Okay, I'm pissed at Intel.

[Arachnotronic]

So take a look at this article: http://newsroom.intel.com/community...neration-high-performance-computing-platforms

They claim that the Xeon E5 delivers up to "2.1 times" raw FLOPs in linpack and up to 70% more performance in real-HPC workloads.

Seems nice, right?

Yeah. 'till you read the fine print:

2S Xeon E5 score of 342.7 based on Intel internal measurements as of 7 September 2011 using an Intel Rose City platform with two Intel® Xeon® processor E5, Turbo Enabled, EIST Enabled, Hyper-Threading Enabled, 64 GB memory (8 x 8GB DDR3-1600), Red Hat* Enterprise Linux Server 6.1 beta for x86_6

Intel Tylersburg-EP platform with two Intel® Xeon® Processor X5690# (6-Core, 3.46GHz, 12MB L3 cache, 6.4GT/s, B1-stepping), EIST Enabled, Turbo Boost enabled, Hyper-Threading Disabled, 48GB memory (12x 4GB DDR3-1333 REG ECC), 160GB SATA 7200RPM HDD, Red Hat* Enterprise Linux Server 5.5 for x86_64 with kernel 2.6.35.10. Source: Intel internal testing as of Apr 2011. Score : 159.40 Gflops.


LAME.

 

[Edrick]

That is very lame. Although HT only supplies a 30%? increase at best? I guess 2.1 times sounds better than 1.7 times (or whatever it would turn out to be).

 

[IntelUser2000]

Chill out man. Remember in Linpack, Hyperthreading is a minor detriment, and that's what they disabled in the details. You know the LinX benchmarks we run and with Hyperthreading its quite a bit lower in score? Well, LinX is Linpack with a different name.

 

[Idontcare]

They claim that the Xeon E5 delivers up to "2.1 times" raw FLOPs in linpack and up to 70% more performance in real-HPC workloads.

The real dissapointment is right there, front and center.

"up to" means "absolutely everything else we threw at it performed worse than this one niche corner case, but we'll reference it anyways in our PR just so we set everyone's expectations unrealistically high".

These are 3-sigma observations, the one benchmark result that fell way outside the distribution.

Divide by 3, bring it back to a 1-sigma reality, and then you'll have a better, more realistic, expectation of the performance boost that is coming down the pipe. de-hype the hype 😀 😛

 

[IntelUser2000]

The real dissapointment is right there, front and center.

http://hothardware.com/News/Intel-Announces-New-Xeon-E5-Series-Talks-Up-Knights-Corner-MIC-Products/

The first slide shows the improvements. The bigger gain ones are likely tied to memory bandwidth increases, which should be quite large over the Westmere EP chips, more than 3 to 4 channels indicate. In HPC, along with the extra cores, will get quite a nice gain.

HPC is also one of the few scenarios where AMD is very competitive.

 

[Idontcare]

http://hothardware.com/News/Intel-Announces-New-Xeon-E5-Series-Talks-Up-Knights-Corner-MIC-Products/

The first slide shows the improvements. The bigger gain ones are likely tied to memory bandwidth increases, which should be quite large over the Westmere EP chips, more than 3 to 4 channels indicate. In HPC, along with the extra cores, will get quite a nice gain.

HPC is also one of the few scenarios where AMD is very competitive.

This graphic is pretty cool!

 

[Nemesis 1]

So IDC does this mean your going to buy knights corner for bragging rights? lol Look at the compute power. 1` knights corner= 9298 pentium xeon

 

[Rifter]

So IDC does this mean your going to buy knights corner for bragging rights? lol Look at the compute power. 1` knights corner= 9298 pentium xeon

That is insane when you think about it, thats a short time for that much of an improvement.

 

[tweakboy]

Wow dual XEON cpu's thats 16 core plus HT for 32 cores and 64GB RAM....Suddenly Sandy doesn't sound as impactful. gl

 

[Arachnotronic]

I'd love to play with an Intel MIC board. Program the hell outta it!

Wait...HT negatively impacts Linpack? Well, shucks. I'll be back...gotta test this for myself.

 

[Arachnotronic]

Well...son of a gun...huge performance increase on my 4GHz 980X (72.3 GFLOPs to 83.8 GFLOPS) without HT. Okay, then new question...why wouldn't Intel disable HT on the 8 core SNB Xeon?

 

[aigomorla]

im wondering how long it takes people to realize even more they are comparing a 16 core machine... top the E5.. i believe they are 8 cores..
(Intel's Xeon E5 chips will support between four and eight computing cores with TDPs ranging from 50W to 130W,)

unless they used the quadcores... but i doubt that... they had to use the octocores...
vs.

Tylersberg.. which is a 12core machine... lol...

OMG did the numbnuts at intel hier the idiots at AMD Marketing?
Please say no....

 

[MisterMac]

im wondering how long it takes people to realize even more they are comparing a 16 core machine... top the E5.. i believe they are 8 cores..
(Intel's Xeon E5 chips will support between four and eight computing cores with TDPs ranging from 50W to 130W,)

unless they used the quadcores... but i doubt that... they had to use the octocores...
vs.

Tylersberg.. which is a 12core machine... lol...

OMG did the numbnuts at intel hier the idiots at AMD Marketing?
Please say no....

X5690 is a 6 CORE machine, with 6 virtual cores.

Both setups have 2 of their respective processors in them.

Leaving 12 cores and 24 threads.


One would assume that intel would test A E5 2600 series with the same ammount of cores.
aka 6.


They could do 8 vs 6, but lets face it - they don't need to.

What will you say if the test results turn out to be with the Quadcores and not hexcores? or octocores?

 

[greenhawk]

LAME.

if you think that was lame, you should have seen the hardware comparisons being done back 10 years or more.

intel machine given more ram, better video card faster HDD vs AMDs with less ram than a standard machine, the $5 video card and a crap drive, then the reviewer would give intel the thumbs up.

All you can do is take it as marketing (and the crap those people dribble) and ignore it. You can not change it and you can not tell people any differently (ie: the review is from intel, it must be the truth).

 

[Idontcare]

I'd love to play with an Intel MIC board. Program the hell outta it!

Wait...HT negatively impacts Linpack? Well, shucks. I'll be back...gotta test this for myself.

It's true.

 

[IntelUser2000]

im wondering how long it takes people to realize even more they are comparing a 16 core machine... top the E5.. i believe they are 8 cores..
(Intel's Xeon E5 chips will support between four and eight computing cores with TDPs ranging from 50W to 130W,)

I don't see the problem there. It's not like it went to 33% more cores then became slower.

The E5 chip tested is an 8 core, which is partially the reason its faster. It doesn't matter anyway, the target market can use more cores. Plus the pricing is similar between this E5 part and the X5690.

2.7GHz(It looks like from Linpack it runs at 2.8GHz with Turbo) with 8 Sandy Bridge cores

versus

3.46GHz(~3.60GHz from Linpack) with 6 Westmere cores

Ideally, it would look like this, assuming Sandy Bridge having 15% IPC advantage.

(2.8GHz/3.46GHz)*(8 cores/6 cores)*1.15 IPC modifier = 24%

 

[aigomorla]

What will you say if the test results turn out to be with the Quadcores and not hexcores? or octocores?

then i would say the X5690 got spanked hard.

but intel typically never does that... pit a new gen lower tier processor with last high tier...
i would assume they kept the same tier lot, saying the X5690 is the best u can get for that platform, so i would assume the highest tier'd E5

 

[jhu]

That is insane when you think about it, thats a short time for that much of an improvement.

It took about that time for DEC Alpha to reach Cray-XMP-level processing power on a chip.

 

[IntelUser2000]

I guess the better question Intel17 should have asked is if they have disabled Hyperthreading on the X5690 but not on the E5 for actual applications they tested. Well here's the notes for the HPC benchmarks they quoted:

For the HPC suite:

Baseline:

2S Intel® Xeon® X5690 HPC suite geometric mean of application measurements by vertical (CAE, Energy, FSI, Life Science, NWS), actual performance will vary by workload. Based on Intel internal measurements as of October 2011 using an ntel® Xeon® 5600 processor platform with two Intel® Xeon® X5690, Turbo Enabled, Best Hyper-Threading configuration, 48GB DDR3-1333 memory, Red Hat* EL5-U5.​

New Configuration: 2S Intel® Xeon® E5-2680 HPC suite geometric mean of application measurements by vertical (CAE, Energy, FSI, Life Science, NWS), actual performance will vary by workload. Based on Intel internal measurements as of October 2011 using an Intel® Canoe Pass platform with two Xeon® E5-2680 (C0 step), Turbo Enabled, Best Hyper-Threading configuration, 32 GB DDR3-1600 memory, Red Hat* Enterprise Linux 6.1, 2.6.39.3 kernel

[/QUOTE]

Now I don't know why its different for the Linpack one. But the numbers say neither the X5690 with disabled Hyperthreading nor E5 with enabled one is getting impacted much.​

2.7GHz x 8 cores x 2 sockets/core x 8 DP Flops/cycle = 345.6 GFlops
3.46GHz x 6 cores x 2 sockets/core x 4 DP Flops/cycle = 166.08 GFlops​