What's left for AMD?

[Junkkyy]

After wandering the web a bit and being unable to find any details, I thought I'd ask you all what is left for AMD in the 32 bit market?

Most AMD fans talk about 64 bit like it's cold fusion or something, but I really don't give a damn. I (and I'm sure most others) won't need to be able to address more than 4G of memory for a few years, and the benefits beyond this are marginal at best. I enjoy having a nice tiny little chip (80mm^2) that only costs me $150 and is proven and reliable. I do not want an Itanium or x86-64 extensions, and I will not want one for at least a good five years.

So what is left for AMD in 32 bit land? I can't find many details on Barton- I read that although they will increase the L2 to 512KB, it will *still* be at the obsolete 133FSB. Anyone know how high clock speeds are expected to get on this? What about after Barton? The AMD roadmap is blank, and I certainly hope they aren't abandoning 32 bits and putting all their eggs in the Sledgehammer due to the 32 bit spanking they are currently receiving from Intel.

What's going to be the end of the road in 32 bit land? I sure hope AMD isn't abandoning this market, because without serious competition Intel may start reaching monopolistic market shares, and prices will go up while progress slows down.

Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway

 

[cellbiogeek]

Hammer will be able to do both 32bit and 64bit. The chip will not be a monster in size and will get smaller as they move to a smaller process. If you always want proven and reliable then buy older technology. No one's forcing you to buy anything right after it debuts.

 

[imgod2u]

Well, Hammer's features won't be exclusively 64-bit extentions. It will have improvements to help in modern day 32-bit code. It just happens to throw in 64-bit extensions because AMD needed this design to be in the server and workstation market as well.

 

[shallowbaby]

i'm not too impressed with amd's 64-bit hipe either. when it releases, there no 64-bit software for it, ohoh! even when os and software is avail, it will be a slow process and who knows the actual benefits of it is.

i wouldn't compare hammer's 64-bit with Itaninum2, they're different classes.

amd's HyperTransport isn't that impressive either, but look at intel's "HyperThreading".. sure is nice to trick software into believing there's two processors.



<font color=green> linux is free.. if your time is worthless </font color=green>

 

[shallowbaby]

imgod2u,

i believe the 32-bit enhancements you're talking about for hammer is SSE2 (yeh, intel has that for over a year now).. and not AMD has abandonned their 3DNow (which is the real enhancement.. heh)!

<font color=green> linux is free.. if your time is worthless </font color=green>

 

[Junkkyy]

Yes of course I understand x86-64 can also run 32 bit code, but you're missing the entire point of my posting.

I DON'T WANT 64 BIT ON MY HOME MACHINE! Extensions or otherwise.

Have you ever looked at the schematic for a 64 bit multiplier next to a 32 bit multiplier? It's not pretty. Adding these extensions are going to make the processor very large, even with 130nano fabs. That means cost will be higher than we've ever seen from AMD, and they will generate even more heat than an XP, and will probably debut at a measly 1Ghz. Junky don't play that.

But you're making me argue this trendy issue that I don't want to argue- I'm just trying to state that I (and many others) will NOT want anything but a nice, small, high clockspeed x86 32 bit CPU, and I want to know what AMD's plans are in this market. How much better will their 32 bit processors get? What's after Barton? Should I switch to Intel next time I upgrade if AMD is abandoning this line?

Pretty please don't try to convince me that Sledgehammer or Itanium2 or Sparc or whatever is good or bad, just please let me know if you have any more information as to what's left for the 32bit CPU markey.

Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway

 

[tbirdXPplus]

personal thoughts: AMD should get what`s left over their pros. i`d say....connect the L1 bridge as done in Tbird AHJYA. that`ll give regular users still 133 FSB for a large number of mobos, while people like u guys(i mean people who build their own systems, enjoy tweeking, overclocking ect.) will move on to the 166 or 200 FSB, with new chipsets like sis746, nforce2, kt400 ect. even chipsets like Ali magiK C or sis745, kt333 provides 166 FSB.

i think i`m just being greedy...lol

Be nice.

 

[IIB]

you are an idiot.
you think to highly of your knowledge.

"Pretty please don't try to convince me that Sledgehammer or Itanium2 or Sparc or whatever is good or bad"

I will try to convince you of nothing.

just so you'll know the actull execution logic - such as a multilier is already 64bit. sence it take cares of floating point multipication as-well which is 64bit.
and in genral execution units are a small <10% of a modern processor die size.

This post is best viewed with common sense enabled

 

[tbirdXPplus]

i get what u mean....

just like those crappy 815 chipsets with unwanted 752 VGA cores, or the nforce 420(i mean, if u want to game, who needs a GF2 mx400? and if u need 2D, then who needs nvidia?), including technology that the users aren`t gonna even use. just makes the price more expensive, while giving the users no benifit.

my opinion: amd did put up a good fight with the athlon, but eventually intel will donimate. even though i`m a AMD zealot, i don`t think amd can turn the tide. maybe with hammer. who knows? but i`m doubtful.

u remember when the first power pc CPUs came out, and the alpha CPUs? they had much better performance than the pentium, but most users ignored them. yes, the power pc gave apple a bit more market share, but that`s about it. how long has it been since the 32bit X86 was the mainstream? since 1989, if i remember right(i386). 13 years, too long for stunts like itanium2, or hammer to change the PC system.

i`m just guessing, the hammer will be a extra-performing 32bit CPU, but that`s it. if it`s not in the mainstream, then its 64-X86 structure will not be much use. and from athlon, i`m guessing that the hammer will not have a market share over, say 25, 35%, even lower at the high end PCs.

Be nice.

 

[tbirdXPplus]

P.S. not saying that the nforce 420 is a bad chipset. pretty good, but i`d personally never use the GF2 included.

Be nice.

 

[tbirdXPplus]

plus....is it true that the next generation intel processor will have the PR of 6000? 4MB of cache? if so...i`d sell all my AMD stocks in a hurry. lol

i do hope that the hammer gives a huge improvement. personally the reason i`m waiting for hammer is that by then, barton prices will be pretty affordable.

Be nice.

 

[texas_techie]

Have you ever looked at the schematic for a 64 bit multiplier next to a 32 bit multiplier? It's not pretty. Adding these extensions are going to make the processor very large, even with 130nano fabs. That means cost will be higher than we've ever seen from AMD, and they will generate even more heat than an XP, and will probably debut at a measly 1Ghz. Junky don't play that.

But you're making me argue this trendy issue that I don't want to argue- I'm just trying to state that I (and many others) will NOT want anything but a nice, small, high clockspeed x86 32 bit CPU, and I want to know what AMD's plans are in this market. How much better will their 32 bit processors get? What's after Barton? Should I switch to Intel next time I upgrade if AMD is abandoning this line?

Lets address your concerns one at a time:

1. Have all those extensions will make hammer a large chip putting out tons of heat.
Nope. Ive seen a Hammer. Its actually a VERY small chip. On par with the P4 in actual chip size. Heat? Dunno yet.

2. Debut at 1 gig? Nope. more like 2 gigs.

3. Dont want 64-bit extensions on your home machine? You wont notice. First, you have to have an OS that supports 64 bit. If you dont, neither the OS or the chip will even address the 64-bit registers. In other words, the chip acts like the 64 bit extensions are not there. It wont slow your performance AT ALL.
You act like the 64-bit part is gonna hurt you in some way, it wont.

4. Whats after Barton? They will continue to ramp up the speeds on Barton and XP for a while. After that is the k-9.

Anyway, i think you have the wrong idea about Hammer. Wait until a few months after its debut and it will be significantly cheaper. Even upon release, im sure it will be less expensive than intel's top-of-the-line stuff.

Benchmarks are like sex, everybody loves doing it, everybody thinks they are good at it.

 

[texas_techie]

when it releases, there no 64-bit software for it, ohoh! even when os and software is avail,

Wow shallowboy, thats one of the oldest arguments out there. Fortunately, its NOT true. I forget the link but at least 5 companies (including IBM) have already made compilers for the Hammer 64-bit extensions.
Secondly, Microsoft has already stated its making a 64-bit OS for Hammer. Go listen to AMDs conference call made around Jamuary. They even stated the name of the guy at MS whose heading up the project.


Benchmarks are like sex, everybody loves doing it, everybody thinks they are good at it.

 

[juin]

Show me the IBM compiler for x87-64 i will be happy to see that.

IBM dont need X87-64

IBM have every reason to kill X87-64

The day i meet a goth queen that tell me Intel suck.I turn in a lemming to fill is need in hardware.

 

[LED]

You act like the 64-bit part is gonna hurt you in some way, it wont.

Will hurt you in pocket. And what other reason is there to buy an AMD system nowadays? Price! And that's it. Take low price away and what've you got? (for the 32bit market)

This sig runs too hot.

 

[imgod2u]

No consumer-level software will benefit or is planned to be ported to x86-64. Database software in the workstation/server market is a different story. AMD doesn't have the resources to independently design 2 core designs so they opted for a one-size-fits-all approach. The x86-64 component of Hammer is for the benefit of their entry into the server market. It's not to benefit the consumers in any way. As for the cost of implementing it, I doubt you'll need to worry. Most likely AMD will overcharge for their server-level processors and keep the price of their consumer-level processor low.
As for the enhancements Hammer has, there's actually a bit more than SSE2. There's the added packing stage that helps scheduling of execution significantly. The integrated memory controller will increase latency reduce the reliance on cache. All in all, it should offer a pretty sizable performance boost over the K7 design in modern code. SOI technology will help scalability somewhat so the K8 will ramp up in clockspeed somewhat as well. It's nothing huge either scalability-wise or IPC wise but it's a nice combination of a little increase in IPC and a little increase in scalability.

 

[Junkkyy]

Ok, you're right, hammer will debut at 2Ghz, I shoulda looked into that. But by then Intel has chips at 3Ghz, maybe 3.5Ghz, and suddenly we're measure the clock speed discrepancy in Gigahertz. Hmm, AMD will have to come up with an even more sneaky bizarre confusing marketing plan to try and conceal this.

You're trying to tell me that I won't notice 64 bit extensions, but I think the fact that the processor says "Sledgehammer" on it will be hard to ignore.

Everyone is still missing my point, I don't want to discuss hammer, and everyone is trying to convince me that it truly is better than cold fusion. I don't want it, and I will refuse to buy any processor that has 64 bit anything. Pretty please stop trying to show me the light. There are a million discussions on the benefits/drawbacks, and it doesn't interest me. I am interested in staying with a 32 bit solution for the next 4-5 years, and I want to know if AMD has any plans to improve things in the area after Barton.

Please don't try to convince me that the 64 bit pill is better than viagra and I must be assimilated into the hammer collective or else. I'm not interested at this point in time. Period. I'd rather hear some good insight and numbers into just how far AMD is planning on pushing their 32 bit solutions.

I have a sneaking suspicion that the AMD fans are avoiding this question because the answer is embarrassing. While Intel will have a 5Ghz Pentium 5 in couple of years, AMD may never surpass 3Ghz in the 32 bit market, and more importantly, they may lose even more ground in benchmarks. So suddenly the only possible line of thought becomes "x86-64 will be more benficial to society than penicillin", and to just stop thinking about 32 bit processors altogether. Is this accurate?

I'll ask one more time, how far is AMD going to push their 32 bit processors? What clock speeds will we see, what die size (90nanos?), how big will the cache get, will they ever pump up the FSB? etc. etc. I can't find anything on this topic on the web- everyone seems to have stopped all discourse on AMD's 32 bit products and they instead chant "Hammer, hammer, hammer" ceaselessly.

Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway

 

[LED]

Hammer IS the next 32bit processor from AMD. Yeah, it has 64bit instructions.....use it or not, up to you.

@2ghz if it performs as good as an Intel 5ghz, does that number really matter? Only problem I see with that is trying to get people to believe it. Remember now, even Moores law is according to performance, not just mhz. Yeah, its sneaky, and tricky to put PRs that look like clockspeeds.....but its sorta not fair that they would have a chip just as good, or even better, but thought of as a lesser product because its clockspeed is slower.

ImGod posts that it wont be too big a hit in the wallet, but Im not too sure about that. I think you will be paying for the added instructions, and the 64bit abilities. The overpaying server market will be hit hardest, cause it can be. I think the consumer is going to be told this is the next big advancement in PC technology, and have to pay for it as well. Initial VERY high dollar amounts, then come down according to market needs. If Hammer takes off, expect to pay a premium for a while. But arguing about it is futile, only time will tell.

Once again for Junky.........Hammer is AMDs next 32bit processor.

This sig runs too hot.

 

[peteb]

Except from what I recall it would not be called Sledgehammer, but Opteron or some such, and even if it wasn't it would have been clawhammer for the PC/WS units.

I don't think everyone is missing your point at all, I think everyone cannot understand why you are making such a big deal out of it.

If, in two years time you have two offerings, a NGhz Intel 32 bit CPU that costs $X and a YGhz AMD 64 bit CPU that costs $X ± 10% and they perform the same what will you care?

Why would you not want 64 bit if the performance were equivalent and price equivalent? What you have not yet done is provide a substantial arguement against 64 bit. You are more than entitled to your opinion, and I'm not trying to make you buy 1 or the other, considereing neither exist for sale today anyway, but why are you so polarised against it?

If you just prefer Intel, that's fine, if you are just trying to force the 32 bit issue because you want to make Intel look good and AMD look bad - why? Are you really a key stock holder of Intel in disguise, hoping to keep your portfolio high?

Help us to understand a valid reason that you don't think 64 bit will cut it. So far none of the arguments have been either valid or substantiated.

Remember:

We don't know what hammers will cost
We don't know what 32 bit 5Ghz P4 will cost, nor if they will ever be released as is.
We don't really know what the release speed or performance of a hammer will be on production software and hardware.

So, how given the above can you still argue against it? Sure we can speculate that it might be slower than the best offering from Intel at the time on 32 bit code, but we won't know that for sure, nor the price of either when that is decided.....

-* <font color=red> !! S O L D !! </font color=red> *-
To the gentleman in the pink Tutu

 

[eden]

ImGod posts that it wont be too big a hit in the wallet, but Im not too sure about that. I think you will be paying for the added instructions, and the 64bit abilities.

We never paid too much when the K7 arrived and it had 3 FPUs, much enhanced caching, and so much more. I do agree price WILL be higher than current XPs, but NOWHERE near Intel's.

And yes clock speed does not matter, it's the PR which sells for AMD.

--
An AMD employee's diary: Today I kicked an Intel worker in the "Willy"! :lol:

 

[eden]

Hammer is the 32-bit processor. 64-bit was just thrown. Come back the clock in the K6 days, you could assume the K7 was not a 32-bit alternative or the next one from AMD, but IT IS. My point is you are trying to deny the future man.

As for chip size, it has been said, we saw some Hammer chips on the web, they ARE NOT gigantic and in fact far from the first Socket 423 Wilamettes. BTW extremly small die sizes like 80mm^2 destroy the whole point of heat transfer and optimum cooling.

Barton has 512K L2 and according to texas techie, may have 166MHZ FSB. Rumors have spread Barton will have some Hammer optimizations, so it can explain why per clock the Barton has 600PR over a Tbred.

You seem to be a clock speed freak, I suggest you stroll down to my thread called 1CC=1NS, Also Pipeline question. I think you will grasp why clock speed means NOTHING if it only goes too fast but with no efficiency in each cycle.
AMD can do a 1GHZ CPU, label it PR5000, and it WOULD perform like an Intel P4 5GHZ. It does not matter what clock speed, the efficiency of the units inside, the data fed and the speed and lower latency of fetching and decoding all make a huge difference in how much a rated MHZ processor can perform.

--
An AMD employee's diary: Today I kicked an Intel worker in the "Willy"! :lol:

 

[Junkkyy]

If what you said were true, then AMD will start losing a lot of money.

How can you claim that a Hammer at 2Ghz will outperform a Pentium 5 at 5Ghz, since neither really exist and there are no benchmarks available? Pretty strange claim. You just have so much faith in AMD that you are utterly confident that they will perform something bordering on the impossible when they are losing money and market share like crazy?

Yes, I am also very curious as to the CPU size. Anyone who claims to have seen the CPU would either have to work at AMD, or AMD let them hang out and rip off the heat spreader which hammer's use to get a look at the core. I find either scenario very unlikely. I still think it's gonna be big and expensive.

But now you've got me really worried. If everyone has convinced themselves that AMD's 32 bit solution is the x86-64, then AMD's "solution" does not address the problem, and this always leads to the shittiest of products. I don't really care if I'm using an AMD or Intel processor, but if AMD can't compete any longer than us consumers are the ones who really suffer. I don't think Moore's law can sustain itself if Intel finds themselves too comfortably in control of the market. I'd much rather AMD solves existant problems and remains competitive.

Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway

 

[eden]

I would also like to add, from what imgod2u said, what and about how much is expected from the ClawHammer alone:

-On-die mem controller, is said to carry a solid 20-30% boost in performance.
-SSE 2 will provide even more competition to what is one of the P4's most strongest selling aspects to many IT aware users. With Hammer's powerful FPU, SSE2 apps will just love working with it.
-Improved all around core, and is said around 5-10% more performance overall from it.
-It has 512K L2 cache, or I hope so, as there have been bad rumors of 256K L2 which would be horrible.
-IMO the FSBless architecture of the Hammer is revolutionary, somethign new and fresh, brings a difference and allows people to create new ways to adapt to it.
-HyperTransport to the guy who falsely attacked it, is more effective anytime than HyperThreading for one reason: IT WORKS WITH ANYTHING ANYTIME. HThreading requires software programming support, and has been tested on current P4s and it reduced performance sometimes. Hyper Transport improves the entire speed of data transfer and from what a guy once told us, it will improve even game load speed as the AGP to chipset HT is very fast. Other than that, its 8-bit interface will allow LOW COST manufacturing. So in fact mobos should cost less.
-Hammer's on-die will now no longer require endless searches for the right chipset, and instead now we just look for the mobo and chipset with the features we like.
-PC2700 support.
-12-stage pipeline allows slightly higher frequencies.
-SOI will allow even more ramping (amount is not known)
-IHS will remove the core crushing problems and core frying.
-Thermal protection will be on any Hammer mobo.

These are some of the ClawHammer's improvements alone. SledgeHammer is EVEN MORE. CH is rumored to debut at 2GHZ with a PR of 3400, which should be able to compete the Intel 3GHZ at the end of the year as Intel has just announced. And since the CH is so efficient per clock, it means PR ramping will be more sensitive. So a 66MHZ jump on CH will result likely in 200PR instead of 100, or a tad less.

I will not however hype it, because we have been victims of several disappointing products recently:
1) P4 Willy
2) Parhelia

These had awesome speccs, yet they sucked ass. So I will keep any hype closed in, because even Prescott has some nice sheet specs, but even then, it's Intel, they could anytime just rip off these and just continue ramping clock speeds only. Whatever it is, we cannot hype either, it's not trustable anymore. I will however beleive Texas-techie's claims, he has proven us he does have AMD contacts, and I would beleive any performance claims he has, as long as it sounds credible.
--
An AMD employee's diary: Today I kicked an Intel worker in the "Willy"! :lol: <P ID="edit"><FONT SIZE=-1><EM>Edited by Eden on 07/25/02 11:49 PM.</EM></FONT></P>

 

[LED]

"How can you claim that a Hammer at 2Ghz will outperform a Pentium 5 at 5Ghz"

I never stated that as fact, it was just an example. Even if I stated it as fact, that still wasnt the point. I wasnt saying "AMD ROCKS INTEL", I was trying to make a point that GHZ doesnt always = the better performer. Maybe you should read it again. You're trying to word it in your favor, cause your argument is dead......In this new post your topic has changed, and youre a little vague.....
Whats this mean?

"then AMD's "solution" does not address the problem"

What's the "problem"? Forgive me if I missed it earlier in the thread, but you didnt state the "problem" in this last post.
Moores law will slow down as soon as cost of producing the next big thing gets too be too expensive. So $ stops Moore, or lack of. Moore never said anything about AMD needing to compete in order for CPUs to scale in performance..

This sig runs too hot.

 

[imgod2u]

I would like to make a few corrections:

-On-die mem controller, is said to carry a solid 20-30% boost in performance.

Of the statement's AMD has made, 20-30% is the overall improvement for Clawhammer vs Athlon. This is including the benefits (if any) of the 64-bit extensions and the increase in clockrate (i.e. 2 GHz vs 1.8 and probably 1.9 GHz of the current K7). How much help the onboard memory controller will bring alone is a mystery.

-SSE 2 will provide even more competition to what is one of the P4's most strongest selling aspects to many IT aware users. With Hammer's powerful FPU, SSE2 apps will just love working with it.

The FPU works independently from SSE2, if anything they're 2 competing standards. I don't see how strong FPU would help SSE2 in anyway.

-Improved all around core, and is said around 5-10% more performance overall from it.

Read above statement.

-It has 512K L2 cache, or I hope so, as there have been bad rumors of 256K L2 which would be horrible.

Not really. As I've stated before the onboard memory controller means less reliability on cache. So the Hammer won't need as much cache at all compared to the Athlon, which doesn't really need all that much cache to begin with.

-IMO the FSBless architecture of the Hammer is revolutionary, somethign new and fresh, brings a difference and allows people to create new ways to adapt to it.

Actually, it limits things as motherboard manufacturers no longer have the option to choose what type of memory and what style to implement that memory they can put on their motherboards. It does help performance of the CPU however it pretty much cuts the motherboard manufacturer's job down to making the south bridge.

-HyperTransport to the guy who falsely attacked it, is more effective anytime than HyperThreading for one reason: IT WORKS WITH ANYTHING ANYTIME. HThreading requires software programming support, and has been tested on current P4s and it reduced performance sometimes. Hyper Transport improves the entire speed of data transfer and from what a guy once told us, it will improve even game load speed as the AGP to chipset HT is very fast. Other than that, its 8-bit interface will allow LOW COST manufacturing. So in fact mobos should cost less.

Hypertransport is a pretty impressive interconnect technology but as far as CPU performance, it really is irrelevent. It does help overall system flexibility though. Hyperthreading on the current P4's is very crude. First of all, the hyperthreading arguement is the same as that for dual CPU's. Because essentially that's what it is. The OS sees 2 CPU's instead of 1 and sends 2 threads of instructions at once. In multithreaded software (software that would original benefit from dual CPU's) this would help. Of course, in the current crude implementation, it can also hurt. Probably why Intel chose to disable it on the P4.

-Hammer's on-die will now no longer require endless searches for the right chipset, and instead now we just look for the mobo and chipset with the features we like.

Chipset problems have traditionally been due to the south bridge, not the north bridge, so any problems that once existed will still exist and you will still need drivers.

-PC2700 support.

Hardly anything significant don't you agree?

-12-stage pipeline allows slightly higher frequencies.

Actually, I don't think so. As I mentioned in a previous post, the added stages aren't simply taking currently existing stages and cutting them up into multiple stages (which is what helps clockrate ramping) but rather completely new functions. This does not help scalability as you're not making stages simpler, you're simply adding more stuff to do. It will, on the other hand, help efficiency of the chip somewhat.

-SOI will allow even more ramping (amount is not known)

SOI is probably the only thing that will help ramping of clockspeed asside from the shift to .09 microns. However, the use of SOI does make yields a bitch. I would imagine AMD is having a pain trying to get good yields with it right now (probably the reason why they demoed Hammer at 800MHz).

-IHS will remove the core crushing problems and core frying.

Again, not something new.

-Thermal protection will be on any Hammer mobo.

Finally.