[HardOCP] Asus DC II 290X max overclock versus GTX 780ti max overclocking review:

[SiliconWars]

http://forums.anandtech.com/showpost.php?p=35899688&postcount=225

The obvious problem here is that the GTX 780 had a sample of 20 while the 290X had a sample of 4. Anyone dealing with statistics? Problem right there.

It's pretty awesome how statistics can be skewed . Let's just play stupid and throw in a sample size of 1. Or four. Versus sample size of 20 including reference cards.

Didn't you bash the reference 290X on a sample count of 2 at toms and techreport?

 

[blackened23]

Do you have more results for oced aftermarket 290Xs? I couldn't find any when I was looking a few weeks back.

Do you want me to fix it? I can put some effort into that later. And since you included reference cards for the 780, i'll include reference overclocks for the 290X. Sample size of 20 for both sides.

Let's re-hash our statistics lesson for the day. Sample size increases == accuracy increases. Sample size decreases == accuracy decreases. Your 290X had a sample size of 4. GTX 780? 20. But we'll do 20 vs 20 for both sides. Including reference overclocks for both sides. Most statistics generally prefer 50 sample size minimum to be representative of a population, but sample size of 50 aint happening.

I will get this going , but will take some time.

 

[Silverforce11]

Well he's bashing the R290X based on the sample size of [H], where they had no scaling in TWO out of FOUR games, in which other sites found good scaling across the same games as [H] and heaps more. [H] has a defective R290 that runs much hotter than others even when they cram in the same volts, at less fan speed than [H] 75%.

So I dunno, when he claims others are cherry picking it screams hypocrisy to me. What next blackened, you gonna start accusing me again of being on an AMD propaganda squad when you clearly yourself are so pro NV its plain as day. Hypocrite.

I mean seriously tell me how again you can claim R290/X don't scale well with OC?

When its not running in silent mode it looks just fine to me.

 

[blackened23]

Didn't you bash the reference 290X on a sample count of 2 at toms and techreport?

For the performance variance that AMD admitted to and released the "Hey let's increase our fan RPMs again" driver?

 

[BallaTheFeared]

All of the reviews I linked had oced aftermarket 780s. Might want to read through the reviews before you disregard the results because they didn't match up with your expectations.

None of them have 290s in them either.

If they had a Titan in those results it would be faster than the 290x at the same clocks at the 780.

 

[Silverforce11]

None of them have 290s in them either.

I believe the original post where you guys were arguing over that, it was due to the use of 780 OC being X,Y,Z faster than Titan therefore its A,B,C faster than R290s.

 

[BallaTheFeared]

I believe the original post where you guys were arguing over that, it was due to the use of 780 OC being X,Y,Z faster than Titan therefore its A,B,C faster than R290s.

I never argued any of this with the Titan, perhaps Black did?

I don't really care about the Titan, all I know is the results Elfer posted were not jiving with my own results and recent 780 OC vs R290/x OC reviews.

 

[Silverforce11]

I never argued any of this with the Titan, perhaps Black did?

I don't really care about the Titan, all I know is the results Elfer posted were not jiving with my own results and recent 780 OC vs R290/x OC reviews.

Yep I do agree recent reviews, they are very close. At average OC, R290 is still a bit faster than 780, but I haven't seen many r290 hit 1.3ghz whereas 780 can. So a fair summary would be that 780 has higher potential OC.

 

[blackened23]

So I dunno, when he claims others are cherry picking it screams hypocrisy to me. What next blackened, you gonna start accusing me again of being on an AMD propaganda squad when you clearly yourself are so pro NV its plain as day. Hypocrite.

I don't' care if anyone has a brand preference. Everyone is biased. I don't give a crap. I'm not giving you the "Oh hey I don't have a brand preference" rigamaroll because everyone has a preference. Where I draw the line is at mis-represented data (intentional or not) and false statements. I think elfears results were skewed, but i'm not saying that was intentionally done. Clearly there was not a lot of data near launch. Yet I am still confused as to why the 290X side did not have reference OC's while the 780 did.

Anyway, I am saying his statistics are mis-aligned due to the incorrect sample size and the fact that the 780 included reference cards while the 290X did not. That heavily skews the results in favor of the 290X. Have you studied statistics? You can't compare sample of 4 to sample of 20, especially when the sample of 20 has reference cards.

Also, to be clear:

I didn't accuse you of that. I did say something else though, which your posts when you first arrived on ATF clearly indicated. Plain as day I might add. You don't wanna go there do you?

 

[Grooveriding]

http://i.imgur.com/uYIe8.jpg

That's not much of a 'source'. No mention of tflops ? Really nothing there but the usual chip pictures on a graph.

Nvidia simply cannot compete on perf/w against Intel. They are going to be using a process that is 2 years behind Intel's 14nm. They don't have a chance.

 

[ams23]

That's not much of a 'source'.

Right. So an Intel slide is fine as a source, but an NVIDIA slide is not?

No mention of tflops ?

That is a very simple calculation because the graph shows projected GFLOPS/w on the y-axis. We know that power consumption will be somewhere ~ 225-250w at most. That means > 3 TFLOPS DP peak throughput.

Nvidia simply cannot compete on perf/w against Intel. They are going to be using a process that is 2 years behind Intel's 14nm. They don't have a chance.

That is false. Architectural considerations drive the efficiency more than anything else at this point. NVIDIA's Kepler architecture already powers the top 10 greenest supercomputers in the world, and Maxwell will be nearly 3x more efficient than Kepler with respect to DP compute.

 

[Grooveriding]

Right. So an Intel slide is fine as a source, but an NVIDIA slide is not?

Given how often nvidia slides referencing performance or release timeframes of unreleased parts are totally off from reality, yes.

That is a very simple calculation because the graph shows projected GFLOPS/w on the y-axis. We know that power consumption will be somewhere ~ 225-250w at most. That means > 3 TFLOPS DP peak throughput.

We know nothing, so guess work on your part.

That is false. Architectural considerations drive the efficiency more than anything else at this point.

Right. Intel's industry-leading fabs, years ahead of anything TSMC can do, have no significant perf/w advantages over TSMC ? I mean, it's not like they can put in place the same number of transistors with a lower power requirement, or more transistors using less or equal amounts of power.

We won't even get into TSMC's initial 20nm not even being the real deal, which we won't see until 2015.

Your posting opinions/hopes as facts.

 

[VulgarDisplay]

This thread is entertaining. No one reads what the other person is saying. One side says 780ti is so fast!!!! Other side says there is something wrong with their 290x. Other side says 290x is so slow 780ti is so fast. Other side says there is something wrong with their 290x..... Rinse repeat.

What reviews I looked at didn't show much for 290x oc scaling in the same titles from this review. 290x is a engineering marvel but it is crippled by its density and die size. 780ti is crazy good and crazy expensive.

 

[ams23]

Given how often nvidia slides referencing performance or release timeframes of unreleased parts are totally off from reality, yes.

Right, so logically we should put more weight in one slide's projection than the other slide's projection? Hilarious.

Intel's industry-leading fabs, years ahead of anything TSMC can do, have no significant perf/w advantages over TSMC ?

Architectural efficiency is even more important than process efficiency in this day and age. Kepler is already at a fabrication process disadvantage compared to Knight's Corner, and yet is still well ahead in the Green 500 Top 10 rankings of the world's most power efficient supercomputers.

Your posting opinions/hopes as facts.

Pot meet kettle? You are the one who claims they "don't stand a chance" without having even a clue of what you are talking about. The reality is that NVIDIA's projected DP compute efficiency in Maxwell is just as impressive as Intel's projected DP compute efficiency in Knight's Landing.

 

[Grooveriding]

You keep on with the red herrings about today's supercomputer efficiency standings and architectural efficiencies while ignoring the massive process advantage Intel has.

From your postings it's clear in your world nvidia can out-engineer Intel as well as overcome a huge process disadvantage.

 

[Imouto]

You keep on with the red herrings about today's supercomputer efficiency standings and architectural efficiencies while ignoring the massive process advantage Intel has.

From your postings it's clear in your world nvidia can out-engineer Intel as well as overcome a huge process disadvantage.

In fact Apple ridiculed Intel a few months ago with the A7 in a worse node (28 nm Samsung no less). Intel is just a newbie in too many fields. We also can't forget that CUDA is the de facto standard for GPGPU computing. All in all I'd say that Intel will release a passing by product.

 

[blackened23]

In fact Apple ridiculed Intel a few months ago with the A7 in a worse node. Intel is just a newbie in too many fields. We also can't forget that CUDA is the de facto standard for GPGPU computing. All in all I'd say that Intel will release a passing by product.

Bay Trail is an excellent product and still has the best CPU performance of any SOC around except perhaps the A7 - BT is even faster than the snapdragon 800. A7 may be better as well. Sure, intel doesn't have the best graphics performance in their SOC yet but I think they will definitely improve there with silvermont. Quite sure of it. Anyway, even if the A7 is faster that doesn't matter. Why? A7 is an iOS only SOC. There will not be any Qualcomm, Mediatek, nvidia, or intel products ever included in an iOS product, so the net effect is nothing. Similarly, A7 will never appear in any android or windows product. So i'm not sure why this comparison is even being made.

This is aside from the fact that cross platform benchmarks are questionable at best, yet with that being said I do like Apple quite a bit - I own the ipad air among other things which I really like using. But I just don't see the point of the comparison. Intel isn't competing with Apple for iOS products. Apple isn't competing for anyone with any android or windows product. So what's the point of comparing any ARM SOC or Bay Trail to A7? There's no point, really. A7 is for its own thing with iOS, period.

Right now, it is true that Qualcomm is winning mobile because of their LTE and CMDA integration. As other vendors catch up to them in that respect, it will be closer to parity IMO. While the SD 800 was an impressive SOC, their LTE integration is literally the biggest reason they've done so well as far as I can tell.

As far as nvidia vs intel? Nvidia has an entire ecosystem with CUDA which is relevant for super computing. As far as I can see it, intel are the ones that need to prove themselves in this field, because software support is everything there. And nvidia is better than intel in this respect. Yes, intel has the process advantage, but i'm not seeing this as an entirely relevant comparison. We are talking "big" Maxwell for super computing right? Or am I missing something?

 

[Imouto]

Or am I missing something?

The whole point as usual. The A7 having a stellar performance in a phone (not a tablet or an activelly cooled handheld console) makes Apple devices attractive versus the competition. It doesn't matter if it can't be found in any other brand or OS it is directly competing with them as Android is more and more polished. Not only that, Apple achieved it in a process considered even worse than TSMC's hence the "ridicule".

 

[Grooveriding]

No I said as much as well and that I was just speaking to raw performance of DP and perf/w from upcoming Intel parts. As far as what they will sell or how they will do against nvidia in terms of the market, who knows.

I think SoC is just starting to move for Intel in mobile as well. It's just a matter of time for them to get going and once they have their designs in place with the advantage of their fabs, they can at the very least have the best performing parts and the most efficient parts. Though there again it will be a case of if they can get their SoCs in devices.

Has Intel stated whether they are going to use any 14nm for mobile parts ?

 

[ams23]

From your postings it's clear in your world nvidia can out-engineer Intel

No, that's not what I said at all. NVIDIA's strength is GPU computing (and they are arguably the best in the world in this area). Intel's strength is CPU computing (and they are arguably the best in the world in this area). In the HPC space, their next gen architectures are projected to have similar DP perf. and similar DP perf. per watt, period.

 

[Elfear]

I don't' care if anyone has a brand preference. Everyone is biased. I don't give a crap. I'm not giving you the "Oh hey I don't have a brand preference" rigamaroll because everyone has a preference. Where I draw the line is at mis-represented data (intentional or not) and false statements. I think elfears results were skewed, but i'm not saying that was intentionally done. Clearly there was not a lot of data near launch. Yet I am still confused as to why the 290X side did not have reference OC's while the 780 did.

Can you link me to the reviews your referring to with overclocked reference cards? I didn't use reference cards for any of the calculations I made. None.

While there aren't a lot of aftermarket 290X reviews out yet, there are plenty of aftermarket 780 reviews. Out of the 20 or so cards tested, the average oced aftermarket 780 is 8-9% faster than a stock Titan. I don't know why there is controversy over a fairly large sample size of 780s except for the fact that the reviews are more than a month old. I don't believe that matters though since we're comparing 780's against a stock Titan.

If you don't think I made the calculations right, feel free to correct my mistakes.

 

[BallaTheFeared]

How did this go from $500-700 desktop GPUs to slow as dirt cellphone chips?

 

[Imouto]

How did this go from $500-700 desktop GPUs to slow as dirt cellphone chips?

Because threads have at least one thing in common with trains.

 

[BallaTheFeared]

Can you link me to the reviews your referring to with overclocked reference cards? I didn't use reference cards for any of the calculations I made. None.

While there aren't a lot of aftermarket 290X reviews out yet, there are plenty of aftermarket 780 reviews. Out of the 20 or so cards tested, the average oced aftermarket 780 is 8-9% faster than a stock Titan. I don't know why there is controversy over a fairly large sample size of 780s except for the fact that the reviews are more than a month old. I don't believe that matters though since we're comparing 780's against a stock Titan.

If you don't think I made the calculations right, feel free to correct my mistakes.

You go with that, I'll go with recent reviews of overclocked 780s and R290/xs instead of trying to math performance from reviews that don't even include either AMD card.

 

[Elfear]

How did this go from $500-700 desktop GPUs to slow as dirt cellphone chips?

:biggrin:

Because threads have at least one thing in common with trains.

So true. I think we all need to go game for a bit.