Re: Openssl on Itanium
Diarmuid Oneill wrote: > I have found out that the 0.9.6 distributions of OSSL do not include > Itanium assembly implementations for much (maybe any Itanium assembly > at all, I didn't look) of the CPU intensive operations, including > RSA/ModExp. So OSSL uses the C routines which are, to say the least, > not optimum. > It's also worth remebering that the Itanium uses EPIC(explicitly > parallel instruction computing) and that in order to see the > performance it's capable of it must be programmed appropriately. It > has been suggested that if there were optimised assmebly routines for > the Itanium it would certainly beat the P3 Mhz/Mhz. > Anyone care to comment on this? If you use GCC, the IA-64 target *IS* an optimizing compiler that can re-order instruction for EPIC. Or weren't you aware of the whole concept of EPIC, compiler-based optimization? -- TheBS -- Bryan "TheBS" Smith mailto:[EMAIL PROTECTED]chat:thebs413 Engineer AbsoluteValue Systems, Inc. http://www.linux-wlan.org PresidentSmithConcepts, Inc.http://www.SmithConcepts.com __ OpenSSL Project http://www.openssl.org User Support Mailing List[EMAIL PROTECTED] Automated List Manager [EMAIL PROTECTED]
Re: Openssl on Itanium
Sorry should have had subject Re OpenSSL on itanium Hi, I have done a bit of research into this topic and some of my findings directly oppose what was said about the itanium not matching the P3 Mhz to Mhz. I have found out that the 0.9.6 distributions of OSSL do not include Itanium assembly implementations for much (maybe any Itanium assembly at all, I didn't look) of the CPU intensive operations, including RSA/ModExp. So OSSL uses the C routines which are, to say the least, not optimum. It's also worth remebering that the Itanium uses EPIC(explicitly parallel instruction computing) and that in order to see the performance it's capable of it must be programmed appropriately. It has been suggested that if there were optimised assmebly routines for the Itanium it would certainly beat the P3 Mhz/Mhz. Anyone care to comment on this? Thanks, Diarmuid Get your FREE download of MSN Explorer at http://explorer.msn.com __ OpenSSL Project http://www.openssl.org User Support Mailing List[EMAIL PROTECTED] Automated List Manager [EMAIL PROTECTED]
Re: OpenSSL on itanium [going offtopic]
On Tue, 21 Aug 2001, chirs charter wrote: > Nice observations. The alpha is gone now? When did DEC > discontinue it? DEC was discontinued. Its corpse was dismembered and sold to various companies, and Compaq got most of the silicon designs (including the aXp and the DS21x4x "Tulip" Ethernet chip) after the manufacturing facilities were sold to Intel.* Now Compaq has sold the processor design itself to Intel, prompting fears that it will die as soon as existing contracts expire. Maybe at least Intel will finally learn something about computer organization by studying it. -- * Cabletron got most of the networking gear, Quantum acquired the disk and tape drives, and Oracle got RDB. Compaq also got the VAX and Alpha gear and the StorageWorks unit. Some outfit I'd never heard of got the terminals and printers unit and was still making VT5xx last I heard. Anybody know whatever happened to the "Dragon" graphics chip? -- Mark H. Wood, Lead System Programmer [EMAIL PROTECTED] Make a good day. __ OpenSSL Project http://www.openssl.org User Support Mailing List[EMAIL PROTECTED] Automated List Manager [EMAIL PROTECTED]
[Off-topic] Re: Alpha history -- WAS: OpenSSL on itanium
[ I guess we're getting off-topic here. I'll make one more go of it for "informational purposes" (I have a EE and worked at an IC design firm for 2 years), but let's take it _off-list_ after this. Please reply off-list. ] chirs charter wrote: > Nice observations. The alpha is gone now? Basically. This was the chronology (with "extra nuggets" for "completeness" which goes a long way to explain things): '90-91: - Digital and IBM release several MIPS R2000/3000-based workstations - Chip vendors consider unifying around the MIPS RISC processor - First Sun bails out (sticks with SPARC RISC approach) - IBM-Moto start talking (becomes PowerPC) - Digital leaves to start RISC-"anal" 64-bit processor VAX replacement - Digital beings talking to ARM about licensing their microcontroller IP '91-92: - Intel takes an interest in Digital's new truly 64-bit "Alpha" project which is ultra-liberal approach to RISC design - Digital beings developing the "StrongARM" microcontroller '92-93: - Intel drops Alpha interest, decides to stick with current GTL+/P6 development (which became the Pentium Pro) - Alpha releases first EV5 (64-bit wide, 66MHz), 64-bit processor, the Alpha 21064, a chip that has only 32-bit and 64-bit instructions -- not even a 8 or 16-bit load/store operation (again, _very_liberally_RISC_)! - NT/Alpha released for x86, Alpha MIPS, PowerPC '93-95: - Other vendors release their 64-bit processors - Intel releases GTL+ bus, 32-bit Pentium Pro - Digital follows up with Alpha EV6 (upto 128-bit wide, 66-100MHz) 21164 (which finally adds 8 and 16-bit load/store operations) at 333MHz while many other chips are barely passing 100MHz -- EV6 also introduces a 16-node, point-to-point bus architecture - Samsung becomes first Alpha licensee and starts fabricating their own Alpha chips - Microsoft NT 3.1 released with x86, Alpha, MIPS and PowerPC support. First Linux port to Alpha released. '95-96: - Digital begins development of Alpha EV7 (upto 128-bit wide, 166/333MHz DDR/effective "reduced EMF") FSB, as well as the same 16-node, point-to-point bus archicture - AMD licenses EV7 bus for forthcoming K7 processor - Digital adds "AV" instructions to new, low-cost 21164PC (21164 w/o the L2 cache) to match MMX (a whole _5_ instructions, that's it!) - Digital also releases FX!32 which is both an emulator and binary, run-time "translator" for running/translating NT/Intel programs to NT/Alpha (and wins numerous awards) - Microsoft cancels NT/MIPS and NT/PowerPC ports leaving only NT/Alpha - Truly 64-bit Linux/Alpha released as GNU/GCC/GLibC are made completely 64-bit capable, SPARC, MIPS and other processors follow as Linux 1.3 (2.0 beta) is developed as a cross-platform OS. '96: - Digital throws lawsuit on Intel for violation of Alpha patents - Claims Intel misused private, but unlicensed information, disclosed during Alpha collaboration of 91-92 - Alpha reaches 500MHz while all other chips are sub-200MHz, SPECfp95 is 3x most other chips of the time, and bests everyone in SPECint95 as well - Digital also sells of networking division (15 of 18 top selling 10/100 NICs are Digital 21140-series "Tulip" based, infamous "Tulip" fragmentation results and continues through today ;-) - Intel beings IA-64 EPIC/Predication design. - StrongARM starts appearing in embedded boards and designs at 133MHz, blows away most other microcontrollers in performance '97: - Digital and Intel settle out-of-court -- Intel buys Digital fabs and licenses both the Alpha and, more importantly (but not well covered by the media) StrongARM (as an i960 replacement) - Samsung and Intel become sole Alpha producers - Alpha Processor, Inc. (API) formed to continue development and oversee support, as well as continue Digital IC interests (like PCI/AGP bridge chips and other interconnect chips) - Intel Pentium II released - Digital releases FX!32 equivalent (cannot remember name) for Linux/Intel to Linux/Alpha binary emluation/translation '98: - Intel Pentium II gains clock speed - Alpha 164 processors still stuck at 500MHz (for over 2 years!) - Alpha 264 delayed until late 98 - Despite Intel's promise of lower fab cost in producting the Alpha, prices don't shrink - Most Alpha chips are still fabbed at 0.35-0.5um while other chips shrink to 0.25um sizes - Digital releases 64-bit version of NT/Alpha 4.0, never publicly made available by Microsoft - AMD begins x86-64 project with first product codename "Sledgehammer." - API beings EV8/364 design which is a 100M transistor Alpha with both run-time optimization and IA-64-like EPIC compiler-optimization (whereas the IA-64 only has later) '99: - Digital sells to Compaq, largely for the MCSE staff (to become largest professional support provider), but also gains Alpha - Compaq promises to support Tru64 and VMS for at least another 4.5 years - NT/Alpha port killed - Max exodus of Digital/API design engineers to AMD - Max exodus of Digital/NT developers to Microsoft - Initial non-FC-PGA Pentium III app
RE: OpenSSL on itanium
You know that DEC's been discontinued (bought by Compaq)? I read that Compaq is selling (sold?) the Alpha to Intel right now. um = micrometer (millionth of a meter) which is the track width of the microprocessor. I thought 0.15um was state of the art, but it seems that it's now 0.13um. 0.35um is older technology. Regards, Steven -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of chirs charter Sent: Wednesday, 22 August 2001 9:44 AM To: [EMAIL PROTECTED] Subject: Re: OpenSSL on itanium Nice observations. The alpha is gone now? When did DEC discontinue it? Lastly in the measurement what does "um" stand for? Thanks --- Bryan-TheBS-Smith <[EMAIL PROTECTED]> wrote: > Diarmuid Oneill wrote: > > When I download and build OpenSSL (which works > fine!) and run the > > openssl speed rsa1024 tests, I get around 68 rsa > signings/sec. When I > > run this on a 4 CPU (700Mhz) P3 machine I get > around 103 private rsa > > signings/sec. I understand that the test is > running on 1 cpu only but > > that's the case for both machines. > > It looks like most of the functions are integer. > Itanium is slower, MHz > for MHz, than just about any x86 Pro+ processor at > integer (even using > optimized code). Only at floating point does > Itanium do about 2x a P3, > MHz for MHz (and the P4 is slower than the P3, MHz > for MHz, unless you > use "lossy"/interpolated SSE instructions). > > -- TheBS > > P.S. It's sad to see a 3-year old design at 0.35um, > the Alpha 264 > 667MHz/4MB, can toast the 0.13um Itanium 733MHz/4MB > at floating point. > Too bad Alpha is gone now. > > -- > Bryan "TheBS" Smith mailto:[EMAIL PROTECTED] > chat:thebs413 > Engineer AbsoluteValue Systems, Inc. > http://www.linux-wlan.org > PresidentSmithConcepts, Inc. > http://www.SmithConcepts.com > __ > OpenSSL Project > http://www.openssl.org > User Support Mailing List > [EMAIL PROTECTED] > Automated List Manager [EMAIL PROTECTED] __ Do You Yahoo!? Make international calls for as low as $.04/minute with Yahoo! Messenger http://phonecard.yahoo.com/ __ OpenSSL Project http://www.openssl.org User Support Mailing List[EMAIL PROTECTED] Automated List Manager [EMAIL PROTECTED] __ OpenSSL Project http://www.openssl.org User Support Mailing List[EMAIL PROTECTED] Automated List Manager [EMAIL PROTECTED]
Re: OpenSSL on itanium
Nice observations. The alpha is gone now? When did DEC discontinue it? Lastly in the measurement what does "um" stand for? Thanks --- Bryan-TheBS-Smith <[EMAIL PROTECTED]> wrote: > Diarmuid Oneill wrote: > > When I download and build OpenSSL (which works > fine!) and run the > > openssl speed rsa1024 tests, I get around 68 rsa > signings/sec. When I > > run this on a 4 CPU (700Mhz) P3 machine I get > around 103 private rsa > > signings/sec. I understand that the test is > running on 1 cpu only but > > that's the case for both machines. > > It looks like most of the functions are integer. > Itanium is slower, MHz > for MHz, than just about any x86 Pro+ processor at > integer (even using > optimized code). Only at floating point does > Itanium do about 2x a P3, > MHz for MHz (and the P4 is slower than the P3, MHz > for MHz, unless you > use "lossy"/interpolated SSE instructions). > > -- TheBS > > P.S. It's sad to see a 3-year old design at 0.35um, > the Alpha 264 > 667MHz/4MB, can toast the 0.13um Itanium 733MHz/4MB > at floating point. > Too bad Alpha is gone now. > > -- > Bryan "TheBS" Smith mailto:[EMAIL PROTECTED] > chat:thebs413 > Engineer AbsoluteValue Systems, Inc. > http://www.linux-wlan.org > PresidentSmithConcepts, Inc. > http://www.SmithConcepts.com > __ > OpenSSL Project > http://www.openssl.org > User Support Mailing List > [EMAIL PROTECTED] > Automated List Manager [EMAIL PROTECTED] __ Do You Yahoo!? Make international calls for as low as $.04/minute with Yahoo! Messenger http://phonecard.yahoo.com/ __ OpenSSL Project http://www.openssl.org User Support Mailing List[EMAIL PROTECTED] Automated List Manager [EMAIL PROTECTED]