Coming 5Ghz+ IBM processors

[halbhh]

http://hardware.slashdot.org/article.pl?sid=07/01/04/1824223&from=rss

on 65nm. Since AMD is their development partner, ya gotta wonder if AMD can get some of that mojo too.

btw, it says around 700M transistors! compare to 230 million in AM2 x2s

my jaw just dropped for a moment

With IBM doing the major game consols, they'll have the revenue to continue progressing, so perhaps the idea of the "duopoly" of AMD/Intel isn't quite right either.

 

[ajfink]

We'll have to see what performance is like before we say how well it will do. Also, the clock speed of those processors is essentially irrelevant, I believe. Completely different core architecture.

 

[BaronMatrix]

http://hardware.slashdot.org/article.pl?sid=07/01/04/1824223&from=rss

on 65nm. Since AMD is their development partner, ya gotta wonder if AMD can get some of that mojo too.

btw, it says around 700M transistors! compare to 230 million in AM2 x2s

my jaw just dropped for a moment

With IBM doing the major game consols, they'll have the revenue to continue progressing, so perhaps the idea of the "duopoly" of AMD/Intel isn't quite right either.

IBM would never try to make X64 CPUs, though they are a major source of processors. Power 6 will be high end server only.

 

[halbhh]

Sure, they're not doing x86....but....

It does make me wonder some even about the possibility of someone writing some code to emulate x86 or whatever in order to run Vista on a machine.

Just for the heck of it.

And then they could benchmark, etc.

Perhaps in time, it would be possible to have some kinda motherboard made in order to do that, and then the Windows world wouldn't be a duopoly of AMD/Intel (currently 99% together of that market).

 

[cwj717]

http://hardware.slashdot.org/article.pl?sid=07/01/04/1824223&from=rss

on 65nm. Since AMD is their development partner, ya gotta wonder if AMD can get some of that mojo too.

btw, it says around 700M transistors! compare to 230 million in AM2 x2s

my jaw just dropped for a moment

With IBM doing the major game consols, they'll have the revenue to continue progressing, so perhaps the idea of the "duopoly" of AMD/Intel isn't quite right either.

IBM would never try to make X64 CPUs, though they are a major source of processors. Power 6 will be high end server only.

IBM has done X86 before... they could do it again. Power6 is definitely not it though.

 

[halbhh]

yeah, but what I'm wondering is....why not? like, what if they did? via a bridge chip or special motherboard, or even just emulation software.

that would be interesting.

sure I know it's not likely, but.....why not?

 

[cwj717]

emulation software

Software emulation is not worth it... sure it would allow you to run X86 software but it would be so much slower it probably wouldn't be much faster (if at all) than current X86 procs. Anyone remember FX32 for the Alpha? it allowed the Alpha to run more software but that software didn't run any faster on the Alpha than the Intel X86 procs at the time.

 

[halbhh]

I don't know enough about the IBM architecture to guess at that overhead cost, but the big 5ghz just makes it interesting.

Although it's a very different animal, wasn't the old G5 (IBM right?)emulation software for Windows overhead around....20%(?)

 

[cwj717]

I don't know enough about the IBM architecture to guess at that overhead cost, but the big 5ghz just makes it interesting.

Although it's a very different animal, wasn't the old G5 (IBM right?)emulation software for Windows overhead around....20%(?)

Im not sure actually, have a link to that information? I do find it hard to believe that it would only be 20% slower than native software. Although, I suppose that its possible... unlikely though. I don't really know enough about the Power 6 to make a guess on that one.

 

[turpit]

http://hardware.slashdot.org/article.pl?sid=07/01/04/1824223&from=rss

on 65nm. Since AMD is their development partner, ya gotta wonder if AMD can get some of that mojo too.

btw, it says around 700M transistors! compare to 230 million in AM2 x2s

my jaw just dropped for a moment

With IBM doing the major game consols, they'll have the revenue to continue progressing, so perhaps the idea of the "duopoly" of AMD/Intel isn't quite right either.

I see no reason to be excited by this, or why you are excited. As everyone already pointed out to you, it is not realistic to see this tech show up in an AMD CPU. Furthermore it is "intelish", and even Intel finally learned, higher clockspeed does directly equate to higher performance. AMD themselves waved that little red flag under Intels nose for 3 years

Regardless, 5Ghz is actaully old news. Even THG got a crap P4 over 5Ghz. They had to use nitrogen to do it, but they got it there.

Old news. An intel cooler for 5Ghz
http://www.bit-tech.net/news/2006/03/08/intel_liquid_cooler/

 

[halbhh]

I was probably remembering the overhead for current Macs emulation using the current C2duos, but....it's beside the point. I understand the profound optimization of x86 and Windows for compatibility, but still can't help but wonder if some kinda new variation on RISC would be interesting, with the "APU" concept out there. I don't know enough about architecture to speculate really.

Another interesting thing about it is just the plain and simple thing about getting 700 million transistors to 5ghz for under 100 watts!!

That's *gotta* mean something important for AMD somehow, and for us eventually.

 

[halbhh]

what do you think about the idea of IBM getting 700M transistors to 5Ghz for under 100 watts?

seems like if AMD is a partner in 65nm, they would be able to somehow get a part of that kinda change too.

 

[BaronMatrix]

yeah, but what I'm wondering is....why not? like, what if they did? via a bridge chip or special motherboard, or even just emulation software.

that would be interesting.

sure I know it's not likely, but.....why not?

Sure, it's possible. I'm sure with something like the Alpha Emulation SW (FX!32) it would probably outdo Intel and AMD. I used it to play games at incredible speed with emulation on Alphas and Win2K. We even setup a DeathMatch server with a dual 500MHz Alpha. Talk about serving!!!!!!!!!!

 

[halbhh]

thanks Jack. I like that explanation, complex enough to mean something but still be partly understandable and partly mysterious.

if you'll indulge my ignorance, I thought the issue for frequency of x86 was really about power use -- they don't want to burn 150 watts, etc., so thus the multicore race now with a 3Ghz (or so) cap.

Are you implying that IBM hasn't done any breakthrough here re power use? Is there no way AMD or Intel could go to 5ghz under 100 watts also? Perhaps this isn't easy to answer, but I don't get it why AMD doesn't just modify the limiting oscillator.

 

[BaronMatrix]

emulation software

Software emulation is not worth it... sure it would allow you to run X86 software but it would be so much slower it probably wouldn't be much faster (if at all) than current X86 procs. Anyone remember FX32 for the Alpha? it allowed the Alpha to run more software but that software didn't run any faster on the Alpha than the Intel X86 procs at the time.

Bull sh i t. I used Alphas to test Win2K and FX!32 worked super great, but you had to run the program a few times to get all of the code converted to binary.

The games of the time were faster on Alpha. EVERYTHING WAS FASTER ON ALPHA. We even used a TinCup as a Quake server.

 

[M_with_one_M]

If IBM's processor should proove to be a high performer, what is the probability that we will see some OS for home use that supports it. Could for instance some macOS be used or modified to work.

 

[HotFoot]

If you're up for using a Linux distro, then they already exist to run on basically anything you want (x86, Power, Sparc, etc.). The problem is as always: finding the programs you want for the OS. I use linux for everything other than gaming. There are too few games out with Linux support.

 

[mesarectifier]

If IBM's processor should proove to be a high performer, what is the probability that we will see some OS for home use that supports it. Could for instance some macOS be used or modified to work.

This is what I'm interested in - a hyper-quick PowerPC-based Mac would be a real boon for some people (like Photoshoppers, until they finally release universal binaries) although Apple are phasing out PPC updates and software. Leopard is going to be the last version of Mac OS that'll run on PowerPC hardware.

Still, could be useful? Although to be honest the current Mac Pro is quite fast enough, don't you think! Plus it'll have updates and new OS's for probably the next decade.

 

[M_with_one_M]

Although to be honest the current Mac Pro is quite fast enough, don't you think!

I'm a PC user, so I wouldent know. I like to build my own computer, with the parts I like.
What I would like is the ability to just change my motherboard and processor and OS, to put some more pressure on Intel and have one more alternative.

I can't help but wondering if there could be something better then x86 out there, that never gets a chance, because of the OS.

 

[cwj717]

emulation software

Software emulation is not worth it... sure it would allow you to run X86 software but it would be so much slower it probably wouldn't be much faster (if at all) than current X86 procs. Anyone remember FX32 for the Alpha? it allowed the Alpha to run more software but that software didn't run any faster on the Alpha than the Intel X86 procs at the time.

Bull sh i t. I used Alphas to test Win2K and FX!32 worked super great, but you had to run the program a few times to get all of the code converted to binary.

The games of the time were faster on Alpha. EVERYTHING WAS FASTER ON ALPHA. We even used a TinCup as a Quake server. Thats depends on what Alpha it was... 21164, 21164PC, or 21264? And what clock speed the Alpha was running at. Also what Intel chip would you be comparing it to? How much cache and what clock speed. FX32 is not horrible really... but I wouldn't expect it to compare to optimized native code. After considering the cost of the Alpha systems... I don't think it would be worth it... get an Alpha to run FX32 generated code? Don't get me wrong though, i think the Alpha was a great processor, it had its place. I still wouldn't mind getting one to run Linux on

 

[cwj717]

yeah, but what I'm wondering is....why not? like, what if they did? via a bridge chip or special motherboard, or even just emulation software.

that would be interesting.

sure I know it's not likely, but.....why not?

Sure, it's possible. I'm sure with something like the Alpha Emulation SW (FX!32) it would probably outdo Intel and AMD. I used it to play games at incredible speed with emulation on Alphas and Win2K. We even setup a DeathMatch server with a dual 500MHz Alpha. Talk about serving!!!!!!!!!!

How did you run games? You mean there were 3D accelerators at the time with Alpha drivers? And on Win2k no less... I thought it was just the beta versions of 2k that supported the Alpha. :S

 

[1Tanker]

http://hardware.slashdot.org/article.pl?sid=07/01/04/1824223&from=rss

on 65nm. Since AMD is their development partner, ya gotta wonder if AMD can get some of that mojo too.

btw, it says around 700M transistors! compare to 230 million in AM2 x2s

my jaw just dropped for a moment

With IBM doing the major game consols, they'll have the revenue to continue progressing, so perhaps the idea of the "duopoly" of AMD/Intel isn't quite right either.

I can't ..for the life of me, figure out why you(the guy who floods the forums with posts/threads downplaying the importance of CPU.. "just get a fast GPU,HD")would care about this. :x

 

[HotFoot]

I can't help but wondering if there could be something better then x86 out there, that never gets a chance, because of the OS.

Of course there's better out there, but industry momentum has to be factored in to what's "best". Academically speaking, base-3 computers are much more efficient than base-2 computers that we use today. However, we may never see them because everyone works in and studies binary. What's already in the field will always dictate 95% of what's coming next.

 

[halbhh]

thanks Jack. I like that explanation, complex enough to mean something but still be partly understandable and partly mysterious.

if you'll indulge my ignorance, I thought the issue for frequency of x86 was really about power use -- they don't want to burn 150 watts, etc., so thus the multicore race now with a 3Ghz (or so) cap.

Are you implying that IBM hasn't done any breakthrough here re power use? Is there no way AMD or Intel could go to 5ghz under 100 watts also? Perhaps this isn't easy to answer, but I don't get it why AMD doesn't just modify the limiting oscillator.

Power is a huge deal and yes, it is part of the equation. Static power has a huge benefit with SOI, but also gets worst as you shrink. Dynamic power gets better as you shrink unless you do not do constant field scaling -- then it goes up, I linked a scaling theory paper earlier in the other thread you can read in much more detail there. Furthermore, the circuit design and as I discussed above complexity also affects power.

http://pllab.cs.nthu.edu.tw/cs5404/low-power/micro00.pdf
See figure 1 showing the rate of increase in static power rising at a steeper slope. Dynamic power increases in this case because the data is under constant voltage scaling.

Overall, it is convoluted such that there is no piece of predictive data that could be generated to say === yeah 5 GHz but power will be X or such.

Frankly, power in today's devices is actually being consume by the PLL clock circuitry above all else:
http://www.cag.csail.mit.edu/6.893-f2000/project/maze_check2.pdf
See figure 6, check out the yellow blocks --- all PLL/Clock circuit driven.

This is why I am little perplexed with AMD's decision to indpendently clock each core --- while the idle power will benefit, the costs is extra power for the extra PLL circuits at full load.

This looks interesting. Thanks again!

 

[halbhh]

http://hardware.slashdot.org/article.pl?sid=07/01/04/1824223&from=rss

on 65nm. Since AMD is their development partner, ya gotta wonder if AMD can get some of that mojo too.

btw, it says around 700M transistors! compare to 230 million in AM2 x2s

my jaw just dropped for a moment

With IBM doing the major game consols, they'll have the revenue to continue progressing, so perhaps the idea of the "duopoly" of AMD/Intel isn't quite right either.

I can't ..for the life of me, figure out why you(the guy who floods the forums with posts/threads downplaying the importance of CPU.. "just get a fast GPU,HD")would care about this. :x

I'm glad you get to re-evaluate your evaluation. 8)

I'm very curious about what's going to happen to cpus in the future. It's a good thing to have more than one premier company in the field. We will certainly have a lot of interesting options coming, and it's an interesting time right now especially, with the struggle to birth more general multi-core code and such.