Xbitlabs Xclusive preview on Athlon64.

[Wingznut]

Originally posted by: Snoop
Further, since the design is based directly on the K7, but with a longer pipeline (2 more stages, so around 20 % longer), you would have to be a complete dolt not to think it will eventually match and exceed the current K7's speeds.

Let's not oversimplify things here. Just because a couple pipeline stages are added, does not directly corolate to higher mhz... Especially when we are talking about throwing SOI into the mix.

I think it's probably true that the difficulties with SOI is the main reason for the delays and low clockspeed.

 

[UlricT]

hey but isnt SOI supposed to help ramp up the clock speeds???

/edited for stoopid keyboard!

 

[oldfart]

I do sympathize with what people are saying about the Prescott. If the P4 architechure gained 10% in going from 256 to 512k L2 Cache, I would expect going from 512 to 1MB to weild at least 10, probably 15%, especially with HT. So the 3.4 Prescott is likely to perform like a 3.7-3.8GHz Northwood, and that says nothing about the scaling of Prescott

It is also increasing from 16K L1 to 32K L1.

 

[Avatar26]

Once again, everyone is getting their panties in a bunch before the product is finished! Patience, when the final processor and chipset revisions hit for the launch, we will get some solid, REAL numbers to compare.

 

[Wingznut]

Originally posted by: UlricT
hey but isnt SOI supposed to help ramp up the clock speeds???

Yes, it does... If all works well. By adding the insulation to the substrate, it does help transistor gate insulation (junctive capacitance, in particular.) However, the downside is that it is much more prone to defects and other difficulties in manufacturing.

 

[UlricT]

Originally posted by: Wingznut

Originally posted by: UlricT
hey but isnt SOI supposed to help ramp up the clock speeds???

Yes, it does... If all works well. By adding the insulation to the substrate, it does help transistor gate insulation (junctive capacitance, in particular.) However, the downside is that it is much more prone to defects and other difficulties in manufacturing.

So.... do you think THIS is where AMD has been having trouble? Because nothing else can really explain why they cannot ramp up in speed. Its kind of ironic if you think about it... The design element that was supposed to help the processor clock higher, is holding it down badly. BUT once they get that sorted out, I'm sure we are all in for a treat!!!

 

[classy]

The thing that gets me here is this. Here's a beta chip and platform putting up numbers with some of the fastest chips and platforms you can buy today. A cpu not scheduled for launch for 6 months is clearly showing some serious potential. And get this it was fairly stable during testing. Two months ago the silicon was only getting to 1.2 g at best now its at 1.6. I think AMD has got a a very powerful cpu here that I believe will be the dominate cpu and platform by the end of the year. CPUs are like athletes, nothing can take the place of raw talent. This cpu has raw talent. Intel has improved on a poor performing cpu with tremendous platforms. The P4 has been a work in process for 2 years now, people forget that. I just think this cpu goes against the grain and is bringing talent to the table like nothing before. Not for desktop use anyway. I look at this cpu like the Pentium Pro, it changed the game. I think the Athlon 64 is going to change the game. But even great raw talent sometimes takes a lil longer to develop, but eventually it surfaces.

 

[Acanthus]

Originally posted by: classy
The thing that gets me here is this. Here's a beta chip and platform putting up numbers with some of the fastest chips and platforms you can buy today. A cpu not scheduled for launch for 6 months is clearly showing some serious potential. And get this it was fairly stable during testing. Two months ago the silicon was only getting to 1.2 g at best now its at 1.6. I think AMD has got a a very powerful cpu here that I believe will be the dominate cpu and platform by the end of the year. CPUs are like athletes, nothing can take the place of raw talent. This cpu has raw talent. Intel has improved on a poor performing cpu with tremendous platforms. The P4 has been a work in process for 2 years now, people forget that. I just think this cpu goes against the grain and is bringing talent to the table like nothing before. Not for desktop use anyway. I look at this cpu like the Pentium Pro, it changed the game. I think the Athlon 64 is going to change the game. But even great raw talent sometimes takes a lil longer to develop, but eventually it surfaces.

The original Release date was months ago.

 

[zephyrprime]

So.... do you think THIS is where AMD has been having trouble? Because nothing else can really explain why they cannot ramp up in speed. Its kind of ironic if you think about it... The design element that was supposed to help the processor clock higher, is holding it down badly. BUT once they get that sorted out, I'm sure we are all in for a treat!!!

That's what the rumors are. It makes a lot of sense really becuase what else could the problem be? Some have said the memory controller but why would something as mundane as a memory controller produce problems?

 

[dexvx]

Originally posted by: classy
The thing that gets me here is this. Here's a beta chip and platform putting up numbers with some of the fastest chips and platforms you can buy today. A cpu not scheduled for launch for 6 months is clearly showing some serious potential. And get this it was fairly stable during testing. Two months ago the silicon was only getting to 1.2 g at best now its at 1.6. I think AMD has got a a very powerful cpu here that I believe will be the dominate cpu and platform by the end of the year. CPUs are like athletes, nothing can take the place of raw talent. This cpu has raw talent. Intel has improved on a poor performing cpu with tremendous platforms. The P4 has been a work in process for 2 years now, people forget that. I just think this cpu goes against the grain and is bringing talent to the table like nothing before. Not for desktop use anyway. I look at this cpu like the Pentium Pro, it changed the game. I think the Athlon 64 is going to change the game. But even great raw talent sometimes takes a lil longer to develop, but eventually it surfaces.

Tell that to DEC, and their Alpha procs. They were the best procs of their time period, and beat the crap out of anyone in the competition. In the end the Alpha's died out simply because they could not market it efficiently. Their sole marketting POV was that it was its raw speed.

You keep iterating the fact that this is a technologically superior CPU without indicating why so. Besides the ability to have 64bit address pointers and an onboard memory controller, I dont see the "raw talent" or the reason it will "dominate" by the end of the year. Quite frankly, I think a lot of people were delusional to what it really is like.

The comment on SSE2. It gives less of a reason to buy current generation Athlon XPs because of their lack of SSE2. With all future procs having SSE2, the current generation Northwoods will see a huge performance boost in next generation software. Longevity wise, it does not look good for people buying AXP's.

 

[Sohcan]

Originally posted by: classy

Its the only processor that has its own integrated memory controller.

Incorrect. Integrated memory controllers are neither new nor revolutionary. Some previous microprocessors with integrated memory controllers include:

- Cyrix MediaGX (1998)
- Transmeta Crusoe TM5600 (2000)
- Sun UltraSPARC IIi and III (1997 and 2000, respectively)
- Alpha 21364 (2003)
- Cray X1 (2003)

The only 64 bit capable processors have been for server use.

64-bit processors are not new to the desktop. DEC introduced low-cost versions of the 21064 for PCs in the early 90s, and the Sun Blade workstations using the i-series of UltraSPARCs have been available and cheep for quite some time...the base US-IIi blade workstation costs around $1400. If 64-bit microprocessors presented a fundamental leap in computing, don't you think that previous ventures in the desktop would have been more successful? There was no technological barrier to introducing 64-bit microprocessors to desktops in the last decade...it's hardly a coincidence that the motivation towards 64-bit computing on the desktop coincides with an upcoming need for efficient memory addressibility beyond 4GB.

* not speaking for Intel Corp. *

 

[KF]

There were a couple of preview benchmark sets of the Athlon/K7 before it was released, the chip having been mysteriously obtained by the respected German tech magazine C'T, and it caused a sensation on the bulletin boards. (Of course it would.) The Athlon was a supposed "engineering sample". The floating point benchmarks were what people we starving to see, because the FP on the AMD K6 were very weak, maybe half or a third the performance of a Pentium. (Even so, some people preferred the K6.) To mitigate the low FP performance AMD added 3DNOW! to the K6. The c't benchmarks had the Athlon about half of or a Pentium Pro/PII, which would have been the Athlon's competition. PATHETIC.... Otherwise the Athlon/K7 looked good, and slightly edged out the PII. c't and everyone else when on and on month after month, ad nauseum, about what a disaster the Athon was going to be. But as people may know, when "real" Athlons began to show up, they kicked butt in FP, and became lengendary to gamers mainly due to that aspect.

c"ts reputation is incomparably greater than this Xbitlabs site. I don't know why c't had such confidence that their "engineering sample" was the real thing, if they ever really did. It was sensation for their site regardless.

At this point I see no reason to think the Athlon64 reviewed at Xbitlabs is beta or alpha silicon, or anything more than a plug-in supplied to mobo designers so they can engineer and test the external bus of the Athlon64 with their mobo, like what c'ts "engineering sample" was no doubt supposed to be used for. AMD is not yet passing out samples of the real thing IMO, if only to forstall what Xbitlabs is doing. AMD wants the real performance a secret. AMD wants the true release clock speed unknown. They want as huge a PR blitz as possible to happen at some optimum time which they set. They aren't going to dribble it out and blunt the impact. This will be the event which will set AMD's course for years. This is obvious, isn't it?

I wouldn't expect the Athlon64 to be over 20% better than an Athlon at the same clock; maybe 30% on apps that can benefit from 64 bit mode optimizations (only available in an OS that operates in that mode).The only point io bringing out a consumer version of Hammer is to compete with Intel. Engineers don't just cross thier fingers and hope. They design to a spec, and that spec was AMDs competitor. I would expect Athlon54's SSE2 performance to match the P4 at the comparable rating, or at least comparable price, just because that is what an Athlon64 will need to be competitive. (Maybe it will need 64 bit mode.) If SSE2 is so poor on this sample, I would expect that AMD disabled some things to foil the benchmarkers. (It's not hard to do. Intel had hyperthreading disabled on its P4 chips for a long time, and they were actually sold to the public. )

 

[RaynorWolfcastle]

Originally posted by: KF
There were a couple of preview benchmark sets of the Athlon/K7 before it was released, the chip having been mysteriously obtained by the respected German tech magazine C'T, and it caused a sensation on the bulletin boards. (Of course it would.) The Athlon was a supposed "engineering sample". The floating point benchmarks were what people we starving to see, because the FP on the AMD K6 were very weak, maybe half or a third the performance of a Pentium. (Even so, some people preferred the K6.) To mitigate the low FP performance AMD added 3DNOW! to the K6. The c't benchmarks had the Athlon about half of or a Pentium Pro/PII, which would have been the Athlon's competition. PATHETIC.... Otherwise the Athlon/K7 looked good, and slightly edged out the PII. c't and everyone else when on and on month after month, ad nauseum, about what a disaster the Athon was going to be. But as people may know, when "real" Athlons began to show up, they kicked butt in FP, and became lengendary to gamers mainly due to that aspect.

c"ts reputation is incomparably greater than this Xbitlabs site. I don't know why c't had such confidence that their "engineering sample" was the real thing, if they ever really did. It was sensation for their site regardless.

At this point I see no reason to think the Athlon64 reviewed at Xbitlabs is beta or alpha silicon, or anything more than a plug-in supplied to mobo designers so they can engineer and test the external bus of the Athlon64 with their mobo, like what c'ts "engineering sample" was no doubt supposed to be used for. AMD is not yet passing out samples of the real thing IMO, if only to forstall what Xbitlabs is doing. AMD wants the real performance a secret. AMD wants the true release clock speed unknown. They want as huge a PR blitz as possible to happen at some optimum time which they set. They aren't going to dribble it out and blunt the impact. This will be the event which will set AMD's course for years. This is obvious, isn't it?

I wouldn't expect the Athlon64 to be over 20% better than an Athlon at the same clock; maybe 30% on apps that can benefit from 64 bit mode optimizations (only available in an OS that operates in that mode).The only point io bringing out a consumer version of Hammer is to compete with Intel. Engineers don't just cross thier fingers and hope. They design to a spec, and that spec was AMDs competitor. I would expect Athlon54's SSE2 performance to match the P4 at the comparable rating, or at least comparable price, just because that is what an Athlon64 will need to be competitive. (Maybe it will need 64 bit mode.) If SSE2 is so poor on this sample, I would expect that AMD disabled some things to foil the benchmarkers. (It's not hard to do. Intel had hyperthreading disabled on its P4 chips for a long time, and they were actually sold to the public. )

I really hope you're right because otherwise AMD is gonna be in a world of pain when Prescott launches.

 

[classy]

Regardless of when it was supposed to be released, it is clear that its true release will be this fall. You would think people would learn to not look at preproduction stuff with such a final outlook. Just a few weeks ago the Raptor didn't look like much either, but it finally turned out to be a pretty high performing hard drive. The Athlon was the same. And people keep harping on Prescott, but there is not a peep anywhere on it except for a couple of demos from last year. Early last year I might add. I remember the days of 3dfx vs Nvidia and its the same here with Intel vs. AMD. I give AMD credit for trying something new. Despite what some guy posted earlier, this will be the first high performance DESKTOP cpu with an integrated memory controller and 64 bit capable. Those two things place this cpu at the forefront of breaking new ground. LOL Remember 3dfx, people laughed at them when they brought out full anti-alasing for games. Hmmm they were right and changed the video card industry. So when someone is bringing something to table different I am going to give them the benefit of the doubt. Just like the Pentium Pro. If some folks weren't sold to a brand name instead of moving forward we might be further along. But some of us have and use our own brains. :Q

 

[sandorski]

From page 4 BIOS bootup screen.."VIA VT8235 Sourth Controller" hehe

I think there are a number of things to consider:

1) as already mentioned, this is a pre-production chip
2) somewhat related to #1: It is very possible that Windows doesn't recognize this chip as having SSE2(ala Athlon XP when it first came out and SSE wasn't recognized).

 

[JeremiahTheGreat]

Isn't an Athlon 64 simply a Opteron with 1 HT missing, and single channel ram? I think i read something that all the cores were exactly the same... so what i don't understand is how come an Alpha-sillicon sample is being tested when Opteron is due to be released in a few days!

 

[Rand]

Originally posted by: SexyKBottom line, some people are going to believe whatever AMD says, no matter what, some people will believe what Intel says no matter what, and a small percentage will actaully look at things objectively and say, in this case, thus far, Athlon 64 is an unmitigated disaster (fact: its a year late, fact: its 20% below target frequence with B0 silicon, fact: it's getting blown out by the Athlon XP in many tests, fact: its going to have much stiffer competition from Intel by the time it's released)

I'm not sure I agree with all of your 'facts'.
You state definitively that the Athlon64 is 20% below target clockspeed at B0 stepping, but provide no absolute proof that AMD is unable to yield beyond
1.6GHz. The tested Athlon64 was clocked at 1.6GHz, I see no reason why that necessarily means that AMD is not yielding in reasonable volume at higher clockspeeds.
We know only that one individual processor is clocked at such a speed, we have no information on what general yields are like.

It's a year late is certainly true.. or at least it will be by launch.
But then, the Itanium was a good 5+ years late and is managing to do passibly well.

Yes it is getting blown away by the AXP 2800+ in many tests, but AMD has not officially announced that the final Athlon64 2800+ will be clocked at 1.6GHz, and it seems foolish to believe absolutely nothing can change between now and official launch.
I can well remember a certain AMD Athlon 500MHz that was being outperformed by the Pentium 2 400MHz 4 months before launch... when said processor was officially released many were comparing it to the Pentium 3 Katmai 600MHz. Quite a dramatic performance difference, despite the clockspeed did not change.

Yes it will see tougher competition from Intel by launch, but then Intel will see tougher competition from AMD as the 2800+/1.6GHz is almost certainly not going to be the highest clocked Athlon64 to be released.


When all is said and done I find the results distinctly underwhelming in most respects. It falls marginally below my expectations and well below my hopes.
I find the main memory bandwidth adequate as I don't believe the Athlon64 will need considerably more then 3.2GB/ in most cases, and the extremely low latency is obviously appreciated.

Early impressions lead me to believe it will be no more then barely competitive with the fastest Prescott however... which is what I was expecting, but I'd hoped for more.
Clockspeed/PR rating could change dramatically....and the old Athlon 500MHz debacle has certainly shown me that the present results at 500MHz don't necessarily correlate well to final launch performance.

64bit capabilities mean precious little IMHO, lack of OS support and very few supported applications should effectively ensure that X86-64 is not a major factor on the desktop.
An unfortunate reality as I personally am already seeing benefits to 1.5GB RAM in some applications... X86-32 isnt exactly very effective beond 2GB, and a 4G system wide maximum leaves much to be desired already and I'll need more then 4GB long before 64bit looks to be widely accepted on the desktop. PAE definitely isnt a desireable solution.


I do believe AMD should be served well by the Opteron at least, as I strongly suspect it may scale remarkably well with additional processors compared to it's primary competition in the Xeon/Xeon MP, and should scale well enough in clockspeed to be more then capable of competing in said market.

 

[SiliconGuy1001]

Does anybody know if the heatspreader thing is patented by intel? It would blow for AMD if it was. They'd have to redesign the asthetics completely. lol

 

[Wingznut]

Originally posted by: SiliconGuy1001
Does anybody know if the heatspreader thing is patented by intel? It would blow for AMD if it was. They'd have to redesign the asthetics completely. lol

Probably... But even so, AMD and Intel cross-license A LOT of technology. Something like an IHS wouldn't be an issue.

 

[Lord Evermore]

If it's patented, it'd have to be patented in a very particular way. Just sticking a metal plate on something to spread heat isn't anything new. Of course, the name "Integrated HeatSpreader" they might try to protect if AMD tried to use it. I hope AMD doesn't make up some stupid name for their big metal plate.

 

[SiliconGuy1001]

I really hope that AMD didn't shoot themselves by integrading a memory controller. It seems to me that AMD does not have such a good track record for production, that's why they had to second source them to companies like SIS and VIA. If they run into production or R&D problems with developing the controllers for new memory solutions, then they might have a big problem on their hands. Mind you that memory technology changes very quickly. Intel already has dual DDR, while AMD is still focusing on single channel. They might have to be one step ahead on the next release of DDR-II, otherwise they are headed for the gutters.

 

[tbates757]

Awesome UT2K3 fly-by score, hopefully these things can scale well

 

[SiliconGuy1001]

A logical explanation to the Athlon 64:

Concerning the yield:

Wouldn't it be logical to say that the yield of silicon would be 50% less for a CPU that combines two CPUs into one? What is the transistor count of the Athln 64?

Outlook on the lack in performance:

AMD really had us hoping that this would be a P4 killer, other than it's a beta sample wouldn't it be logical to consider that if a CPU is dedicating half of its resources to one "32 bit CPU" and half to a 64 bit one, then wouldn't the performance theoretically be deminished? A CPU that has both 32-bt and 64-bit to deal with may suffer the very consequences described above.

Perhaps AMD engineers could tweak the CPU to fully devote its resources to one ISA, rather than both 32 and 64 bit. Does anybody know if it continuously waits for 64 bit instructions while handling 32 bit ones? Otherwise its just wasting resources.

Concerning the onboard memory controller:

As much as AMD thought the Athlon 64's onboard memory controller was inovative, it may have very well shot itself. By being the SOLE producer and designer of memory controllers for the Athlon 64, AMD has to obligation to be punctual. As we can see from history, AMD's track record isn't so great for time to market releasing. An example would be the continuous delays of the Athlon 64 itself! How long have we been waiting for this CPU? Weren't we supposed to get it by Chirstmas!? Given that AMD's track record isn't great, it doesn't help to have not any second sourcers. Intel is smart enough to license their architecture to SIS, ALI, and now VIA. All these companies can do for AMD is to produce the north and south bridges. As of now (most likely not in the future), AMD has denied the Athlon 64 from Dual DDR. The memory might very well be strangled by AMD itself.

 

[Lord Evermore]

If a CPU combines two cores in one, then the yield isn't going to be lower based on that. It would be lower based on the simple existence of twice as many transistors and circuits (probably more than twice due to interconnecting). A single core with 500 million transistors would have pretty much the same yield as a 2-core CPU with 250 million transistors per core. Obviously if you have a 2-core die with 500M transistors, it's going to have lower yields than a single core die with only 250M transistors. The Athlon64 doesn't contain two cores though, just a LOT of transistors, obviously making the yield lower just like Intel raising the transistor count makes their yield lower for awhile.

The processor is NOT a 64-bit processor. It's not doing the same thing that the Itanium does, where it has to process 32-bit code differently from 64-bit code. The CPU is still a 32-bit processor, it just has 64-bit extensions that allow it to address higher amounts of memory, and give it more registers and some other things. It's still just a 32-bit processor using the 32-bit x86 ISA. So the resources in the chip aren't "dedicated" to any particular type of processing.

It uses all the resources it can presumably, whether in 32 or 64 bit mode. The OS won't be able to make use of some things I presume if it's only 32-bit, because it can't address 64-bit memory space, but the CPU should be able to make use of the new registers just like a Pentium3 or Athlon makes use of more than the 8 registers of the basic x86 ISA by register renaming.

The onboard memory controller would have been awesome if they'd gotten the part out on time originally. Now it's just falling behind the times compared to everything else. However I was under the impression that it may be possible, either immediately or with the next version, for a chipset maker to integrate a memory controller that takes over and shuts down the A64's controller, so you get the better memory types and speeds, but at the loss of the low-latency from the integrated controller.

Incidentally, whether the A64 needs dual-channel memory or not is entirely dependent on how the internal memory to core bus is designed. I don't know yet whether it is capable of throughput between the memory controller and the core that's essentially limited only by the core speed, or if there's a "bus" that will limit it.

Of course, to AMD they're not "limiting" it at all. They're making the Athlon64 with a single channel because they are focusing it on one market segment, and have another product line that they've put dual-channel into.

 

[SexyK]

Originally posted by: Rand

I'm not sure I agree with all of your 'facts'.
You state definitively that the Athlon64 is 20% below target clockspeed at B0 stepping, but provide no absolute proof that AMD is unable to yield beyond
1.6GHz. The tested Athlon64 was clocked at 1.6GHz, I see no reason why that necessarily means that AMD is not yielding in reasonable volume at higher clockspeeds.
We know only that one individual processor is clocked at such a speed, we have no information on what general yields are like.

It's a year late is certainly true.. or at least it will be by launch.
But then, the Itanium was a good 5+ years late and is managing to do passibly well.

Yes it is getting blown away by the AXP 2800+ in many tests, but AMD has not officially announced that the final Athlon64 2800+ will be clocked at 1.6GHz, and it seems foolish to believe absolutely nothing can change between now and official launch.
I can well remember a certain AMD Athlon 500MHz that was being outperformed by the Pentium 2 400MHz 4 months before launch... when said processor was officially released many were comparing it to the Pentium 3 Katmai 600MHz. Quite a dramatic performance difference, despite the clockspeed did not change.

Yes it will see tougher competition from Intel by launch, but then Intel will see tougher competition from AMD as the 2800+/1.6GHz is almost certainly not going to be the highest clocked Athlon64 to be released.


When all is said and done I find the results distinctly underwhelming in most respects. It falls marginally below my expectations and well below my hopes.
I find the main memory bandwidth adequate as I don't believe the Athlon64 will need considerably more then 3.2GB/ in most cases, and the extremely low latency is obviously appreciated.

Early impressions lead me to believe it will be no more then barely competitive with the fastest Prescott however... which is what I was expecting, but I'd hoped for more.
Clockspeed/PR rating could change dramatically....and the old Athlon 500MHz debacle has certainly shown me that the present results at 500MHz don't necessarily correlate well to final launch performance.

64bit capabilities mean precious little IMHO, lack of OS support and very few supported applications should effectively ensure that X86-64 is not a major factor on the desktop.
An unfortunate reality as I personally am already seeing benefits to 1.5GB RAM in some applications... X86-32 isnt exactly very effective beond 2GB, and a 4G system wide maximum leaves much to be desired already and I'll need more then 4GB long before 64bit looks to be widely accepted on the desktop. PAE definitely isnt a desireable solution.


I do believe AMD should be served well by the Opteron at least, as I strongly suspect it may scale remarkably well with additional processors compared to it's primary competition in the Xeon/Xeon MP, and should scale well enough in clockspeed to be more then capable of competing in said market.

Great response Rand, thanks for the infusion of reason.