AMD SSE5 ?

D

duby229

2[H]4U
Joined
May 1, 2005
Messages
3,014
Check it out folks...... Seems like the integration of ATi's GPU technology into the x86 ISA has begun.

http://developer.amd.com/sse5.jsp

My take on it, is that these new instructions will more likely execute not on the CPU, but rather on the GPU portion of Fusion. We'll have to wait and see, but I believe that I'm corretc.
 
I thought intel was the one who made all of the SSE instruction sets, and AMD just had free reign to use them as per a lawsuit settlement.

is AMD making there own instructions now again?
 
Looks to me that this is AMD slapping Intel by releasing SSE5 just when Intel is about to start really advertising for SSE4. LOL
Legal hilarity will probably ensue.
 
What legal hilarity? They have cross licensing agreements. Kudos to AMD for being the first one to release a new instruction set. Unfortunately, they have to give it to Intel as well but this means that AMD processors will (hopefully) have it first. Then AMD can find a benchmark that uses SSE5 and use that to demo their new processors.
 
Eickst said:
Then AMD can find a benchmark that uses SSE5 and use that to demo their new processors.
Click to expand...
And what good would that do besides inflated benchmark scores that tell you nothing of how the good the processor really is? Never been a fan of benchmarks, they give you a general performance area but we have all seen how benchmarks can be skewed, unless all you do is run benchmarks they are pretty much useless.
 
yevaud said:
Looks to me that this is AMD slapping Intel by releasing SSE5 just when Intel is about to start really advertising for SSE4. LOL
Legal hilarity will probably ensue.
Click to expand...

So will AMD just skip SSE4? Could be a mistake if the early SSE4 numbers (100%+ improvement) mean anything.
 
munkle said:
And what good would that do besides inflated benchmark scores that tell you nothing of how the good the processor really is? Never been a fan of benchmarks, they give you a general performance area but we have all seen how benchmarks can be skewed, unless all you do is run benchmarks they are pretty much useless.
Click to expand...

what all of the SSE's are is a set of instructions that the CPU uses to calculate whatever, the question wouldn't be if AMD could make a benchmark to utilize it, but what programs would take advantage of it in the first place, and since these instructions don't yet exist on any chip, no one can write programs to take advantage of something that doesn't yet exist intill it exists and is available for sale.

The only reason AMD is doing this is to say fuck you to intel, because as we know, the last instruction set AMD made was 3dnow, and intel could of given a fuck less, but since intel has made every instruction set that anyone has ever needed for x86 (mmx,sse1-4.1/ssse3), either AMD took what intel was going to use for what would be called SSE5, and developed it themselfs or intel already developed it and AMD wants to get a jump on implementation.

If AMD did develop it, I got 10 bucks on intel doesn't use it, or disregards it or downplays it in some way to say fuck you back, and intel has the power to do so.
 
Simpson5774 said:
The only reason AMD is doing this is to say fuck you to intel, because as we know, the last instruction set AMD made was 3dnow, and intel could of given a fuck less, but since intel has made every instruction set that anyone has ever needed for x86 (mmx,sse1-4.1/ssse3), either AMD took what intel was going to use for what would be called SSE5, and developed it themselfs or intel already developed it and AMD wants to get a jump on implementation.
Click to expand...

I believe AMD developed AMD64, which ended up being x86_64. Intel implemented that. It's been pretty sucessful too.
 
Simpson5774 said:
what all of the SSE's are is a set of instructions that the CPU uses to calculate whatever, the question wouldn't be if AMD could make a benchmark to utilize it, but what programs would take advantage of it in the first place, and since these instructions don't yet exist on any chip, no one can write programs to take advantage of something that doesn't yet exist intill it exists and is available for sale.
Click to expand...
I know what SSE's are, I was replying to his comment on the need for an Amd benchmark on SSE5 which would only provide skewed results, if a benchmark instead of a useful program is what you are judging the processors performance by, what good are the results to anyone. Thats what I was trying to say.
 
So Intel will have iSSE5 and AMD will have ...

:D

You might want to rethink that name, AMD.
 
RCGodward said:
I believe AMD developed AMD64, which ended up being x86_64. Intel implemented that. It's been pretty sucessful too.
Click to expand...

Intel was forced to use amd64 because of microsoft didn't want to implement two 64 bit instructions, and because AMD had implemented a 64 bit x86 processor first, intel got screwed over.

its known as EM64T for intels fyi.

If intel makes an instruction set, AMD gets free reign on it without cost (due to a former lawsuit), if AMD makes an instruction set, Intel has to shell out the fee in order to use it, and they will have to it goes mainstream (x64)
 
Simpson5774 said:
If intel makes an instruction set, AMD gets free reign on it without cost (due to a former lawsuit), if AMD makes an instruction set, Intel has to shell out the fee in order to use it, and they will have to it goes mainstream (x64)
Click to expand...
That is not true. AMD's lawsuit that was settled in the early 1990s gave it the right to clone the i386, and give Intel money for that privilege and other tech. :p

AMD and Intel have different license agreements that allow each other access to various CPU technologies. This is the last long term one: http://news.com.com/2100-1040-257059.html Looks like AMD is still paying Intel royalties, not the other way around.
 
munkle said:
And what good would that do besides inflated benchmark scores that tell you nothing of how the good the processor really is? Never been a fan of benchmarks, they give you a general performance area but we have all seen how benchmarks can be skewed, unless all you do is run benchmarks they are pretty much useless.
Click to expand...

Its called PR and marketing. Something that tends to make a company some green stuff.
 
Honestly, this doesn't really matter. An instruction set is only as good as the programs written for it. Intel has always been the industry leader in hardware and software support for instruction sets, so that's what is going to get the support. What was the last killer app you used that took great advantage of the technology joke known as 3DNow!? :p
 
It depends where the instruction set is aimed how it will fare.

If SSE5 is being made for HPC applications, it doesn't matter if Intel supports it or how it does in the desktop space. I think it really is made for this segment. It gives AMD an edge over Xeons, which won't have similar acceleration*, in low cost nodes, and an edge in price vs Itanium (Tukwila) which is also getting SIMD enhancements.

If AMD is hoping for desktop adoption, then it is critical for Intel to support it for it to gain acceptance. It's not the same case as AMD64, which Intel pretty much had to adopt when MS said it would only support 1 x86-64 instruction set. I haven't read over all the opcodes yet, but I don't see SSE5 as something critical for Intel to support. Intel can already automatically fuse certain instructions so it doesn't need to have explicit new instructions.

* Larabee co-processors will smoke initial and probably even later on-die SIMD implementations. I have no idea where Larabee fits in though.
 
AFAIK AMD64 succeeded because AMD released it as an open standard.. if you had two ISAs to support and one was free to develop for, which one would you choose?

Hypertransport succeeded the same way - by building and open standards consortium, and SSE5 will probably rely on much the same approach to build support amongst the developer community.

As for 3Dnow! - it was pretty nice to have when it came out for the K6-2 (I had one :D ) - don't go calling it a joke, I'm fairly sure it was more use than MMX was back in the day :)
 
MMX,3DNow(Pro),SSE et al.

it all comes down to software being written to take advantage of the instruction set(s). if software does not take advantage of the intruction set....its never used. 3D now was a good attempt,and provided a decent performance boost, but there was/is no software written to utilize it. MMX was the samething but MMX being exstensibly integer...no code = no gain.

if SSE5 is sucessful or not depends on adpation. regardless it is good to see AMD attempting to innovate. the same people who slam AMD for SSE5, are many of the same who slammed them for not leading or innovating prior to Athlon.

I hope Intel and AMD can overcome their differences and ego so consumers win :cool:
 
xxGriff said:
MMX was the samething but MMX being exstensibly integer...no code = no gain.
Click to expand...
MMX was and is still being used now, believe it or not. It is useful for handling host signal processing (modem, sound, others) and some current compilers (gcc, icc) still output auto-vectorization MMX code by default. I'm pretty sure the MS compilers do too.
 
pxc..i think you missed my point. MMX et al were marketed as the "turbo" charger for intels x86, and "promised" big. now if you manufacture modems, or write code (gcc, icc)...then you may be using MMX, but just how much "boost" is there? say compared to just a faster CPU?

my point is that these instruction sets hold promise, but if the software used by regular users is not coded for it, then it is more silicon, more etching, more power, more heat, more hype and more $$$

i will say that you are correct in MMX use, but if you can remember the marketing onslaught that preceded the Pentium w/MMX...and compare it to the pay off. well i remember having to get a new board that supported split rail voltages and for my 133mHz to 233mHz upgrade, i was underwhelmed with the added instruction set. MMX was pushed as a Multi-Media, and Gaming boost that did neither.
 
You must log in or register to reply here.
Back
Top