1. CryptoGraphics: Cryptography using Graphics Processing Units Bachir Babale CSEPtu 590 3/8/2006

2. 3/9/2006Bachir Babale2 Outline Motivations GPU Operations Stream Ciphers Block Ciphers (omitted) Building a Secure System PVP-UAB Conclusion

3. 3/9/2006Bachir Babale3 Motivations New CPU performance model: parallelism GPUs are already highly parallel 24 fragment shading pipes 100s of hardware threads 150 Giga FLOPS (Nvidia’s G70)

4. 3/9/2006Bachir Babale4 Motivations (2) Decrypting inside the GPU would allow secure visualization application

5. 3/9/2006Bachir Babale5 GPU Operations 32 bit pixels processed as floating points RGB or RGBA formats in front/back buffer Operations used for cryptography: Copying pixels between coordinates Color maps Logical XOR Slowest operations in GPU No modular arithmetic for large integers

6. 3/9/2006Bachir Babale6 Stream Ciphers Synchronous Binary Additive Stream Cipher Stream of pseudo- random digits XOR plaintext/ciphertext Algorithm for GPU Pre-compute keystream Load into GPU memory Load ciphertext into memory area with XOR enabled Benefits of parallel processing diminished if data is too small XOR rate of 105.0MB/s for a 600x600 area (75.5% of CPU)

7. 3/9/2006Bachir Babale7 Block Ciphers (omitted) Bit level operations (shifts, rotates) not available No 32 bit data structure AES representation in GPU AES-GL rate is 50% of AES-C1, 2.4% of AES-C2 CPU usage for AES-GL is still 100%

8. 3/9/2006Bachir Babale8 Building a secure system Server, Proxy, Client over an untrusted network GPU contains pre-installed certificate The data remains encrypted until in the GPU Dynamic session key -> Need for authentication Close to ideal 24fps for videoconferencing 500 updates/s for thin client application

9. 3/9/2006Bachir Babale9 PVP-UAB Content Output Path Protected Video Path - User Accessible Bus Encrypts premium content on the bus to prevent hardware snoop hacks OS must verify valid subsystem Unique keys A challenge that leverages the complex arrangement of a large number of gates and a complex state model in the GPU 128 bit AES MPG2 and Windows Media 9 compression 2048 bit Diffie Hellman

10. 3/9/2006Bachir Babale10 Conclusion GPUs can be used to offload the CPUs to apply stream ciphers on large segments GPU can serve as the base for a trusted computing visualization platform PVP-UAB brings authentication and compression Need to develop ciphers that can better exploit the capabilities of modern GPUs

11. 3/9/2006Bachir Babale11 References CryptoGraphics: Secret Key Cryptography Using Graphics Cards Debra L. Cook., John Ioannidis., Angelos D. Keromytis., Jake Luck http://www1.cs.columbia.edu/~dcook/pubs/CTRSA-corrected.pdf Remotely Keyed Cryptographics. Secure Remote Display Access Using. (Mostly) Untrusted Hardware Debra L. Cook., Ricardo Baratto., Angelos D. Keromytis http://www.ncl.cs.columbia.edu/publications/icics2005.pdf Protected Interactive 3D Graphics Via Remote Rendering David Koller, Michael Turitzin, Marc Levoy), Marco Tarini, Giuseppe Croccia http://graphics.stanford.edu/papers/protected/protected.pdf GPGPU IEEE Visualization 2005 TUTORIAL http://www.gpgpu.org/vis2005/http://www.gpgpu.org/vis2005/ (03/05/2006) Output Content Protection and Windows Longhorn http://www.microsoft.com/whdc/device/stream/output_protect.msp xhttp://www.microsoft.com/whdc/device/stream/output_protect.msp x (03/05/2006)