Asus' First AMD-Based ROG Motherboard Is Coming Soon

[damric]

Before this gets a bit too out of hand (a benchmark-off!), I'd like to point out that the Athlon 750K (4.3 ghz overclock) performed roughly on par with the FX-4350 (stock 4.2 ghz). Also, I'd like to (re-)point out that the Athlon 760K is the updated, Richland-based variant, with all the changes already listed.

I love data; I would enjoy seeing a benchmark comparison.

Well I tried to give a variety of free benchmarks. One to showcase single thread (Pentium should win), one to showcase multitask (3D mark), and one to show brute force calculation using only 4 threads (IBT AVX linpack).

 

[RedJaron]

Hmm, I might run some of those on my 2600K. Not that it's necessarily applicable ( $80 CPU vs $300 one, ) but it might make an interesting data point.

 

[Calculatron]

I did a touch of digging, and Anandtech has the Cinebench R15 benchmark with the G3258. It achieved a result of 124 at stock, and 173 at a 4.7 ghz overclock.

 

[childofthekorn]

My bad, we're on two completely different subjects. I'm referring to APUs vs CPU and dedicated GPU lol

And by the way, you're the only person I've ever seen say that Athlon's are better than the newest FX series. Especially considering a dual-core Pentium (the G3258) beats the 750k pretty well.

A10-7850k can compete against an 8350 in non-CPU intensive tasks where multithreading is key. the APU's are fine for every day use and even gaming. Its the multithreaded number crunching where the FX series thrives, but still compared to intel its only marginally slower in multitasking, with as low quality ALU and FPU units they have, rumor mill is turning that they're creating a custom higher-quality ALU/FPU combo for their excavator chips, but we've seen where thats led us thus far.

 

[RedJaron]

Second run of benches after disabling HT and two of my cores and OCing to 4.4 GHz ( I could probably reach 4.6 or more if I wanted to take time to fine tune it. ) So this is the closest I can do to approximating a 3258 right now ( take into account IPC improvements between Haswell and SB. )

Cinebench - 154cb
3DMark - 4757
IBT - 58.5 GFlops


And then, just for fun I ran the tests on the full 4C/8T CPU OC'd to 4.4

Cinebench 8T - 765cb
3DMark Physics - 10313
IBT - 94 GFlops

I think I'll try them with the full four cores but HT disabled. I have a sneaking suspicion . . .

UPDATE: And yep, it seems IBT doesn't favor HT too much. Running my 2600 at 4C/4T I get 112 GFlops instead of the 94 at 4C/8T. Interesting.

 

[bgunner]

Phenom II 965 @ 4.073 with 3 cores disabled

Cinebench - 84cb
3Dmark - N/A
ITB 54.3 GFlops <-- done with all cores enabled.

for an older CPU the Athlon is not to far ahead of it. But we are also talking 3 gens older also.

EDIT: After figuring out how to run Cinebench's single core test and leave my other core's enabled in the Bios, I got a score of 106. Your Athlon is still lacking because the lack of having L3 cache on the chip.

Proof:

 

[damric]

Phenom II 965 @ 4.073 with 3 cores disabled

Cinebench - 84cb
3Dmark - N/A
ITB 54.3 GFlops <-- done with all cores enabled.

for an older CPU the Athlon is not to far ahead of it. But we are also talking 3 gens older also.

EDIT: After figuring out how to run Cinebench's single core test and leave my other core's enabled in the Bios, I got a score of 106. Your Athlon is still lacking because the lack of having L3 cache on the chip.

Proof:

The L3 cache on Phenom II is night and day compared to that on the AM3+ CPUs in that it is actually beneficial. Compair it in AIDA 64. You'll see that the AM3+ L3 cache has latency as high as 60ns.

The reason your score is high is because Phenom II @4GHz has more IPC than anything Bulldozer/Piledriver. Also Phenom II x4 should rightly score higher FLOPS since it has four floating point units compared to only two in the Athlon. The Athlon mitigates this by calculating FLOPS with its integer cores with AVX. While not quite as efficient, it does a pretty good job.

But anyways, we aren't comparing Phenom IIs. If we were, I could show you that I could also get 50GFLOPs on my old 4GHz Athlon II x4, but this isn't 2009.

 

[damric]

Second run of benches after disabling HT and two of my cores and OCing to 4.4 GHz ( I could probably reach 4.6 or more if I wanted to take time to fine tune it. ) So this is the closest I can do to approximating a 3258 right now ( take into account IPC improvements between Haswell and SB. )

Cinebench - 154cb
3DMark - 4757
IBT - 58.5 GFlops


And then, just for fun I ran the tests on the full 4C/8T CPU OC'd to 4.4

Cinebench 8T - 765cb
3DMark Physics - 10313
IBT - 94 GFlops

I think I'll try them with the full four cores but HT disabled. I have a sneaking suspicion . . .

UPDATE: And yep, it seems IBT doesn't favor HT too much. Running my 2600 at 4C/4T I get 112 GFlops instead of the 94 at 4C/8T. Interesting.

Did you make sure it was the AVX version of IBT? It might help.

 

[RedJaron]

Did you make sure it was the AVX version of IBT? It might help.

It's v2.54, looks the exact same as the screen you posted. Unless there's a v2.54 that doesn't use AVX, I'm guessing I've got the right one.

 

[bgunner]

So because the Phenom ii is older and can keep up with your Athlon means that the newer CPU's can't? There has been benchmarks put in links in this thread that dispute your claim. But you keep fighting this, If my son brings his PC over I will bench his FX chip and prove that your claims are under a faulty assumption.

 

[Calculatron]

Yes! Yes! For science!

 

[RedJaron]

So because the Phenom ii is older and can keep up with your Athlon means that the newer CPU's can't?

That's not what we've said at all. We've said that due to differences in architectures and memory controllers from the older Phenoms to the FX and the new Athlon designs, the effectiveness of RAM and cache have drastically changed in the newer chips. I've already elaborated on this. The memory controller and cache access speed in the FX chips was a step back from the Phenoms to the BD/PD chips. The Richland chips improved the memory controller ( though not the cache speed, I believe, ) so that they can access system RAM nearly as fast as the FX chips can hit their L3 cache. With fast RAM and proper tuning, the lack of L# becomes less an issue in many tasks. If you look at this review, compare the bars for the 750K @ 4.3GHz to the FX-4300 @ 4.2GHz. You'll see in many cases they're very similar. Yes, the 750K still trails a little overall, but keep in mind that the 750K is Trinity, not Richland, so it doesn't have the improved memory controller, and the RAM used was only 1866. Damric has already explained the difference in FLOPS, since the BD modules have half the FP arrays the Phenoms do.

There has been benchmarks put in links in this thread that dispute your claim.

All the links provided were for the 750K, not the 760K.

But you keep fighting this, If my son brings his PC over I will bench his FX chip and prove that your claims are under a faulty assumption.

I actually would like to see that. Not to be antagonistic, but to see the actual numbers. I'd find it an interesting comparison.

 

[damric]

Just look at how pathetic the L3 latency is. Not worth the extra die space and heat.

FX-8320

Spoiler

50ns L3 latency (face palm)

FX-6300

Spoiler

65ns (double face palm)

Athlon 760K

Spoiler

RAM latency 53ns

In order for L3 to be worthwhile, the latency should be a magnitude faster than the RAM latency as it is on Intel CPUs.

 

[simplyjackc]

Right as I bought my a88x-pro...

 

[anonimous]

So spending more on a motherboard than on a CPU has always been foolish...

Asus ROG has always been overpriced, one in ten people that get them do not ever use the "functions" it provides.

Asus must have done a market analysis and concluded that there are enough suckers on this planet who will buy this board (doesn't even include the AMD APU) instead of a G3258/H81/used 7770 for a little bit more.

 

[damric]

So spending more on a motherboard than on a CPU has always been foolish...

Asus ROG has always been overpriced, one in ten people that get them do not ever use the "functions" it provides.

Asus must have done a market analysis and concluded that there are enough suckers on this planet who will buy this board (doesn't even include the AMD APU) instead of a G3258/H81/used 7770 for a little bit more.

The ROG boards on every other socket come with a huge premium and not much more features. The Crosshairs are the most notorious for being very low value for price. There is a market for premium overpriced stuff like Dominator RAM, ROG, Titan-Z, ect.

 

[RedJaron]

Did a little more tweaking and got my system up to 4.7GHz ( if I could get some expert help, I think I can safely hit 4.8GHz. ) Also, I had a derp moment when I ran my last IBT tests with "All" threads rather than four, so you can disregard those. Strangely, adding the threads didn't seem to impact it all that much.

Here are some final numbers:

Cinebench - 162 ( 271 combined, both cores )
3DMark Physics - 5063
IBT - 62 GFlops

So the 760K isn't keeping up in either Cinebench ( which we knew would happen, ) or IBT ( which is surprising to me. ) And even with the Haswell IPC improvements, I don't see the G3258 making up 500 points in a physics test.

 

[damric]

Did a little more tweaking and got my system up to 4.7GHz ( if I could get some expert help, I think I can safely hit 4.8GHz. ) Also, I had a derp moment when I ran my last IBT tests with "All" threads rather than four, so you can disregard those. Strangely, adding the threads didn't seem to impact it all that much.

Here are some final numbers:

Cinebench - 162 ( 271 combined, both cores )
3DMark Physics - 5063
IBT - 62 GFlops

So the 760K isn't keeping up in either Cinebench ( which we knew would happen, ) or IBT ( which is surprising to me. ) And even with the Haswell IPC improvements, I don't see the G3258 making up 500 points in a physics test.

Good work!

 

[RedJaron]

Any thoughts about the strange IBT scores?

 

[damric]

Any thoughts about the strange IBT scores?

IBT calculates floating points. i7 not only has strong FPUs, but it also has AVX instructions, so really either way it's going to do pretty well. The Pentium lacks AVX so it's probably not going to do nearly as well as your i7 simulated.

Hyper Therading isn't quite as efficient in tests like this but it still helps some.

 

[RedJaron]

Ahh, I didn't know the Pentiums and Celerons didn't have AVX. Well, that invalidates most of my comparison. At least it was fun to push my chip higher than I've ever done before.

My main questions on the IBT were why four threads on a 2T CPU only dropped the performance 5%. I would have thought it'd have much more impact. I mean, running eight threads on 4C/8T only gets me 94, but running four threads on 4C/4T gets 112. Things are just looking weird here.

 

[damric]

Ahh, I didn't know the Pentiums and Celerons didn't have AVX. Well, that invalidates most of my comparison. At least it was fun to push my chip higher than I've ever done before.

My main questions on the IBT were why four threads on a 2T CPU only dropped the performance 5%. I would have thought it'd have much more impact. I mean, running eight threads on 4C/8T only gets me 94, but running four threads on 4C/4T gets 112. Things are just looking weird here.

I'm hoping someone has the Pentium to test for just that reason. It has stronger FPU than the Athlon, but the Athlon can make up for it with AVX. Would be interesting to see which is faster in that regard.

Hyper threading is weird. Often it works better with lighter loads. It also seems to benefir i3s more than i7s.

 

[bgunner]

Ahh, I didn't know the Pentiums and Celerons didn't have AVX. Well, that invalidates most of my comparison. At least it was fun to push my chip higher than I've ever done before.

My main questions on the IBT were why four threads on a 2T CPU only dropped the performance 5%. I would have thought it'd have much more impact. I mean, running eight threads on 4C/8T only gets me 94, but running four threads on 4C/4T gets 112. Things are just looking weird here.

ITB is known to have ups and downs in its Gflops. As an example Both my Phenom and Athlon scores will range by about 5 GFlops from test to test. My Phenom II will get between 50.3 to 55.2 depending on the run. One run I will get the 50 -53 score and teh next run will get 52- 55 score. My Athlon II will get 38 - 41 one test and 42- 45 the next. Add this in to the variable of some of the higher reading and the rest being a difference in HT. A extra run or two wont hurt as long as you have the temp head room to make sure there is no fluke score coming out for what ever reason.

 

[james pinnix]

So my big question is about this board

Is amd going to bring out anything worth putting in it

Because an amd a10-7850k isn't
Good enough to have a board like this

I will say this is a very good thread I'm really curious about the amd athlon x4 860k as well

 

[RedJaron]

ITB is known to have ups and downs in its Gflops. As an example Both my Phenom and Athlon scores will range by about 5 GFlops from test to test. My Phenom II will get between 50.3 to 55.2 depending on the run. One run I will get the 50 -53 score and teh next run will get 52- 55 score. My Athlon II will get 38 - 41 one test and 42- 45 the next. Add this in to the variable of some of the higher reading and the rest being a difference in HT. A extra run or two wont hurt as long as you have the temp head room to make sure there is no fluke score coming out for what ever reason.

I noticed a +/- 3 variance on my scoring runs as well, but that doesn't explain 18 additional GFlops. Even if the 112 was the +3 side and 94 was the -3 side, that still leaves a gap of 12, and not in HT's favor. I know HT has some overhead associated with it and it can't double the processing, but this is a net negative affect. The only thing I can think of is that disabling the HT opened thermal headroom so each core to spin up and maintain its maximum freq for as long as possible. But that doesn't seem quite right.