SSE64 or 3DNow64?

HDTVMan

Banned
Apr 28, 2005
1,534
0
0
Ok We have SSE3 in 32bit and 3dnow is 32bit.

Does either of these exist in 64bit form? If we migrate to windows 64 is the slate clean and were starting from just base 64 bit instructions with no SSE or 3Dnow?

Does either side plan on on bringing enhances instructions in the near future?
To me AMD should be thinking about that because you know Intel is.

Some instructions to help specific processes along. I would like to see some Physics instrcutions begin to be implemented and maybe H264 support?

Thoughts?
 

Furen

Golden Member
Oct 21, 2004
1,567
0
0
Correct me if I'm wrong but "64bits" applies to integers, floats on the x87 fpu run at 80bit precision and SIMD (sse, mmx, 3dnow) run at 64bit precision. The SIMDs could be extended, conceivably, but I dunno if there'd be much point outside hardcore math applications.
 

aka1nas

Diamond Member
Aug 30, 2001
4,335
1
0
I am pretty sure SSE and SSE 2 are available in x86-64. MMX has been dropped from the ISA and I am unsure about 3dnow. They have little if anything to do with the "bitness" of the chip as they are usually dealing with registers larger or equal to 64 bits anyway.
 

HDTVMan

Banned
Apr 28, 2005
1,534
0
0
As I understood it they were extensions which are basically shortcuts to doing specific tasks that instead of doing the regular caluclations the additional functions of SSE, MMX, and 3DNow were implemented to shorten the need for doing things the long way. I believe they would be 32bit calls to the processor. Would they be reimplemented in a 64bit format? For the day when we abandon 32bit applications or have similar calls already been implemented into the 64bit instructions on the processors?

Like stated they could be used for hardcore math like the functions of Physics which seems to be the main bottleneck in gaming today.

Seems Intel got a good software hold and boost when they implemented their own additional instructions. SSE was a good thing for Intel and AMD rushed to add them in because it did make a difference in some applications.

Intel could concievably do the same with a 64 bit version of SSE, Saying SSE4 or SSE64. That implements a few shortcuts which software companies might jump onto and we would have the same thing again. AMD rushing to implement the additional instructions or Extensions. However AMD could be pro active in implementing some sort of 3DNOW but with a better name to do the reverse of what Intel did for years to AMD.

Im thinking Video Encoding Instructions for WMV9, HD-Divx, Physics, H264, how about MD5 or other encryptions as a CPU call instead of calculating with software algorythms?
 

HDTVMan

Banned
Apr 28, 2005
1,534
0
0
I guess what I am getting at is AMD needs to think beyond just speed to keep ahead of Intel. Granted the architecture is currently better however Intel has money that AMD doesnt and AMD should consider implementing such things in future processor releases so that they can stay competitive. Intel is not going to be slower forever and eventually may solve current leakage. If so then we could see a Ghz/Mhz war again.

Intel had a hard time swallowing the 64bit pill from AMD it would be wise for them to continiue to serve up pills that AMD had to swallow for many years.

This is what I believe AMD needs to do early on to distinguish innovation and true leadership for long into the future beyond the scope of basic chip design. Otherwise Intel could potentially come back and hard and if they do they definaly wont let up on AMD ever again.
 

CheesePoofs

Diamond Member
Dec 5, 2004
3,163
0
0
um ... amd is developing new technologies. They aren't planning on using the K8 core forever.

AFAIK, SSE and 3DNOW have nothing to do with the bitness of the chip. They are still there in 64bit mode.
 

Fox5

Diamond Member
Jan 31, 2005
5,957
7
81
AMD was planning a new SIMD thingie for x86-64, however they decided that SSE would be an official part of x86-64.(not sure if 3dnow is supported under x86-64, but it's basically dead anyhow, I think mmx has to be supported since its part of SSE and 3dnow and is the only way to do 128-bit integers)
 

Furen

Golden Member
Oct 21, 2004
1,567
0
0
hmm... I'll jump in again:

OK, just a clarification: 64 bit operations are not faster than 32 bit operations, they are just as fast, 64 bit operations are just "wider". The only reason why there is a performance improvement between 32 bit and 64bit modes in some apps is because a) there are more registers available to the programmers--which has nothing to do with 64bits, but rather is part of the "definition" of AMD64 (or EMT64, for that matter)--and b) because the cpu can operate on 64bit operands natively (which takes more than one operation using 32bit cpus). So you wouldnt get more physics, just better physics. Unless there is a need for higher precision, there's not much point to making the SIMD instruction sets support "wider" operands.

If you mean that they should make more instruction sets, then yes, I agree. Unfortunately AMD got burned by the lack of support of the 3Dnow instruction set (mostly everyone just ignored AMDs instructions while they jumped all over intel's SSE) so investing money into putting more SIMD support onto their CPUs doesnt make a whole lot of sense unless they can either gain more market share or have many developers (kind of like they did with amd64) pledge support for it. Frankly, I'm a bit surprised they were able to pull off AMD64. After delay after delay of windows 64 I feared that it was going to get pushed under the carpet and everyone was going to forget about it.
 

Aenslead

Golden Member
Sep 9, 2001
1,256
0
0
SSE3 is supposed to execute up to 128bit instructions.

3DNow are code optimizations.

AMD64 are the instructions A64's use to... er... use 64bits.
 

HDTVMan

Banned
Apr 28, 2005
1,534
0
0
Originally posted by: Aenslead
SSE3 is supposed to execute up to 128bit instructions.

3DNow are code optimizations.

AMD64 are the instructions A64's use to... er... use 64bits.

This makes more sense. SSE is supported up to 128 bit so 64bit SSE should be faster than 32bit. Because it would only have to process 2 words instead of 4. If 128bit precision is needed which is highly doubfull. I would bet that the SSE implementations word would be the 32bit version with possibly zeroed out additional 32bits. So Furen is correct in saying better precision physics are possible even if its not needed. So I can understand here why 64bit code may have no additional benefit to that 32bit code. My question is probably more that if programming entirely in 64bit are the SSE/MMX instructions in a 64bit format the cpu understands or have new opcode replaced what is known as SSE and MMX. When someone says they are it still seems they are 32bit codes possibly extended in a 64bit word. You wouldnt call a 32bit word from a 64bit word unless the word was extended to 64bit. Yes I am ware you could call the 32bit extension but in future processors the 32bit extensions will be dropped eventually. Muke like early Xeon dropped 8 and 16 bit opcodes.

Where we could see a benfit is where there are currently no enhanced instructions like MD5. Additional instructions could be added to the CPU to do MD5 or other encryptions that would normally take many more passes since the encoding is I believe is something from 128bit to 4096bit encryption. Using special 64bit opcodes one could encrypt or decrypt them much faster. Instead its done within a program so portions of the program could be implemented into a CPU to enhance. Because its much faster to put one code through to get the answer than it is to ask for this piece and wait for the result to feed that result back in again to get another result and to pass that again and again and again.

I also understand that 64bit is not faster than 32bit unless precision is beyond 32bit then 64 bit will definately be faster. I programmed in assembly language in the 8bit days and doing 16bit math requires a lot more work than passing 2 8 bit math equations. Consider that 8 bit only really goes from -128 to 128. Which could be interesting boost for 48bit color depth on a 64bit cpu but that may be entirely done on the video card.

Technically we could have gone with 48bit CPU's instead of 64bit cpu's and saved a lot of transistor space. I dont believe any of us need the precision 64bits can produce in a single word pass that 48bit couldnt do. But I understand that AMD design of 64 is really so they can handle 2 32bit floating point words in a single pass. This is a benefit for 32bit but once you go beyond 32bit its just wasted space. 48bit would have been the next logical step. Who knows maybe we will see a 48bit CPU from someone. Granted it would need an OS but its a thought.