Intel Delivers New Architecture for Discovery with Intel Xeon Phi Coprocessors

[Xzibit]

Originally Posted by HumanSmoke

Bullshit . Verifiable numbers or STFU.
Since you can't produce any K20 numbers I'll presume you'll carry on pulling numbers out of thin air to continue your trollathon.

Nvidia GK110 Whitepaper

Kepler GK110 will provide over 1 TFlop of double precision throughput with greater than 80% DGEMM
efficiency versus 60‐65% on the prior Fermi architecture.

now STFU

I bet you cling to the vagueness of it all.

[Steevo]

Just used the provided link in the other article. No need to be caustic.


Interesting to do the math on cores/score for some of the systems.

[HumanSmoke]

Originally Posted by Steevo

Just used the provided link in the other article. No need to be caustic.

Sometimes it gets hard to tell motivation when trolls like Xzibit and Co. are doing their thing. Case in point is the post directly above yours. A whitepaper put out shortly after GK110 taped out, and the phrase "greater than" equating to "equal" in the trolls book constitute its argument- the same troll who discounts one vendors figures while relying upon a competitors PR blurb, and seemingly bolstered by Intel PR via your post. What's a guy to think?

Originally Posted by Steevo

Interesting to do the math on cores/score for some of the systems.

If it comes down to core efficiency then its a no contest- Power7 pretty much wins all. Every GPGPU computer in existence relies on x86 CPU's, and x86 CPU's pretty much suck in relation to GPU's for double precision workloads. Since IBM don't use GPU, it would seem logical that pairing the most efficient GPU (K20X) with the most efficient double precision capable x86 (Xeon E5) would yield the most efficient big iron.

@Xzibit
Maybe you try reading from your own quote

Kepler GK110 will provide over 1 TFlop of double precision throughput with greater than 80% DGEMM

Last time I checked with my common sense, "Greater than 80%" is larger than the 80% you claimed in your post. Date of that whitepaper...15th May 2012

[Xzibit]

Originally Posted by HumanSmoke

@Xzibit
Maybe you try reading from your own quote


Last time I checked with my common sense, "Greater than 80%" is larger than the 80% you claimed in your post. Date of that whitepaper...15th May 2012

Go complain to Nvidia about.

If you have a link to solid numbers provide them or follow your own advice

Originally Posted by HumanSmoke

Verifiable numbers or STFU.

[Cortex]

If Larrabee succeeded now we would have (at least) 48 x86 cores @ ~800GFLOPS graphics card, for $500-600.

Good efficiency for GCN but still no Fire Pro S supercomputers . There are two Radeon powered though


http://i.top500.org/system/177430
http://i.top500.org/system/177154

[Steevo]

ATI/AMD has always done a poor job of following through with the software to backup their hardware. Are we surprised? Nope.


VLIW & GCN I am guessing are going to be remarried at some point in the future, as VLIW can do more heavy lifting, but requires more overhead, but can be programmed to do more than GCN, but GCN is easier to program for and execute on.

[HumanSmoke]

The SC lists are out

Top500

and

Green500
1. Beacon (Intel Xeon + Xeon Phi.....#253 overall)
2. SANAM (Intel Xeon + AMD S10000...#52 overall)
3. Titan (AMD Opteron + Nvidia K20X...#1 overall)
4. Todi (AMD Opteron + Nvidia K20....#91 overall)
5 thru 29 IBM BlueGene/Q systems