this is gonna' be gud - nVidia calls out Intel for cheating in GPGPU benchmarking | updated

[zMeul]

source: http://arstechnica.com/gadgets/2016/08/nvidia-intel-xeon-phi-deep-learning-gpu/

 

 

Quote

Nvidia has called out Intel for juicing its chip performance in specific benchmarks—accusing Intel of publishing some incorrect "facts" about the performance of its long-overdue Knights Landing Xeon Phi cards.

Nvidia's primary beef is with the following Intel slide, which was presented at a high performance computing conference (ISC 2016). Nvidia disputes Intel's claims that Xeon Phi provides "2.3x faster training" for neural networks and that it has "38 percent better scaling" across nodes.

 

 

sad day for Intel ... I though they were done with bullcrap like this, instead they pull shit out of the same bag AMD used to

 

nVidia accuses them for comparing their previous gen Maxwell cards against Intel's Xeon Phi, instead of using Pascal

furthermore, nVidia points out that Intel used an older version of the AlexNet benchmark - the updated one shows improvements for Maxwell

Quote

Intel claimed that a Xeon Phi system is 2.3 times faster at training a neural network than a comparable Maxwell GPU system; Nvidia says that if Intel used an up-to-date version of the benchmark (Caffe AlexNet), the Maxwell system is actually 30 percent faster. And of course, Maxwell is Nvidia's last-gen part; the company says a comparable Pascal-based system would be 90 percent faster.

 

---

 

 nVidia made a blog post addressing Intel "skewed" benchmarking: https://blogs.nvidia.com/blog/2016/08/16/correcting-some-mistakes/

 

Quote

Correcting Intel’s Deep Learning Benchmark Mistakes

 

Benchmarks are an important tool for measuring performance, but in a rapidly evolving field it can be difficult to keep up with the state of the art. Recently Intel published some incorrect “facts” about their long promised Xeon Phi processors.

 

Few fields are moving faster right now than deep learning. Today’s neural networks are 6x deeper and more powerful than just a few years ago. There are new techniques in multi-GPU scaling that offer even faster training performance.

 

In addition, our architecture and software have improved neural network training time by over 10x in a year by moving from Kepler to Maxwell to today’s latest Pascal-based systems, like the DGX-1 with eight Tesla P100 GPUs.

For example, Intel recently published some out-of-date benchmarks to make three claims about deep learning performance with Knights Landing Xeon Phi processors:

• Xeon Phi is 2.3x faster in training than GPUs
• Xeon Phi offers 38% better scaling that GPUs across nodes
• Xeon Phi delivers strong scaling to 128 nodes while GPUs do not

We’d like to address these claims and correct some misperceptions that may arise.

 

Intel used Caffe AlexNet data that is 18 months old, comparing a system with four Maxwell GPUs to four Xeon Phi servers. With the more recent implementation of Caffe AlexNet, publicly available here, Intel would have discovered that the same system with four Maxwell GPUs delivers 30% faster training time than four Xeon Phi servers.

 

In fact, a system with four Pascal-based NVIDIA TITAN X GPUs trains 90% faster and a single NVIDIA DGX-1 is over 5x faster than four Xeon Phi servers.

Intel is comparing Caffe GoogleNet training performance on 32 Xeon Phi servers to 32 servers from Oak Ridge National Laboratory’s Titan supercomputer. Titan uses four-year-old GPUs (Tesla K20X) and an interconnect technology inherited from the prior Jaguar supercomputer. Xeon Phi results were based on recent interconnect technology.

 

Using more recent Maxwell GPUs and interconnect, Baidu has shown that their speech training workload scales almost linearly up to 128 GPUs.

 

---

 

Intel replied to the accusations:

Quote

It is completely understandable that Nvidia is concerned about Intel in this space. Intel routinely publishes performance claims based on publicly available solutions at the time, and we stand by our data.

in a sense, I think Intel is kinda' right

are nVidia's Pascal based Tesla accelerators widely available?

 

---

 

if there isn't a market for independent GPGPU benchmarking, there should be one

but the problem is: who's gonna afford these cards or entire ecosystems based around them - not a lot

 

---

 

related reading: Tom Forsyth on Larabee - Why didn't Larrabee fail?

Edited August 18, 2016 by zMeul

[Misanthrope]

Intel probably forgot that unlike AMD, they can't just beat up Nvidia without consequences. I hope a bloody battle (or at least a legal one) ensues. 

[zMeul]

3 minutes ago, Misanthrope said:

(or at least a legal one) ensues. 

legal battle, nah ... Intel can claim that at the time of the testing, Pascal based Teslas weren't available

moreover, the new Pascal based Titan X is a GTX card when they "market" it as a deep learning card - nVidia fucked up this one really good

 

 

what nVidia could do, and probably will, is to publish a bunch of benchmarks of their Pascal based GPUs against Xeon Phi

[4960X]

[samcool55]

They won't be able to do anything, look at the small text, it clearly says "Software and workloads used in peformance tests may have been optimized for performance only on intel microprocessors"

How i understand this is "this stuff might be biased and cherry picked" and nvidia can't do anything.

The note is there...

[GoodBytes]

It's "NVIDIA", not Nvidia, or nVidia, or NVdia, or nvidia or Nvidiadia. (from their PR stuff).

I know the company likes to change its name format, but I think their marketing team thinks that screaming the company name is best for brand recognition. lol!

 

 

 

 

[Curufinwe_wins]

Ehh, whatever either way.

[zMeul]

7 minutes ago, GoodBytes said:

It's "NVIDIA", not Nvidia, or nVidia, or NVdia, or nvidia or Nvidiadia.

I know the company likes to change its name format, but I think their marketing team things that screaming the company name: NVIDIA, is best for brand recognition.

MicroSoft

[Shahnewaz]

Top comment on Ars Technica:

Quote

Publishing biased benchmarks? Surely not. That definitely sounds like something Nvidia have *never* done

 

[GoodBytes]

7 minutes ago, zMeul said:

MicroSoft

I think they should rename themselves to MacroSoft, I mean Windows is fairly big.

[Citadelen]

1 hour ago, Misanthrope said:

-snip-

There were consequences for Intel, they had to pay AMD upwards of $1bn in a single case, there have been multiple cases against them in multiple countries.

[CtrlAltELITE]

1 hour ago, Shahnewaz said:

Top comment on Ars Technica:

Publishing biased benchmarks? Surely not. That definitely sounds like

something Nvidia have *never* done

 

They are all at it :D.

Nvidia turning up the tesselation dial wherever they can

AMD using that one game that is optimised with vulcan

Intel trying to save CPU based supercomputers

 

Hell its the same in every industry, German vehicle manufacturer efficiency numbers anyone?

[Notional]

Haha what is Nvidia going to do against Intel? Intel iGPUWorks!

[Haquer]

1 hour ago, GoodBytes said:

It's "NVIDIA", not Nvidia, or nVidia, or NVdia, or nvidia or Nvidiadia.

I know the company likes to change its name format, but I think their marketing team things that screaming the company name: NVIDIA, is best for brand recognition.

 

 

 

nViDia amD inTEl

[-BirdiE-]

Seems rather ironic that NVIDIA is accusing someone of shady marketing techniques....

 

 

GTX 1080FASTER THAN TWO 980s in SLI

 

 

 

 in VR

[zMeul]

2 minutes ago, -BirdiE- said:

Seems rather ironic that NVIDIA is accusing someone of shady marketing techniques....

 

GTX 1080FASTER THAN TWO 980s in SLI

 

 in VR

 

[-BirdiE-]

1 minute ago, zMeul said:

 

Sorry. I don't follow ... Are you noting that it shows a Paragon background, but none of the performance benchmarks are based on Paragon?

[zMeul]

Just now, -BirdiE- said:

Sorry. I don't follow ... Are you noting that it shows a Paragon background, but none of the performance benchmarks are based on Paragon?

it's marketing .. who cares ?!

the GTX1070 material has Mirror's Edge Catalyst in background:

[LAwLz]

2 hours ago, GoodBytes said:

It's "NVIDIA", not Nvidia, or nVidia, or NVdia, or nvidia or Nvidiadia. (from their PR stuff).

I know the company likes to change its name format, but I think their marketing team thinks that screaming the company name is best for brand recognition. lol!

Their logo spells it with a lower case n and the rest upper case.

If they wanted it to be NVIDIA then maybe they should have written it that way in their logo.

[-BirdiE-]

1 minute ago, zMeul said:

it's marketing .. who cares ?!

the GTX1070 material has Mirror's Edge Catalyst in background:

-snip-

I don't care.. I'm just saying it's ironic that NVIDIA is throwing a big stink about Intel doing this, when they employ very similar tactics...

[zMeul]

Just now, -BirdiE- said:

I don't care.. I'm just saying it's ironic that NVIDIA is throwing a big stink about Intel doing this, when they employ very similar tactics...

you were quoting me because of the image in the background

please don't take me out of context

[-BirdiE-]

Just now, zMeul said:

you were quoting me because of the image in the background

please don't take me out of context

I specifically said I didn't follow the point you were trying to make with that.... I was asking for clarification..

[zMeul]

6 minutes ago, -BirdiE- said:

I specifically said I didn't follow the point you were trying to make with that.... I was asking for clarification..

the image I posted should be enough to clarify what you "claimed" nVidia did

 

1st off, it wasn't 980Tis, it was 980

2ndly, VR is labeled appropriately and not a minuscule footnote as you make it look like

[-BirdiE-]

Just now, zMeul said:

the image I posted should be enough to clarify what you "claimed" nVidia did

 

1st off, it wasn't 980Tis, it was 980

2ndly, VR is labeled appropriately and not a minuscule footnote as you make it look like

Clearly I wasn't referencing that piece of marketing material....

 

It was the stuff during the announcement press conference, when they kept saying that the 1080 was twice as fast as 980s in SLI... And only after the conference did they specify that they meant in VR.

[zMeul]

Just now, -BirdiE- said:

Clearly I wasn't referencing that piece of marketing material....

 

It was the stuff during the announcement press conference, when they kept saying that the 1080 was twice as fast as 980s in SLI... And only after the conference did they specify that they meant in VR.

the writing was on the wall quite literally since that slide was shown off during the event - some people are more busy with their social media bullshit rather than paying attention to what's going on