same CPUs once CB 11.5 was realeased with ICC compilations and Intel Threads Profiler and MKLs.

And another:

http://www.anandtech.com/bench/CPU/1025

This one is total BS, it run on X87 on AMD CPUs and up to SSE3 in Intel ones, really the most biased and deseperate possible bench, that is, you have to litteraly put AMD s CPUs in ISA disadvantage to prove a point.

I guess they could ALL be total BS... I'll leave it to the gentle readers to decide.

And they are all total BS, not that Intel doesnt have the ST advantage but for the record Anand used 7zip to measure integer IPC, notice that both CB and 3Dparticle are FP, wich are the most pointeless benches in respect of average users but perhaps Povray wich is FP should be taken into account since you seems interested by the thing :

And 7zip for Integer, as used by Anand to test servers IPCs :

Do the maths in respect of frequencies...

 

[crashtech]

I'm not disagreeing with you, Intel certainly does better at single threaded performance. But what does being able to score a 2.05 in Cinemark single threaded vs. 1.21 mean in real life practical terms? In this age of sub 15 watt CPU's running Windows, what does this mean to an average user? Would they be able to tell the difference between an i3, i5, even an i7 and an FX (assuming they all have an SSD)?

Are all those benches listed that show single threaded performance actually being artificially limited to one thread? (honest question, I know Cinebench is, what about the others) If so, how does that represent real applications, that in this day and age tend to be multithreaded if they need to be, or single threaded if they don't need to be as any modern processor is going to be fast enough?

I'm not suggesting the FX is better, I want to be clear on that. But I don't think it is very much worse if not neck and neck with pricier options from Intel in an actual use scenarios, not just watching SuperPi numbers.

Benches that attempt to measure ST performance are mainly to show what execution resources are available at lighter thread loads. Just like benches that load all cores at 100%, they are not realistic representations of what a PC is used for all the time. They exist to provide data points that should be considered as part of the whole.

 

[crashtech]

Sysmark is basicaly an Intel Bench, ask Bapco, nicely choosen, anyway...snip

I'm not going to fall for it. My disclaimer made it clear why I am reluctant to post benches. If you want to directly refute my point, post an ST bench that includes an i3-4340 or better, and a heavily OCed FX.

 

[Cerb]

Actually those 44xx i5s tend to have a pretty poor Turbo multiplier. Depending on the need, a faster i3 can be better.

For a Flight Simulator X box on a budget, sure. That's not much different than the cherry-picking to try to make the AMDs look great, but pushed down by some conspiracy, instead of accepting that their cost is due to their value relative to the competition.

i5-4570: $200, 3.2/3.6GHz, 6MB
i5-4460: $180, 3.2/3.4GHz, 6MB
i3-4370: $160, 3.8GHz, 4MB
i3-4150: $120, 3.5Ghz, 4MB

+33% cost for +8.57% performance in the best case, on the i3-4370. At least the i5 gets you from 2C to 4C for the extra cost.

 

[Abwx]

I'm not going to fall for it. My disclaimer made it clear why I am reluctant to post benches. If you want to directly refute my point, post an ST bench that includes an i3-4340 or better, and a heavily OCed FX.

Because you think that a i3 will have better ST perf than a i5 4670k..??

Do you know what single thread means.?.

You can take any CPU from a given uarch to do such a test.

A disclaimer is needed when one know that his choices are not that good, i put no disclaimer for the benches i posted, i think that they are quite reliable unless someone can prove me the contrary.

 

[Cerb]

Indeed, at which point even with cherry picked arbitrary scenario's like "heavy background video encoding whilst heavy gaming" (LOL) the A10 gets crushed. It's a CPU that makes sense at $130-$150 but as it is, even the slowest i5 is roughly +35-40% faster averaged across a dozen apps (7-zip, Lightroom, Visual Studio, X264, etc). A +40% boost for +1% price-tag increase ($180 vs $182) is a total no brainer to all but die-hard fanboys (not even factoring in power consumption differences).

As for "encoding background video whilst playing games" (odd usage comparisons which only ever seem to spring up when comparing AMD quad's to Intel's duals... :sneaky it makes far more sense on ANY CPU to encode when you're not doing anything else (ie, when you're eating dinner, showering, taking the dog for a walk, browsing the web, or for huge workloads of dozens of queued up vids even leave it running overnight then set it to switch itself off when finished, etc). Common sense / intelligent time management is a good thing regardless of what CPU you own.

The A8-7600 is, IMO, the main Kaveri with appeal. With one good reason to have a modern AMD Radeon GPU, it can offer some serious value. Intel's drivers may be good these days in many ways, but options aren't among them, nor productivity application optimization (for the desktop driver versions, anyway). Intel's is good enough, but AMD's is still better.

 

[crashtech]

For a Flight Simulator X box on a budget, sure. That's not much different than the cherry-picking to try to make the AMDs look great, but pushed down by some conspiracy, instead of accepting that their cost is due to their value relative to the competition.

i5-4570: $200, 3.2/3.6GHz, 6MB
i5-4460: $180, 3.2/3.4GHz, 6MB
i3-4370: $160, 3.8GHz, 4MB
i3-4150: $120, 3.5Ghz, 4MB

+33% cost for +8.5% performance in the best-case, on the i3-4370. At least the i5 gets you from 2C to 4C for the extra cost.

You are right, but I don't think that necessarily negates my point for a machine that won't see heavy loads.

 

[crashtech]

Because you think that a i3 will have better ST perf than a i5 4670k..??

Do you know what single thread means.?.

You can take any CPU from a given uarch to do such a test.

A disclaimer is needed when one know that his choices are not that good, i put no disclaimer for the benches i posted, i think that they are quite reliable unless someone can prove me the contrary.

So, you can't provide proof. Let's move on.

Edit:

The reason I won't do as you suggest is that first off, performance does not always scale linearly with speed. Second, how many readers even know what the turbo clock speed of all these CPUs are in ST? It's too confusing.

 

[Abwx]

So, you can't provide proof. Let's move on.

The i5 has same ST perf as i3, the proof is here, it s just that you are left finding some harbour in denial by asking the same graph but with a i3, for the record a FX scale almost perfectly with frequency, it s not difficult to extrapolate the scores, also it is clear that the FX has more disadvantage in the FP department than in the Integer related softs, it can match a 4770K in Povray but in Integer code if maxed out it will handily beat it, so in this last case, ie Integer ST perf, the deficit is significantly less.

Edit:

The reason I won't do as you suggest is that first off, performance does not always scale linearly with speed. Second, how many readers even know what the turbo clock speed of all these CPUs are in ST? It's too confusing.

Turbos are 4.2 for the 8350, 4.3 for the 8370/8370E, in this latter case one should compare with the 8350 to see if the turbo effectively reach 4.3 but in principle it should.

 

[crashtech]

The i5 has same ST perf as i3, the proof is here, it s just that you are left finding some harbour in denial by asking the same graph but with a i3, for the record a FX scale almost perfectly with frequency, it s not difficult to extrapolate the scores, also it is clear that the FX has more disadvantage in the FP department than in the Integer related softs, it can match a 4770K in Povray but in Integer code if maxed out it will handily beat it, so in this last case, ie Integer ST perf, the deficit is significantly less.



Turbos are 4.2 for the 8350, 4.3 for the 8370/8370E, in this latter case one should compare with the 8350 to see if the turbo effectively reach 4.3 but in principle it should.

Oy. Fine. Graph you chose. Adding 20% to the 8370's ST score nets 301.82. The 4670K scores 372.53. This suggests that a fast Haswell i3 would have better ST than a 9590 to me. Are you suggesting otherwise?

 

[Abwx]

Oy. Fine. Graph you chose. Adding 20% to the 8370's ST score nets 301.82. The 4670K scores 372.53. This suggests that a fast Haswell i3 would have better ST than a 9590 to me. Are you suggesting otherwise?

Not at all, i pointed that the FX is more at a disadvantage in FP than in Integer, my point was that benches that are obviously favouring Intel CPUs will inflate considerably this number and distort the real picture even in FP.

You can imagine how much misleading it is then to assume that those phony FP benches are representative of Integer ST perf, that is adding a second bias to a previous one, litteraly.

Integer ST perf can be extracted using any soft that max out the CPUs MT wise assuming some corrections are made to account for scalability, that s why i used Anand s servers scores since theses are sure scores without any need to extrapolate.

That said i find funny that some people do rely on single thread scores of...multithreaded softs, that is to rely on data that show the CPUs in an utilisation that is completely stupid since no none would use MThreaded softs in ST situations, the real comparison is on MT , in real scenarios , not in partial and pointeless usages that do not exist.

 

[crashtech]

Good grief, how many hundreds of words does it take to not substantively disagree with me? This fixation on supposedly rigged benchmarks did not really help. I pull data from the Anandtech bench when possible because they are our generous hosts. Complain to them if you don't like them.

 

[SlowSpyder]

Oy. Fine. Graph you chose. Adding 20% to the 8370's ST score nets 301.82. The 4670K scores 372.53. This suggests that a fast Haswell i3 would have better ST than a 9590 to me. Are you suggesting otherwise?

For whatever it is worth, I just ran this bench @ 5.1GHz (222 x 23). I scored 307.21 PPS single threaded, 1601.19 PPS with all eight cores.

*edit - Was throttling, changed some settings, scored 1895PPS at the same clocks.

 

[Yuriman]

For whatever it is worth, I just ran this bench @ 5.1GHz (222 x 23). I scored 307.21 PPS single threaded, 1601.19 PPS with all eight cores.

Pretty respectable. 60% better multithreaded and 82% of the single threaded performance of a stock i5.

Overclocked to 4.6 (and accounting for stock turbo frequencies) which I think is a fairly commonly achievable overclock with recent Haswell chips, I'd estimate about 475/1225 from the i5. That would give the i5 a ~55% lead in singlethreaded performance, with the FX having a 30% lead in MT.

Or, in other words, I'd expect a heavily overclocked FX-6300 to match a heavily overclocked i5 in multithreaded scenarios, with 2/3 the single-threaded performance (with a 500mhz clock advantage).

The i5 carries a price premium over lower-end AMD octa's, but runs about the same as an FX-9370.

 

[crashtech]

For whatever it is worth, I just ran this bench @ 5.1GHz (222 x 23). I scored 307.21 PPS single threaded, 1601.19 PPS with all eight cores.

Yeah, good enough to beat a 2700K in ST, and within 2% of what I predicted. FX CPUs do indeed seem to scale quite linearly on this bench.

 

[Abwx]

Good grief, how many hundreds of words does it take to not substantively disagree with me? This fixation on supposedly rigged benchmarks did not really help. I pull data from the Anandtech bench when possible because they are our generous hosts. Complain to them if you don't like them.

No need to complain since i know the limitations of thoses benches, as for Anandtech that s their problem if they give some credibility to those that point some of their benches as being phony, i m particularly pointing 3Dparticle movement, personaly i would be more professional on this matter if asked to manage a site bench suite, but whatever, to each his own...

For whatever it is worth, I just ran this bench @ 5.1GHz (222 x 23). I scored 307.21 PPS single threaded, 1601.19 PPS with all eight cores.

ST score is on line but MT score is well below the expectations, Computerbase.de scored 1711 at 4.7, you should be at about 1850.

http://www.computerbase.de/2014-09/amd-fx-8370e-im-test/2/#diagramm-pov-ray

 

[crashtech]

Abwx, this is the statement to which I was originally responding:

... i3 is locked and even tho has considerably higher ST IPC, FX clocks MUCH higher and can make up that deficit completely...

In the light of our recent dialogue, do you agree or disagree with inf64's statement?

 

[Abwx]

Abwx, this is the statement to which I was originally responding:
In the light of our recent dialogue, do you agree or disagree with inf64's statement?

Depending of the bench one could make him wrong or right, he may be wrong for CB 11.5, of course..., or even Povray but he s right for Fritzchess or 7zip.

So now who is the one who is "the more right".?.

As already pointed all thoses softs, without exception, are Multithreaded softs wich are used as such, what matters is what your CPU can extract from thoses softs, that said i d like to hear about a single threaded soft that would be annoyingly slow on a FX, i guess that there wont be much if any, that s why i find all this debate quite pointeless, what matters are the final perfs in realistic scenarios.

 

[Phynaz]

i3 vs fx83xx? That gotta be a trick question or what? i3 in 2014/2015 Vs FX83xx is no brainer, FX83xx is just a better CPU. i3 is locked and even tho has considerably higher ST IPC, FX clocks MUCH higher and can make up that deficit completely. Then we go to MT workloads which are becoming more and more prevalent each day goes by and i3 looks like a toy compared to FX. Sorry to say it but i3 has nothing on FX83xx.

Quicksync.

 

[frozentundra123456]

And an igp which means you dont have to add a discrete card like you do for the already power hungry FX. For the vast majority of users, that dont overclock and use heavily multithreaded software, the i3 is actually a better choice.

 

[Abwx]

On another thread a member has discovered that his G3258 at 4.6 has about half the FP throughput of a stock Kaveri, that is, the same as a 2C 4.0 Kaveri, i guess that no one will use IBT as single thread perf reference...

for the already power hungry FX.

I would call this an automated reflex.

For the vast majority of users, that dont overclock and use heavily multithreaded software, the i3 is actually a better choice.

The vast majority use multithreaded softs, wake up , we are in 2014 with 2015 being at the corner...

 

[SlowSpyder]

ST score is on line but MT score is well below the expectations, Computerbase.de scored 1711 at 4.7, you should be at about 1850.

http://www.computerbase.de/2014-09/amd-fx-8370e-im-test/2/#diagramm-pov-ray

You are absolutely right. I watched it more closely this time, some of the cores are going down to a 7x multiplier during the run (~1.5GHz). Looking into that now, score should be better...

 

[Sweepr]

I wonder how many years will it take till an AMD CPU actually matches the ST performance of a 4.6GHz Pentium G3258 using 24/7 cooling.