AMD gearing up for speed

2

Comments

  • Reply 21 of 53
    chagichagi Posts: 284member
    Quote:

    Originally posted by Kecksy

    True, they're in the process of transitioning now and most claim they'll be ready for volume production in Q2 2004. It's reasonable to assume then that IBM will be able to take the G5 to 3GHz by summer in time for an Apple update. A 50% increase in clock speed by summer is probably better than what Intel will be able to pull off. I doubt we'll see 5GHz Pentiums in the next 12 months. Prescott is supposed to dissapate over 100 watts at 3.4GHz, and that's on 0.09-micron! No way they're going to make it 5GHz without skrinking the process again.



    PowerPC 970 is looking very strong, I think. The only thing Apple has to worry about is the Athlon64. It could could scale just as well.







    What the hell is that supposed to mean?




    Just for interest sake, here's an unofficial Intel roadmap, courtesy of http://pc.watch.impress.co.jp (the site is not in English btw).



    A couple of criticisms of the roadmap to consider though. It's doubtful that Prescott will ship 4th quarter of 2003, so we'll more likely see it 1st or even 2nd quarter of 2004. Given the problems that Intel is currently experiencing, I highly doubt that Intel will be shipping 4GHz by mid-2004.



    If IBM and Apple are able to deliver on the promise of a dual 3GHz system next summer, it's going to be very competitive with the PC platform offerings. Not to mention that most PC desktops don't run dual-CPU configs, due in part to Intel's retarded marchitecture. In recent months, Xeon is the only option for dual-CPU and greater on the Intel platform, as you can't use the P4s for MP systems. This may change a bit soon though, as AMD Opterons have become a decent desktop alternative to the AMD Athlon series chips, as opposed to being strictly relegated to servers.



  • Reply 22 of 53
    powerdocpowerdoc Posts: 8,123member
    I have read somewhere : but is it good infos ?, that the Prescott chip will be more deeply pipelined than the P4, and thus at equal mhz it will be lighty slower.



    If this info is true, a 3 ghz G5 will kick the ass of a 4 ghz Prescott.
  • Reply 23 of 53
    eugeneeugene Posts: 8,254member
    Quote:

    Originally posted by Chagi



    A couple of criticisms of the roadmap to consider though. It's doubtful that Prescott will ship 4th quarter of 2003, so we'll more likely see it 1st or even 2nd quarter of 2004. Given the problems that Intel is currently experiencing, I highly doubt that Intel will be shipping 4GHz by mid-2004.




    The question becomes: Do you think Intel could have shipped a much faster P4s by now if AMD had been more competitive in the year before the release of the Athlon 64? I still think a 4 GHz P4 is doable by 2H04. The .13µ Northwood 3.4 GHz P4 EE will be out soon. That's not even on this roadmap. 3.6 GHz Prescott in February/March. 4.0 GHz in July/August sounds completely reasonable given the move to .09µ.
  • Reply 24 of 53
    Quote:

    Originally posted by Powerdoc

    I have read somewhere : but is it good infos ?, that the Prescott chip will be more deeply pipelined than the P4, and thus at equal mhz it will be lighty slower.



    If this info is true, a 3 ghz G5 will kick the ass of a 4 ghz Prescott.




    It is true that the Prescott will be slightly deeper pipelined (more drive-stages afaik, to help accelerate clock-frequency), but Intel has also done efforts to make the Prescott more efficient (as in higher IPC). I don't know exactly what, but it will have SSE3 (minor upgrade, really), PNI (Prescott New Instructions), so I guess it won't be much slower per clock than the Northwood, if slower at all.



    But I too think that a 3GHz G5 with 1MB of L2 cache would be more competitive with a 4Ghz (not that they'll reach such a high frequency in time) Prescott, than the 2GHz G5 is with the 3.2Ghz P4 today.



    Now, if only Apple/IBM could reduce their damn cache and memory latencies with the G5.
  • Reply 25 of 53
    drboardrboar Posts: 477member
    With Intel in the 3 Ghz range (2.8-3.2) for quite a while that would be an opportunity for Motorolas G4 to acually cross that "half clock speed threshold" that they have been under since 1999 or early 2000.



    Will they ever cross 2 GHz will anyone care
  • Reply 26 of 53
    powerdocpowerdoc Posts: 8,123member
    Quote:

    Originally posted by Zapchud

    It is true that the Prescott will be slightly deeper pipelined (more drive-stages afaik, to help accelerate clock-frequency), but Intel has also done efforts to make the Prescott more efficient (as in higher IPC). I don't know exactly what, but it will have SSE3 (minor upgrade, really), PNI (Prescott New Instructions), so I guess it won't be much slower per clock than the Northwood, if slower at all.



    But I too think that a 3GHz G5 with 1MB of L2 cache would be more competitive with a 4Ghz (not that they'll reach such a high frequency in time) Prescott, than the 2GHz G5 is with the 3.2Ghz P4 today.



    Now, if only Apple/IBM could reduce their damn cache and memory latencies with the G5.




    The IBM engineers have claimed that the G5 core is build to deal with such latencies : the vast number of registers allow to feed continualy the beast without suffering from latencies problems.



    Thanks for confirming the deeper pipelining. But even if Intel did some tricks to make the Prescott more efficient, a deeper pipelining means nearly always less speed, unless great changes in the design are made. For example the G4 7400 have 4 units and the 7450 7 units, and they have the same performance level per mhz only after optimisation of the code.

    If the Prescott do not have more executive units than the P4, even with a larger cache, an improved BPU, the chip will be slower at equal mhz.

    A optimist guess for Intel will be a 3,2 ghz P4 equal a 3,4 ghz P5 (prescott).



    A last note :

    The P4 EE are Extremely Expansive : near 1000 ? at least in France (if you can find one) for the 3,2 ghz model.

    I think that AMD is still in the competition with his Athlon 64 bits.
  • Reply 27 of 53
    kecksykecksy Posts: 1,002member
    Intel can lengthen it's pipelines as much as it wants. It's chips are still going to hit a brick wall because of heat. I heard somewhere at the current rate the heat density of chips is increasing, within a decade or so, they'll be as hot as a nuclear furnace.
  • Reply 28 of 53
    Quote:

    Originally posted by Powerdoc

    Thanks for confirming the deeper pipelining. But even if Intel did some tricks to make the Prescott more efficient, a deeper pipelining means nearly always less speed, unless great changes in the design are made. For example the G4 7400 have 4 units and the 7450 7 units, and they have the same performance level per mhz only after optimisation of the code.

    If the Prescott do not have more executive units than the P4, even with a larger cache, an improved BPU, the chip will be slower at equal mhz.

    A optimist guess for Intel will be a 3,2 ghz P4 equal a 3,4 ghz P5 (prescott).




    I seriously doubt the Prescott will be less efficient per MHz than the Northwood. The lengthened pipeline is only 3 stages longer, or 15% (IIRC). Since the Prescott do feature a larger L2 cache, double the L1 data cache, an improved BPU, and more, I would be seriously surprised if it performed as bad as you predict "optimistically"



    If a 3.4GHz Prescott equals a 3.2GHz P4, it is terrible for Intel. Good for us mac fans
  • Reply 29 of 53
    chagichagi Posts: 284member
    Quote:

    Originally posted by Zapchud

    I seriously doubt the Prescott will be less efficient per MHz than the Northwood. The lengthened pipeline is only 3 stages longer, or 15% (IIRC). Since the Prescott do feature a larger L2 cache, double the L1 data cache, an improved BPU, and more, I would be seriously surprised if it performed as bad as you predict "optimistically"



    If a 3.4GHz Prescott equals a 3.2GHz P4, it is terrible for Intel. Good for us mac fans




    We'll have to wait and see what happens with Prescott. I'm trying to find the link right now, but there was a website with leaked preliminary Prescott benchmarks, and they were slower than the current model. Of course, this is pretty much moot when faster clocked models are shipped.



    The Pentium 4 EE (Extremely Expensive edition) chips are little more than Intel attempting to pee on AMD's Athlon 64 FX parade. According to www.theinquirer.net, good luck getting your hands on the EE variant from anyone but Dell, and even then it may not be possible to get a shipping system before the end of 2003.
  • Reply 30 of 53
    powerdocpowerdoc Posts: 8,123member
    Here is a link dealing with the benchmarks of a 2,8 ghz prescott chip : http://oc.com.tw/article/0309/readgo...le.asp?id=1974



    Basically the chip is on par with the nortwood or have a little advantage. There is only an issue with CPU mark 99, but at the light of the others tests, we should forget this one.



    An another article pointed it out the fact that the bigger L2 cache was the main reason for these good results.



    SSE3 seemed fun also : i can't wait for SSE 4. Intel is bringing more and more instructions generations after generations : what's a pain for programmers ...
  • Reply 31 of 53
    chagichagi Posts: 284member
    Quote:

    Originally posted by Powerdoc

    Here is a link dealing with the benchmarks of a 2,8 ghz prescott chip : http://oc.com.tw/article/0309/readgo...le.asp?id=1974



    Basically the chip is on par with the nortwood or have a little advantage. There is only an issue with CPU mark 99, but at the light of the others tests, we should forget this one.



    An another article pointed it out the fact that the bigger L2 cache was the main reason for these good results.



    SSE3 seemed fun also : i can't wait for SSE 4. Intel is bringing more and more instructions generations after generations : what's a pain for programmers ...




    Thanks for the link.



    I have to agree that the larger L2 cache is helping the Prescott benchmark, but if that's the case, it's just a smart design decision (aside from the added cost to manufacture). Of course, part of the whole reason to switch to 90mm fabbing is to reduce manufacturing cost, so griping about the cost of L2 is kind of moot...
  • Reply 32 of 53
    powerdocpowerdoc Posts: 8,123member
    Quote:

    Originally posted by Chagi

    Thanks for the link.



    I have to agree that the larger L2 cache is helping the Prescott benchmark, but if that's the case, it's just a smart design decision (aside from the added cost to manufacture). Of course, part of the whole reason to switch to 90mm fabbing is to reduce manufacturing cost, so griping about the cost of L2 is kind of moot...




    I expect that the 90 nm G5 will feature a 1 MB L2 cache as well
  • Reply 33 of 53
    We also don't know of the benefits, if any, of recompiling and/or optimizing code for the Prescott. We'll indeed have to wait and see.
  • Reply 34 of 53
    ed m.ed m. Posts: 222member
    Does anyone know how well these Athlon-64 CPUs are selling? And I mean selling as in actual systems going to consumers... We already know that the Opteron is the second slowest selling CPU right behind the Itanium. The G5s (aka: PPC970) have been selling like crazy. Anyone have any numbers?



    --

    Ed
  • Reply 35 of 53
    chagichagi Posts: 284member
    Quote:

    Originally posted by Ed M.

    Does anyone know how well these Athlon-64 CPUs are selling? And I mean selling as in actual systems going to consumers... We already know that the Opteron is the second slowest selling CPU right behind the Itanium. The G5s (aka: PPC970) have been selling like crazy. Anyone have any numbers?



    --

    Ed




    I hate to critique you without statistics to back me up, but you're definately wrong about a couple of things.



    I highly doubt that the Opteron is behind the Itanium (Itanic as it's been lovingly dubbed) in sales. Thor's Hammer (the forthcoming supercomputer built from Cray's new "Red Storm" product) alone would probably outstrip all Itaniums sold to date. Keep in mind that Itanium has been doing absolutely horrible in sales so far.



    As for Athlon 64 sales figures, we're probably going to have to wait for at least one AMD fiscal quarter to go by, as the chips were just launched a month or so ago.
  • Reply 36 of 53
    kurtkurt Posts: 225member
    One of the things I think is interesting about this whole thing is the lack of information from IBM. It seems that Apple must have some words in their contract so that IBM keep quiet about any chips designed for them. It was the same with Motorola. Both companies talk about their future chips in other areas but chips that will be going into Macs get very little advanced discussion.



    Look at Intel and AMD. Both companies are touting (sp?) future designs while IBM and Motorola seem to only have occasional slips where information gets out.



    There was a computer company from the early days of computing that preannounced its next product and ending up losing sales of its current product. It ended up going bankrupt. I think Steve Jobs remembers this lesson very well. (certainly better than I since I can not remember the name of the company)
  • Reply 37 of 53
    Lol, i doubt that apple will have problems with heat from the G5. The heatsink on those babies are crazy!
  • Reply 38 of 53
    eugeneeugene Posts: 8,254member
    Quote:

    Originally posted by Chagi

    I hate to critique you without statistics to back me up, but you're definately wrong about a couple of things.



    I highly doubt that the Opteron is behind the Itanium (Itanic as it's been lovingly dubbed) in sales. Thor's Hammer (the forthcoming supercomputer built from Cray's new "Red Storm" product) alone would probably outstrip all Itaniums sold to date. Keep in mind that Itanium has been doing absolutely horrible in sales so far.




    Er, you've definitely misinterpreted his message. On a list of slowest selling processors, Itanium would be first and Opteron would be second. He didn't say Itanium was outselling Opteron. Whether that statement is true is another issue altogether.



    Quote:

    As for Athlon 64 sales figures, we're probably going to have to wait for at least one AMD fiscal quarter to go by, as the chips were just launched a month or so ago.



    Athlon 64 sales will be abysmal until a major OEM actually starts selling machines with them. You know it won't be Dell. The only major OEM that sells AMD equipped PCs right now is HP, and they're probably waiting for Windows XP Pro 64-bit Edition.
  • Reply 39 of 53
    kecksykecksy Posts: 1,002member
    Quote:

    Originally posted by Eugene

    Er, you've definitely misinterpreted his message. On a list of slowest selling processors, Itanium would be first and Opteron would be second. He didn't say Itanium was outselling Opteron. Whether that statement is true is another issue altogether.





    Athlon 64 sales will be abysmal until a major OEM actually starts selling machines with them. You know it won't be Dell. The only major OEM that sells AMD equipped PCs right now is HP, and they're probably waiting for Windows XP Pro 64-bit Edition.




    You underestimate the power of the enthusiast market. The original Athlon, never had huge success with OEMs. Most sales where made directly to hobbyists who built they're own machines and didn't have to worry about wrath from Intel. I don't see this picture changing with the Athlon64. Hobbyists will embrace it because of its great performance, but Intel will keep the OEM market because they're considered the standard.
  • Reply 40 of 53
    Quote:

    Originally posted by Zapchud

    ...I don't know exactly what, but it will have SSE3 (minor upgrade, really), PNI (Prescott New Instructions), so I guess it won't be much slower per clock than the Northwood, if slower at all...



    Just once I would like to see Intel introduce a new chip which doesn't add new instructions to an already cluttered ISA. What a freakin' mess. This doesn't help their IPC anymore because practically nobody will take advantage of these.
Sign In or Register to comment.