What's the catch?

Intel17 · 05-16-10, 09:14 PM

Okay, so help me understand this:

If a top end, $1000 Intel CPU can only do, say, 100 GFLOPS but a $500 nVidia or AMD graphics board can do >500 GFLOPS (in dp) and >1TFLOP (in sp), why is anything computationally expensive (compression, physics simulation, CADD, etc.) done on a CPU any more? What is the catch? Why haven't GPUs basically taken over with CPUs doing minimal work?

trivium nate · 05-16-10, 10:21 PM

english? thanks

**snowmanwithahat** · 05-16-10, 10:38 PM

1. Not all problems can be parallelized to the degree of benefit from gpgpu computing
2. So far the language has been proprietary. CUDA, Stream, from nvidia and ATi respecitvely. OpenCL stands to change that.

Basically only a small subset of the computing problems can leverage this technology. The ones that can have been, but development is hard and slow because of the fact that no commercial developer, and most open-source coders for that matter, don't want to develop for a shifting target as far as language is concerned.

Highly parallel problem solving is really only good in.

Graphics
Physics simulation
Encryption (sometimes)
Molecular simulation (sometimes)

TLDR; It has little impact on programs used by consumers, so don't expect to see much anytime soon.

Ninja Prime · 05-16-10, 10:39 PM

Because GPUs can't run general purpose code, essentially. To really explain it more than that would take longer than I want to type.

05-16-10, 09:14 PM	#1
Intel17 Is not an Intel fanboi Join Date: May 2004 Location: Burlington, VT Posts: 1,368	What's the catch? Okay, so help me understand this: If a top end, $1000 Intel CPU can only do, say, 100 GFLOPS but a $500 nVidia or AMD graphics board can do >500 GFLOPS (in dp) and >1TFLOP (in sp), why is anything computationally expensive (compression, physics simulation, CADD, etc.) done on a CPU any more? What is the catch? Why haven't GPUs basically taken over with CPUs doing minimal work? __________________ Intel Core i7 980X Extreme Edition @ 4.00GHz, 2.94GHz uncore\| Corsair H60 \| ASUS P6X58D-E \| Zotac GTX 580 @ 872/1744/2145\|Western Digital Caviar Black 1.5TB (Games/Steam)\| Seagate 1.5TB (media/misc.) \|Seagate 1 TB (OS/Main) \| 24 GB G.Skill RipJaws X 1.5V DDR3-1333 \| Creative X-Fi Titanium \| Sennheiser HD 202 Headset \| Dual Gigabit Realtek LAN \| Antec True Power Quattro 1200W \| Acer G235Hbmbd 23" LCD Display \| Logitech K320 Wireless Keyboard \| Logitech Performance MX Mouse

05-16-10, 10:21 PM	#2
trivium nate Registered User Join Date: Dec 2007 Posts: 4,023	Re: What's the catch? english? thanks __________________ CM-HAF-932//AMD Phenom-II-X4-965-BE-Deneb-3.4 GHz-Quad Core//CorsairH50 ASRock 770 EXTREME3 Mobo//EVGA GTX-480(1536MB GDDR5)-Arctic Cooling Accelero Xtreme+cooler 180GB SSD//3TB//G.SKILL Ripjaws X Series 8GB//DVD-RW//blu-Ray//1000 Watt Corsair PSU 42"Vizio TV-1920x1080-1080P-120Hertz//Win 7 Ultimate(64)//XBox360-pc controller Saitek Eclipse 2 keyboard//Razer DeathAdder Mouse//250gb-Xbox 360//160gb-PS3

05-16-10, 10:38 PM	#3
snowmanwithahat Snowy Join Date: Jul 2004 Location: Michigan Posts: 973	Re: What's the catch? 1. Not all problems can be parallelized to the degree of benefit from gpgpu computing 2. So far the language has been proprietary. CUDA, Stream, from nvidia and ATi respecitvely. OpenCL stands to change that. Basically only a small subset of the computing problems can leverage this technology. The ones that can have been, but development is hard and slow because of the fact that no commercial developer, and most open-source coders for that matter, don't want to develop for a shifting target as far as language is concerned. Highly parallel problem solving is really only good in. Graphics Physics simulation Encryption (sometimes) Molecular simulation (sometimes) TLDR; It has little impact on programs used by consumers, so don't expect to see much anytime soon. __________________ ---Gaming Rig--- Q6600 3.4ghz (378x9) - 1.5v Gigabyte EP45-UD3P 8gb (4x2gb) OCZ Gold DDR2-800 (5-4-4-12) MSI+ASUS GTX 470 SLI Dell u3011 IPS Display HP 22" Auxiliary Monitor 256gb Western Digital Silicon Edge Blue SSD 5x2tb RAID-5 Array 750W PC P&C PSU Windows 7 Pro 64-Bit & Ubuntu 10.04 64-bit ---Gaming Laptop--- ASUS G53JW Core i7 740QM 16gb DDR3 Nvidia GTX 460m 1tb WD HDD 120gb Corsair SSD Join the NvNews Folding @ Home Team

05-16-10, 10:39 PM	#4
Ninja Prime Registered User Join Date: Mar 2003 Location: Utah Posts: 2,263	Re: What's the catch? Because GPUs can't run general purpose code, essentially. To really explain it more than that would take longer than I want to type.