The REAL reason AMD is in trouble is because of their laziness with chipsets. Who hasn't tired of the stupid PCI bus and other chipset problems with Via, etc. AMD does make a better CPU, but they have trusted Via and others for too long for chipsets and are now going to pay the price. After all the Via problems of the last few years with the 686b and up chipsets, who is going to trust them with Hammer? Not me. The problem is, it takes too long to determine the real stablity of a chipset. Just look at the issues just now coming to light about the nVidia chipsets -- a year later? Sure Intel has had it's problems with chipsets in the past, but for some time now seems to have fixed them all. Who would argue that there are stability or feature problems with the Intel 845 series chipsets? The only AMD system I would purchase today is the dual processor ones based solely on AMD's own chipsets. Problem is, I and most others do not need dual processors and don't want to pay the several hundred dollar premium. Unfortunately, Intel is looking better every day.

PatrickR, you should read the qualifiers - maybe. AMD/Athlon/Hammer utilises a lot of ex-Alpha tech and idea's. Go have a look at the specs for Alpha EV7. Lot's of Rambus channels.

Having said that, RDRAM has a bad odour and doesn't look like becoming a commodity. But things might change. I would be surprised if AMD didn't have a few samples floating around that sported quad-channel RDRAM. If purely from an R&D perspective.

Multiple channels can hide latency on predictable data, but only looses out to random memory accesses. This hasn't hurt the P4 too much as nobody says the P4 sucks because it performs poorly on office-type applications (a common situation where most memory reads will be random).

But don't go and quote Sysmarks Office scores and say it doesn't suck becasue SysMark artificially promotes heavy bandwidth usage in their Office tests (by the way in which they are run and measured) - the end result of which advantages an architecture with memory bandwidth. Look at Winstone 2001 scores instead.

What I'm getting at is the P4 isn't bagged for having poor latencies - because the types of applications which generally require them are more than satisfied by todays hardware anyway. And if Hammer went down that path, it would still have better latency characteristics than Intel anyway. Now that I think about it (thanks PatrickR), Opteron (dual channel DDR) should be an absolutely awesome database server. I guess we can look forward to another of AnandTech's "site upgrade" articles...

Addressing the 2-3 sticks of RAM issue...

I believe there was a motherboard review, perhaps a Tyan, just a short while ago that had four slots for DDR. However, if you populated all four slots, it would drop into 100Mhz mode, for signal integrity. You could run 2 or 3 at 133, but if you added the fourth, the mobo would automatically slow everything down for stability. Given that was only PC2100, I find it very easy to believe we are reaching a limit on DDR. sure, I could get PC3000 or higher, but the memory is really reaching it's limit, much like rambus was when it first came out (heat spreaders and fans on the memory, anyone?).

I guess that's one reason I never really liked rambus...

Originally Posted by patrickr
rdram?? Are you sure about that....I though the whole point of the hammer was to have reduced latency times. Why would AMD opt for high latency memmories? I though DDR-II was suppose to be better anyways. Why won\'t AMD release the hammer with out of box support for DDR 400 and DDR-II? It kind of sucks that I would have to by all this new equipment just to switch ram type...But hey I guess that\'s technology

None yet - I was talking about the limitations of DDR and why some (early) boards will only sport 2 DIMM's. But I'm sure that DDR-II is in the Hammer picture - maybe even RDRAM.

Samsung apparently has engineering samples of DDR-II, but I would think that it is twelve months away at least. This also does expose the limitation of on-board memory controllers. With Athlon, you could move the chip from an SDRAM board to a DDR board.

With Hammer, that won't be possible. It'll be new chip, memory and new mobo.

Originally Posted by patrickr
im confused..which hammers will be able to support DDR-II?

OK 37

..."Hate to tell I have both a Asus A7V333 and a giga-byte 7vrxp+
w/ 3 512mb sticks of samsung DDR333 and the both boards see all the memory at the correct speed of 333 as far as signal integrity you don't see the problem until you have four slots and only use slot one w/ a big stick (i.e. 1gb) better to use two 512 for better performace "...

Specific experience v sweeping generalisation. Some companies are better at R&D than others. Some solve them for some cases but not for all. Because you can run some DIMM's in all slots @ max doesn't mean the problem isn't there. In fact you have validated (perhaps) my comments by saying that the problem does appear with large capacity DIMM's. This too will be fixed in later versions of both the board and memory. The only way to know for sure (if the problem doesn't exist) is to run a logic analyser over it. As time goes on and boards a revised, the problem will become less significant, or even be solved, until faster memory appears. It's a never ending cycle.

But my original post was in response to some who saw only two DIMM's on some boards as a problem. This is funny in two ways becasue most likely some (most) of the boards sporting two DIMM slots at Computex were non-functional. They could have put 16 slots on them for all the good it would have done...

The AMD760 had a limit of only 2 DIMM slots to ease the load on board designers - and it was successful even though people poo-pooed the idea. Such will probably be the case with the AMD8000. Then other chipsets will emerge as designers get clever and the limit will ease. But things should get much better with DDR-II for aforementioned reasons.

"Why only 2 DIMM's? Signal bounce/integrity. Even the KT333 boards you can buy now that have 3 DIMM's only allow you to use 2 @ 333MHz because of this issue. This is something that will really only be solved with DDR-II (trace termination), so I don't see this as an issue."

Hate to tell I have both a Asus A7V333 and a giga-byte 7vrxp+
w/ 3 512mb sticks of samsung DDR333 and the both boards see all the memory at the correct speed of 333 as far as signal integrity you don't see the problem until you have four slots and only use slot one w/ a big stick (i.e. 1gb) better to use two 512 for better performace

Bloody Lava tags....

dmonsees,

the hammer doesn't have a front side bus because there is no external northbridge/memory controller. What it doesn have is a point-to-point link in the chip itself, that according to AMD whitepapers is 800MHz. I don't think it is a HT link either as the protocol itself may not be suited to this task.

Even if you want to use the "800MHz" as a psuedo-FSB "speed" that doesn't tell the whole story as the on-chip memory controller has much lower latencies - so even if you wanted to compare Hammer "psuedo-FSB" with some other products FSB, the Hammer is still a winner. HTH

P$ is now only slated for 667MHz FSB in next rev AFAIK.

Why only 2 DIMM's? Signal bounce/integrity. Even the KT333 boards you can buy now that have 3 DIMM's only allow you to use 2 @ 333MHz because of this issue. This is something that will really only be solved with DDR-II (trace termination), so I don't see this as an issue.

Opterons use registered ECC memory so they aren't as affected - but the memory latencies are higher and the price is higher - you gets what you pay for.

64-bits. Don't really need it for home use - but AMD's plans is for one chip to satisfy a broad range of markets, and bring economies of scale to bear. Consider it a freebie. Any software compiled (by a compiler and not hand tuned) will benefit fairly significantly from the extra registers as compilers arent as good at scheduling code as good assembly writers. 64-bit code will be it's nature be slower than 32-bit code (size, reduced amount of instructions will fit in a cache), but the extra regs will more than offset this.

Back on topic. AMD realised a long time ago that this would be a difficult time. That is why you aren't seeing increased cache, SOI, faster FSB. All of this takes time, effort and money to do and (this is the important bit), every time you introduce architecture changes, yields take a hit. Lower yields means it costs more to make each chip. Which means either AMD's profits take a hit or AMD has to charge more. AsS AMD won't be able to charge more because Intel willl be fighting an aggresive price war, profits will take a hit. Most profits come from the top two speed grades. Did you notice that Intel dropped all P4 prices last week except] the speed gradde where AMD doesn't have a competitor? That was no accident. Intel are making sure that AMD gets very little cash for their chips but doesn't cut too much into their own profits. So AMD's strategy (identified a long time ago for this period) has been clear. Make Athlons as cheaply as possible. Unfortunately this means no new or exciting tech. Notice how the promised new/improved Athlons appear only after Hammer is established?

By then, profits should be healthier so they can devote the time and resources into boosting the Athlon - which they will still need till at least the end of 2003, and Intel will be pressuring with faster Celeron 4's with more cache etc.

My take anyway...

Anon. Coward: 666 FSB for P4 doesn't impress me much. 400->533 brought a big boost in performance, but I think it was more because a bottleneck was removed. 400->533 = big boost, 533->666 = smaller boost. We'll have to wait and see, though, I suppose.

#31: the FSB speed of Satan? I think so!

Originally Posted by Transman88
Hammer, Hammer, Hammer.....geez, all I really want right now is Thoroughbred. This Athlon B 1.0ghz is overclocked to the hilt, I\'d REALLY like a .13 XP in this 8kha+.

Hurry up T-bred!!!

Anand says:

666mhz FSB and 1mb L2 Prescot will have.

(OK, he said it [i]ma/i] have the 1mb L2. He also noticed that AMD's non-Opteron sample chips have 1mb L2.)

Yoda says:

Great Processor K8 will be

Nobody (with a brain) thinks AMD will close their fabs during the next 6 month to come.

But it WILL be a tough time for them, no matter how you personally like or dyslike AMD.

BTW: I heard rumors that Intel will have a x86-64 processor ready in 2003: The Prescott core could include such a feature + SMT. We will see...

think on monday

does anyone know when the 2200+ is scheduled to be released?

sroylance:

[q]Database servers and scientific computing desperately need 64-bit CPU's. I'm sure plenty of people here have 1-2GB of RAM in their systems now[/q]
Yeah, but does Dell really care? Why shouldn't they just push Itaniums with Intel's big discounts, big name, and big adversising budget. As for the rest of us, I run 512mb, 768mb and 1.5gb myself, and sure as heck can't see how I can possibly double that by Xmas. RAM has, like hard drives, progressed into excess. Now I need CPU power, memory speed and graphics speed. Sure AMD's 64-bt extenstions expose new registers for custom-compiled programs, but how will that compare to Intel's P4 at the time? People talk about "64 > 32" as if it matters, what about P4's ultra-high future core clockspeed and /FSBmemory speeds? Or on the server front, what about Itaniums and their super large tripple-layer on-die cashes? That [iound/i] pretty fast too!

Transman88: [i](regarding 4gb of RAM with Hammer)[/i]

[q]Not likely, with Hammers' two memory slot limitation. Those would be some big, expensive DIMMS your talking about.[/q]
My definition of "enthusiast" includes multi-cpu, and hence, more slots for RAM. :)

sroylance:

[q]Wha? I can't find the article, but I think that the demo 2-CPU hammer system had like 16 DIMM slots. It's a weird NUMA-like memory architecture that I think required paired population of banks, but I don't think it's limited to 2 RAM slots.[/q]
Don't conduse the dual-channel Opteron chips with their lesser desktop brethren (or the dual CPU, one channel each Opteron, either).

Well, none of the boards currently on display are dual processor boards, right? I.e., they are targeted at the desktop, not servers. I expect server class boards to have more (usuable) memory slots.

Originally Posted by Transman88
Dunno. Got my info from overclockers.com. Guy over there brought up the point that all the motherboards they have on display at Computex all have 2 slots on them anyway, and those that have 3 will only work with 2 of the slots populated.

[q]Not likely, with Hammers' two memory slot limitation. Those would be some big, expensive DIMMS your talking about.
[/q]

Wha? I can't find the article, but I think that the demo 2-CPU hammer system had like 16 DIMM slots. It's a weird NUMA-like memory architecture that I think required paired population of banks, but I don't think it's limited to 2 RAM slots.

-Steve

Can't it support > 2 DIMMs if registered memory is used, just like some of the current Socket A boards?

Company I'm working for right now has an application that could benefit from >4GB of RAM...

Originally Posted by Transman88
[q]>4GB in an enthusiasts system might be realtively common by the time hammer hits its stride. [/q]

Not likely, with Hammers\' two memory slot limitation. Those would be some big, expensive DIMMS your talking about.

AG#15:
Database servers and scientific computing desperately need 64-bit CPU's. I'm sure plenty of people here have 1-2GB of RAM in their systems now, >4GB in an enthusiasts system might be realtively common by the time hammer hits its stride.

-Steve

AMD is not in trouble yet. This is just the calm before the 64 bit storm. True, most people don't need 64 bit computing, but I could probably make the case that most people could get by on 8 bit machines since all they do is word processing and small spread sheets. 64 > 32 so it will carry the day. You don't need a degree in marketing to figure it out. Should there be any doubts, MS is going to spend millions (if not billions) to plug XP-64 and stimulate demand. AMD has a really good strategy so long as they can execute on schedule, and they look to be roughly on schedule so far.

Intel may have reclaimed the desktop crown for the next 6 months, but the damage is already done. The success of the Athlon has breathed new life into AMD, a company that should have been vanquished along with Cyrix. Now Intel is left with 32 desktop chips and a 64 bit server chip that is ungodly expensive to produce and lacks (in practical terms) backward compatability. Intel is caught in a trap. If they adopt x86-64 they legitimize AMD, but to ignore it they will likely wind up losing the emerging 64 bit desktop market.

Worse yet for Intel, IA-64 is yet to set the heart of the enterprise aflutter. Old RISC unix boxes still rule the day. Displacing Sun in the enterprise will be quite a task, especially with the sinking husk that is HPQ leading the fleet. Indeed, Intel may have overestimated the influence in the high end market.

what a bunch of internettards we got here.. LEARN TO READ...

I had to read this thing about 4 times just to make sure I wasn't missing the part you llamas are reading... Nowhere does it say AMD is going away, he doesn't even speculate... all it said was it will be a HARD 6 months for AMD...

damn what a bunch of gerbils... even you registered guys need some help... with the exception of the first 6 or so posts that is...

Later

I agree with most of the senitiments here. AMD will weather this fine. They aren't idiots, to have come this far and completely blow it, the hammer is exactly where they need to put resources now, so that one year from now their CPU line remains viable and strong. Their reply to "hollow megeahertz" is "64 bits because we can". Smart move on both company's parts. They each have something to sell.

Intel is doing great, and will continue to do so. But don't forget who pushing them.

I think that this is a rather pessamistic view of AMD. I don't think the next 6 months is going to make or break AMD. The Athlon is selling well and if anyone has had supply problems it is definitely Intel and not AMD. AMD has pretty much shipped processors as soon as they have been announced unlike Intel.

My only concern for AMD is chipset and processor stability for the athlon hammer. I am also concerned about Microsoft shipping a consumer version of win xp 64 on a timely basis. If those 2 concernes turn out to be no problem at all then AMD is going to have another hit on their hands. AMD can coast on the hard work they did on the athlon for the next six months. It is Still the best bang for the buck by far and probably will continue to be for the next 6 months.

Don't get me wrong Intel has come back in a big way with there p4 pricing schemes and better chipset support. They definitely deserve credit for executing products on time and without having to recall anything.

In my exp. it seems that once people get over the Professional Grade FUD that is spread on message boards and build or buy there first AMD systems they soon realize that there is not much difference other than price. You have to be a real DOLT IMHO to pay more for the same sh!t.

Yeah right, all this "64 bit is needed" talk strikes me as ignorant. Why the hell do we need 64 bits all of the sudden? Cause it exists? Intel has [i ot/i] of things that could keep the Dell's of this world happy, what with the 800mhz FSB on their next P4 that I keep reading about (for example). That and the usual Intel brand name and advertising game.

It also seems to me that all this worrying about a tough half a year of AMD is silly... they've seen worse. The Athlon is not about to start looking like a K6-2.

I still don't know anyone who owns a P4. The primary home machines around here are powered by at least 8 Athlons, 3 Durons, 3 P3's, two G4's and a Celeron... and one of the Duron guys wants to go to an Athlon.