Overclock.net › Forums › General Hardware › General Processor Discussions › Why can't Intel beat NVIDIA with Xeon Phi on FLOPS?
New Posts  All Forums:Forum Nav:

Why can't Intel beat NVIDIA with Xeon Phi on FLOPS? - Page 3

post #21 of 29
Quote:
Originally Posted by maarten12100 View Post

What is the same(minor): interface, GDDR5, parallel cores
What is different(major): x87(x86 based arch), non host dependent(runs its own OS), calculates with extreme precision compared to a gpu(ECC and a x86 like arch), can be programmed within a short time frame for specific code, vastly less but bigger cores.

It is not nor was it ever going to be a gpu if they wanted that they should've gone ASIC based instead of using their old arch and optimizing the hell out of it.

It was definitely intended to be a GPU. They bought out a shop that had developed a software based x86 renderer. Those guys were intended to program the driver that used the tiny x86 cores on the card to provide Direct3D in Windows and stuff.

This was the software: http://www.radgametools.com/pixomain.htm
post #22 of 29
Quote:
Originally Posted by deepor View Post

It was definitely intended to be a GPU. They bought out a shop that had developed a software based x86 renderer. Those guys were intended to program the driver that used the tiny x86 cores on the card to provide Direct3D in Windows and stuff.

This was the software: http://www.radgametools.com/pixomain.htm
We are talking the Xeon Phi not Intel's MIC fiasco.
The difference is:one is a finished product the other is a wafer shown in a presentation giving false hopes.
Using a cpu based arch for rendering is almost as stupid as using an FPGA to do it.
post #23 of 29
MIC is this current Xeon Phi, not the graphics part. No idea what that was named, just that it had the code name Larrabee.

It's just fun to think it might have been originally not intended for computing and the current product might just be a lucky mistake. They had probably done the work for the chip and for a PCI-E card, all the VRM stuff around it and cooling, etc. Then NVidia came out with Tesla and they just jumped on that idea as their graphics performance sucked.
post #24 of 29
Thread Starter 
I've no idea what's the point of beating around the bush. Its main job is to use OpenCL type software and it's highly parallel. It is fundamentally a very close relative of GPUs.

But that's not the point of the thread in the first place. I was wondering why Intel can't beat AMD or NVIDIA in GPUs. And I still suspect it's not their "low skill" or IP.
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
post #25 of 29
Quote:
Originally Posted by fateswarm View Post

I've no idea what's the point of beating around the bush. Its main job is to use OpenCL type software and it's highly parallel. It is fundamentally a very close relative of GPUs.

But that's not the point of the thread in the first place. I was wondering why Intel can't beat AMD or NVIDIA in GPUs. And I still suspect it's not their "low skill" or IP.
it is not it is to port existing code(in a smaller time frame) and keep the high precision of a cpu.
post #26 of 29
Thread Starter 
What are you on about? Its main job is not to port old code.
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
post #27 of 29
"Old code" means here, you can use Fortran, C++, etc. Code used by programs that are running on normal CPUs. A shop that already has code like this, code that's already parallelized to make use of normal multicore CPUs, doesn't have to replace their existing stuff with OpenCL to use this card. It can run on the bunch of tiny x86 cores it has. For Tesla, you'd have to program new code.
post #28 of 29
Thread Starter 
And something tells me that's not Intel's main intent.
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
4790K 4.6G 1.24v/1.74v, 5.1G 1.35v validation. GA-Z97X-Gaming 7 Tri-X R9 290 1100/1350 +0.012v G.Skill 2400 c10 
CoolingMonitorPowerCase
Noctua NH-D15 24EA53 IPS 76Hz OC EVGA 1000 G2 Phanteks Enthoo Pro 
Audio
SoundMagic E10 
  hide details  
Reply
post #29 of 29
Quote:
Originally Posted by fateswarm View Post

And something tells me that's not Intel's main intent.
Those datacenters and the Scale MP deal make a boatload of money but there is a slight possibility it wasn't Intel's original intention.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: General Processor Discussions
Overclock.net › Forums › General Hardware › General Processor Discussions › Why can't Intel beat NVIDIA with Xeon Phi on FLOPS?