Xbitlabs Xclusive preview on Athlon64.

[Nemesis77]

Originally posted by: Lord Evermore
B0? Early? In an age when A0 chipsets are being shipped on retail products?

Considering that the launch of the chip is still about 5 months away, I think it's fair to say it's an early revision. That chip was made in january, 9 months before the official launch. There will be few new revisions before the actual launch IMO.

 

[Lord Evermore]

But as has been pointed out repeatedly, they're WAY behind their original target already. If they're already up to B0 and that's STILL such a lame-dog for performance, what's to make anybody think they're really going to get something good by September? Even though it may be an "early" revision because they're going to go through many more revisions before launch, it's still relatively poor in the eyes of the people who expect B0 to be at least something close to ready.

 

[Nemesis77]

Originally posted by: Lord Evermore
But as has been pointed out repeatedly, they're WAY behind their original target already.

Yes they are. So?

If they're already up to B0 and that's STILL such a lame-dog for performance,

Is B0 a late revision? Fact is, that they still have 9 months to tweak that core before it's launched. And besides, we all know Via isn't the fastest platform out there.

what's to make anybody think they're really going to get something good by September?

It's already pretty good IMO. And early benchmarks of original Athlon weren't that good either.

Even though it may be an "early" revision because they're going to go through many more revisions before launch, it's still relatively poor in the eyes of the people who expect B0 to be at least something close to ready.

B0 isn't meant for general public, so it doesn't really matter what general public thinks about B0 silicon. What matter is the end-product. For all I care, early revision could be as slow as 486, all I care is the performance of the chip that they are going to ship.

 

[Lord Evermore]

If a company ships B0 products, as does happen pretty commonly now, then they're meant for general public. Since I assume that when Intel reaches B0 stage, it's probably gone through as much work as another company's product that has reached B0 (pretty much every site looks at the revision for any brand as if they're all similar in "doneness" at a certain point), then I'd assume that AMD has done a certain amount of work and reached B0 and still doesn't have a very good product considering what it's intended for, and the fact that they'll still be another 5 months finishing it meaning something like a year and a half of delay, just doesn't inspire confidence. This isn't about anybody saying with assurance that they'll either have a great product in September or they won't, it's about whether they APPEAR to be on the road to meeting that deadline and having a great product by then. These numbers and the delays just don't make anybody really think there's much of a guarantee of it happening; if we were betting on it, most people would be a little less confident in betting on AMD because of past performance and knowledge of what they have right now. That doesn't mean if it works great and comes out in September, they won't buy it. It just means right now, people are considering that maybe they won't be blowing our minds like they first expected.

 

[Nemesis77]

Why don't you just wait and see how the actualy SHIPPING PRODUCT performs, instead of damning the entire CPU based on a prerelease chip running on prerelease chipset?

 

[PrinceXizor]

<Puts his boots on before entering this forum thread>

1. Even though this is a very early preview on a non shipping product. The results are still marginally disappointing at this stage. Yes, AMD has 9 months to tweak, and yes the original Athlon had bad early benches. Guess what, the benches were and still ARE being referenced as "BAD" benches despite the "GOOD" turnout of the final shipping product. The fact that the final Athlon64 COULD be and probably WILL be a better performing chip doesn't negate the fact that these preliminary scores are "bad", "worrisome", "disappointing" or whatever.

2. For those claiming the shipping product can't be THAT much better. How do you know? You are all probably the same nay-sayers that claimed there was no way a new firmware would make THAT much difference in the whole Raptor pre-release hoopla. The same thing happened on the original Athlon, the Raptor and probably countless other hardware products that I don't know about. Let's wait for final production "reviews" before condemning this chip to oblivion.

3. As far as a delay, this will enable them to 1. Continue to improve the yield on their higher frequency cores. 2. Wait for a commerically available 64-bit OS from the boys in Redmond. 3. Perhaps it will allow them time to rework the internal memory controller to function at 400MHz DDR speeds. Of course, these are just all my conjections and opinions.

P-X

 

[Nemesis77]

Originally posted by: PrinceXizor
3. As far as a delay, this will enable them to 1. Continue to improve the yield on their higher frequency cores. 2. Wait for a commerically available 64-bit OS from the boys in Redmond. 3. Perhaps it will allow them time to rework the internal memory controller to function at 400MHz DDR speeds. Of course, these are just all my conjections and opinions.

P-X

Athlon 64 already works with DDR400.

 

[PrinceXizor]

Originally posted by: Nemesis77

Originally posted by: PrinceXizor
3. As far as a delay, this will enable them to 1. Continue to improve the yield on their higher frequency cores. 2. Wait for a commerically available 64-bit OS from the boys in Redmond. 3. Perhaps it will allow them time to rework the internal memory controller to function at 400MHz DDR speeds. Of course, these are just all my conjections and opinions.

P-X

Athlon 64 already works with DDR400.

Hmm...indeed, somehow I missed that in the article. Faulty reading on my part. Let me substitute DUAL Memory Controller for #3

P-X

 

[Lord Evermore]

Opteron is just Athlon64 with a dual channel controller. Well, other things, but that for one.

I'm not condemning the final, shipping product at all. There's still a good chance they'll come out with a whiz-bang product, and I expect that to happen. But there's also a chance that it won't be quite as good as we might have hoped, sort of like the GeforceFX of the CPU world. And given the PR system, they really CAN'T come out with a bad product -- they just have to reduce the gap between actual frequency and PR number, and make sure the clock frequency is ramped up enough to have PR numbers matching Intel; obviously, if they just called it a 1.6GHz CPU and priced it lower than a 2.6GHz P4, they'd have a winner because it's going to perform WAY better than a 1.6GHz P4. What's really disappointing is that they're seemingly stretching the PR system as much as they can, in order to get PR numbers as high as possible with the lowest possible frequencies; rather than giving us a frequency which guarantees the performance will ALWAYS be at or above the PR number, they give us a stretch where maybe in some places it's not REALLY a 2800+ performance. Of course, the final product could be so significantly better that this becomes completely untrue; then again, it might not.

All I'm saying is these benchmarks are a bit disappointing given how long they've already delayed, and don't LOOK very good in terms of how much better it could get with only 5 more months. We expected to be seeing these sort of benchmarks and performance levels a YEAR ago, and they're just getting to that point now. Whatever the reasons for the delays, all it looks like to someone considering the Athlon64 as an option is they're having a much harder time reaching their goal, so MAAAAAAAAAAAYBE we'll go ahead and start looking at other products rather than continuing to be sure that Athlon64 will be what we want to use. It's just a seed of doubt for some people, where previously they may have had little to no doubt.

 

[SexyK]

Originally posted by: Nemesis77
Why don't you just wait and see how the actualy SHIPPING PRODUCT performs, instead of damning the entire CPU based on a prerelease chip running on prerelease chipset?

Man, I just don't understand where you're getting this stuff. No one is "damning the entire cpu" based on this article. All that anyone has said is that for this stage of development, these benchmarks are disappointing. Is that really so unfair? I don't think so.

 

[Duvie]

I thought it stated that the opteron has a built in dual channel 333ddr controller but the athlon64 (the desktop model) is only a single channel ddr333....

Check out the article at TECH-REPORT

One thing that needs to be noticed is this B0 chip is basically an opteron with the single channel memory controller yet the opteron is out at the same 1.8ghz (shipping now).....The opteron alos in reviews I have seen under desktop situations is performing a lot like xbitlabs has shown. In many instances this chip will need to optimized 15-20% or increase speed to minimum of 2ghz just to stand even with already released p4 3ghz 800fsb HT chip. I have no doubt they can do that, but haven't we all been bombarded with so much hammer hype that just being equal to intels best or just a marginal fraction ahead will be considered rater disappointing??? I mean we have heard reports of the hammer going to destroy the p4 yet the biggest threat to the hammer is still yet unknown as too its performance increase over the existing northwoods. If the prescott comes in at 3.2-3.4ghz w/ 1mb of l2 cache, increased l1 cache, .09 process, enhanced HT, are we not to assume that could be a ggod size gain over current higher perfromer from Intel???


The one thing I agree about is the platform...yes via was used and I am sure an nforce board could enhance performance, but how much will it do that when the memory controller is onchip??? that is one of the biggest enhancements I see in chipset upgrades and nforce boards over via chipset....

 

[Nemesis77]

Originally posted by: Duvie
I thought it stated that the opteron has a built in dual channel 333ddr controller but the athlon64 (the desktop model) is only a single channel ddr333....

Like I said, Athlon 64 works just fine with DDR400. DDR333 might be the fastest thing with official support, but DDR400 works just fine as well.

One thing that needs to be noticed is this B0 chip is basically an opteron with the single channel memory controller

We don't know that for sure untill Athlon 64 is released. I wouldn't be surprised if the mem-controller in Athlon 64 is more tweaked than the one in Opteron.

The opteron alos in reviews I have seen under desktop situations is performing a lot like xbitlabs has shown. In many instances this chip will need to optimized 15-20% or increase speed to minimum of 2ghz just to stand even with already released p4 3ghz 800fsb HT chip.

It performs just fine on desktops. Reasons for slowdowns might be the poor current support in Windows and no x86-64. And that's why they are pushing Opterons to servers right now, because servers can take advantage of NUMA and x86-64 right now. Windows-desktops will follow a bit later (although you could run it on Linux right now and get all the benefits)

but haven't we all been bombarded with so much hammer hype that just being equal to intels best or just a marginal fraction ahead will be considered rater disappointing???

Um, no. The reviews I have seen in Aces shows that it usually considerably outperforms the Xeon. And that's WITHOUT x86-64!

If the prescott comes in at 3.2-3.4ghz w/ 1mb of l2 cache, increased l1 cache, .09 process, enhanced HT, are we not to assume that could be a ggod size gain over current higher perfromer from Intel???

Of course Intels CPU's will improve. But same will happen to Opterons and Athlon 64. New revisions would have tweaked mem-controllers for example, and that would boost performance, clock-speed would go up etc. etc. And die-shrinks could bring even more improvements.

And I do think that Athlon 64 will move to 2x64bit mem-bus in the future.

I am sure an nforce board could enhance performance, but how much will it do that when the memory controller is onchip??? that is one of the biggest enhancements I see in chipset upgrades and nforce boards over via chipset....

AGP-controller? IDE-controller? Yeah, those have exactly zero relevance when talking about performance.

 

[alexruiz]

I am personally satisfied. The results show the potential is there.

Some comments:

1) If the CPU is based heavily on the Athlon, it should have ALL the Athlon strenghts.... but is being beaten very badly in some cases by a humble Barton..... why?
2) Workstation performance is less than stellar, but I suspect is because the SSE2 programs ARE NOT recognizing/using the SSE2 capabilities (sysmark / wme anyone?)
3) I think the market segmentation AMD is doing is smart.... single DDR channel for mainstream, dual for high end. Time will prove it.
4) It is only a sample....

So, let's wait and see..... I am still betting on the clawhammer and the preview gave me more confidence.... gaming performance is higher, workstation performance is bad but we don't even know if the applications are seeing the processor correctly.....

 

[Duvie]

Nemesis, read the review I linked....it talks about how the architecture of the athlon 64 relies on the hyperlink and packaging info to be sent to agp controller through what it calls a serial like port. It also talks about how this has possible extreme latency penalties.

The sad thing to remember in many of the mp3 and mpeg2 stuff most of those are not heavily or not even SSE2 optimized so there is no excuse they run slower then xcurrent 2800+ barton and around the performance of a 1.6ghz athlon as it did in one test.

I have said and agree this is a preview but I have the right to comment on this "midterm report"...there is definitely some work to be done...

 

[PrinceXizor]

Actually, there IS an "excuse" or more appropriately a reason that certain tasks are performed better i.e. faster on Intel architecture. As should be evident by now there is no one "best" way to design a microprocessor. There are multiple concepts that can be used. Intel uses a longer pipeline and higher frequency (yes, this is way oversimplified) and AMD uses a shorter pipeline and lower frequencies.

The point is, the current K7 microprocessor family has its strength in FPU and in situations that favor its shorter pipeline and thus its lesser penalty for branch mis-predictions. In situations where branch mispredictions are low and thus the full "theoretical" bandwidth is being utilized most efficiently the Intel microprocessors SHOULD excel, because that's what its architecture was designed for.

People need to actively consider what they truly use their computer for.

Me, I use it mostly for games and general business items. Therefore an AMD processor is currently the best fit for me. If I did a large deal of media encoding, I'd get an Intel based system. It's just like a car. I'm interested in horsepower, torque, handling, etc. but in the end, I pick the car that best fits my budget and my needs. It doesn't mean its the best car period, just the best car for my particular circumstances.

P-X

 

[Duvie]

Originally posted by: PrinceXizor
Actually, there IS an "excuse" or more appropriately a reason that certain tasks are performed better i.e. faster on Intel architecture. As should be evident by now there is no one "best" way to design a microprocessor. There are multiple concepts that can be used. Intel uses a longer pipeline and higher frequency (yes, this is way oversimplified) and AMD uses a shorter pipeline and lower frequencies.

The point is, the current K7 microprocessor family has its strength in FPU and in situations that favor its shorter pipeline and thus its lesser penalty for branch mis-predictions. In situations where branch mispredictions are low and thus the full "theoretical" bandwidth is being utilized most efficiently the Intel microprocessors SHOULD excel, because that's what its architecture was designed for.

People need to actively consider what they truly use their computer for.

Me, I use it mostly for games and general business items. Therefore an AMD processor is currently the best fit for me. If I did a large deal of media encoding, I'd get an Intel based system. It's just like a car. I'm interested in horsepower, torque, handling, etc. but in the end, I pick the car that best fits my budget and my needs. It doesn't mean its the best car period, just the best car for my particular circumstances.

P-X

Hmmm....I wasn't stating about two different platfroms like Intel and AMD as that is quite obvious to me....example doesn't work...I am stating why in several instances the Barton 2800+ an AMD chip was quite drastically ahead of the Athlon64??? If the case is about pure IPC and the athlon64 only running 1.6ghz while the 2800+ is over 2ghz this leads to the problems many have with AMD's PR rating!!!

 

[Nemesis77]

Originally posted by: Duvie
Nemesis, read the review I linked

The Tech-report article? It's not a review

it talks about how the architecture of the athlon 64 relies on the hyperlink and packaging info to be sent to agp controller through what it calls a serial like port. It also talks about how this has possible extreme latency penalties.

There might be concerns, but it's nothing that can't be fixed. Wait 'till they ship the product OK.

The sad thing to remember in many of the mp3 and mpeg2 stuff most of those are not heavily or not even SSE2 optimized so there is no excuse they run slower then xcurrent 2800+ barton and around the performance of a 1.6ghz athlon as it did in one test.

In the xbit-labs tests? Because the chip was prerelease? Because the MoBo and the chipset were prerelease? Because the chipset-drivers were beta? There are SO many reasons, that one shouldn't even think about these kinds of things! We have all seen the huge impact drivers can have on performance. And here they are benchmarking a beta-product with unfinished drivers.

Again: you can not base your opinions regadring the performance of Athlon 64 based on the "review" at xbit.

I have said and agree this is a preview but I have the right to comment on this "midterm report"...there is definitely some work to be done...

Of course there's still work to be done! And that's why this "review" is meaningless when talking about the performance of the final shipping product!

 

[sandorski]

Concerning Desktop performance: When Win64(or whatever it'll be called) is released, Desktop Performance will increase. Some reviews have explained why, such as Memory use methods and such.

 

[Duvie]

Originally posted by: Nemesis77

Originally posted by: Duvie
Nemesis, read the review I linked

The Tech-report article? It's not a review

it talks about how the architecture of the athlon 64 relies on the hyperlink and packaging info to be sent to agp controller through what it calls a serial like port. It also talks about how this has possible extreme latency penalties.

There might be concerns, but it's nothing that can't be fixed. Wait 'till they ship the product OK.

The sad thing to remember in many of the mp3 and mpeg2 stuff most of those are not heavily or not even SSE2 optimized so there is no excuse they run slower then xcurrent 2800+ barton and around the performance of a 1.6ghz athlon as it did in one test.

In the xbit-labs tests? Because the chip was prerelease? Because the MoBo and the chipset were prerelease? Because the chipset-drivers were beta? There are SO many reasons, that one shouldn't even think about these kinds of things! We have all seen the huge impact drivers can have on performance. And here they are benchmarking a beta-product with unfinished drivers.

Again: you can not base your opinions regadring the performance of Athlon 64 based on the "review" at xbit.

I have said and agree this is a preview but I have the right to comment on this "midterm report"...there is definitely some work to be done...

Of course there's still work to be done! And that's why this "review" is meaningless when talking about the performance of the final shipping product!

Kind of a broken record with nothing much to add, huh???? Nothing is meaningless to talk about....Whether the opteron shedded light on a small factor of the athlon64's performance is worth discusssing and then when it is released we can see whether or not optimizations helped or whether it helped just some things.

It is a preview!!! Get it through your head!!!! We all know there can be advances!!! We will accept this as truth but we want to speculate...heck half of the discussions in the general hardware thread speculate about upcoming rpoducts....Go Figure!!!

If it isn't ATI r350 and NV35, It was Canterwood and springdale, orPrescott and hammer...At least here we some tangible numbers from the core logic of the athlon64 just from the more souped up model for the server market...

 

[Duvie]

Originally posted by: sandorski
Concerning Desktop performance: When Win64(or whatever it'll be called) is released, Desktop Performance will increase. Some reviews have explained why, such as Memory use methods and such.

Yep I saw some of that at Ace's...Looks like it can be nothing on some to +25% on a host of others....Good stuff...

But maybe according to Nemesis I can't discuss the Athlon64 until Microsoft releases their 64bit OS so we can see how effects theses apps!!! Otherwise I would be ASSuming...