Athlon64 Preview: nForce3 at 2.0GHz @ anandtech

[mrgoblin]

Originally posted by: AgaBooga

Originally posted by: shady06

Originally posted by: mrgoblin
Shame its gonna cost upwards of 1k for an fx cpu/mobo combo and some reg ddrs.

i thought registered was not required for athlon64???

Same here. I don't think it is, but I think its an option like the 875P boards. Either that or its required for the Opterons which are for workstation and server markets.

The fx model does require those because its a rebadged overclocked opteron. The 939 pin model wont need reg dimms. The 754 pin single channel is fine with normal dimms.

 

[DaveSimmons]

Originally posted by: shady06

Originally posted by: DaveSimmons

So, the A64 at 2.0 seems to be competitive with a 3.2 P4, but not dominant in anything except Gunmetal. It's also seriously weaker in media encoding, at least until developers figures out how to optimize for it.

10% lead in quake III <---against a 3.0 GHz !
12.5% lead in ut2k3 flyby
24% lead in ut2k3 botmatch

those numbers seems pretty dominate to me

The article is using a 3.0 GHz P4 -- that's my main criticism, phrases like "best P4" and "fastest" led people like you to think a 3.2 GHz was being used in the review, but it's just a 3.0.

 

[glugglug]

Originally posted by: DaveSimmons
those numbers seems pretty dominate to me

The article is using a 3.0 GHz P4 -- that's my main criticism, phrases like "best P4" and "fastest" led people like you to think a 3.2 GHz was being used in the review, but it's just a 3.0.[/quote]

Actually it is being compared against a dual 3.06 GHz P4 Xeon. There is no 3.2GHz Xeon chip yet.

 

[Bonesdad]

lets all get along...dont make them lock this thread by creating an AMD vs Intel out of it. As far as I can see, competition is good, keeps prices down and gives us options. I for one don't want to see AMD fail. If either AMD or intel had the only processor, none of us would be happy (and we'd all be poor).

 

[DaveSimmons]

Originally posted by: glugglug

Originally posted by: DaveSimmons
The article is using a 3.0 GHz P4 -- that's my main criticism, phrases like "best P4" and "fastest" led people like you to think a 3.2 GHz was being used in the review, but it's just a 3.0.

Actually it is being compared against a dual 3.06 GHz P4 Xeon. There is no 3.2GHz Xeon chip yet.

Yes, 3.06 xeons and a 3.0 GHz P4. My point is simply that the phrasing in the article will lead some people to assume that the single P4 is a 3.2 when it is actually a 3.0. Also dual xeons don't really shine in gaming benchmarks, since games really only use one of the CPUs.

 

[Cat]

Are there 800Mhz FSB Xeons? Edit: no.

 

[Alkaline5]

Originally posted by: DaveSimmons
It will also be interesting to see how much single-channel will hurt the reasonably priced version of the A64, if it drops performance by 10% then the 2.0 A64 is down to matching the 3.0 P4 used in the review.

I'm surprised that no one has mentioned the ocworkbench A64 2900+ Review yet. Unfortunately they only compare it to the Barton 3200+, but even running at only 1.8 GHz the single channel A64 shows some good performance gains compared to the 2.2 GHz Barton..

 

[Eug]

:Q

VERY impressive, but I should point out this is on a 222 MHz bus. We'll see when the true 2.0 appears. Plus no 3.2 GHz P4 chips are included in the mix.

Prescott 3.4 should prove interesting too...

And the G5 2.0 with IBM's XL compilers will be interesting too (although not as fast as Prescott or the 2.0 Opteron)...

A very interesting year indeed.

 

[pelikan]

I am very excited to see the gaming performance. I was tempted to go with a 2.4C O/C set up for the 10% or so performance gain over what I have now, but after seeing this article I'll wait!

 

[Dug]

Why does it take so long for articles to come up? Each page takes minutes to load, sometimes having to refresh just to get pictures. It's been like this for weeks.

 

[Regs]

Originally posted by: Eug
:Q

VERY impressive, but I should point out this is on a 222 MHz bus . We'll see when the true 2.0 appears. Plus no 3.2 GHz P4 chips are included in the mix.
Prescott 3.4 should prove interesting too...
And the G5 2.0 with IBM's XL compilers will be interesting too (although not as fast as Prescott or the 2.0 Opteron)...
A very interesting year indeed.

That's one point I don't get . I thought the Athlon 64 didn't have a bus. Memory controller is on the cpu itself while using the nforce3's hypertransport.

 

[alexruiz]

Originally posted by: Regs

Originally posted by: Eug
:Q

VERY impressive, but I should point out this is on a 222 MHz bus . We'll see when the true 2.0 appears. Plus no 3.2 GHz P4 chips are included in the mix.
Prescott 3.4 should prove interesting too...
And the G5 2.0 with IBM's XL compilers will be interesting too (although not as fast as Prescott or the 2.0 Opteron)...
A very interesting year indeed.

That's one point I don't get . I thought the Athlon 64 didn't have a bus. Memory controller is on the cpu itself while using the nforce3's hypertransport.

You are correct! The change of this "multiplier" was only to overclock this baby.....

 

[dullard]

Originally posted by: Regs

Originally posted by: Eug
:Q

VERY impressive, but I should point out this is on a 222 MHz bus . We'll see when the true 2.0 appears. Plus no 3.2 GHz P4 chips are included in the mix.
Prescott 3.4 should prove interesting too...
And the G5 2.0 with IBM's XL compilers will be interesting too (although not as fast as Prescott or the 2.0 Opteron)...
A very interesting year indeed.

That's one point I don't get . I thought the Athlon 64 didn't have a bus. Memory controller is on the cpu itself while using the nforce3's hypertransport.

It is since Eug had a typo, it is on a 222 MHz hypertransport. But really, to me, there isn't much difference. You bump up the hypertransport and the CPU with its fixed multiplier goes faster (as does anything else using hypertransport for its timing).

 

[Eug]

It is since Eug had a typo, it is on a 222 MHz hypertransport. But really, to me, there isn't much difference. You bump up the hypertransport and the CPU with its fixed multiplier goes faster (as does anything else using hypertransport for its timing).

Yeah, not quite the same thing, but close enough. And yes, HyperTransport is a bus, even if it isn't a bus in the traditional sense.

The other concern with this demo is the fact it's testing a dual channel DDR system (as others have mentioned), whereas the consumer chips are single channel DDR. On some benches it should make quite a significant difference. I know both single and dual channel versions will be released, but one has to wonder how much the FX is gonna cost. I'm betting the FX is going to have a significant premium even some time after initial launch.

Count me in the camp that reads this (admittedly impressive) bench with a healthy helping of salt.

 

[First]

Originally posted by: Pariah
Overclocking a CPU to attain a clock speed that already exists (Opteron 246) in an attempt to predict the performance of a CPU that is not officially released and whose specs aren't even known yet? I thought Anandtech was supposed to be above such tabloid material.

Is it somehow unbelievable to think that we know the A64 specs in advance and so are able to publish a preview that will show almost exactly how the high-end A64 will perform at 2.0GHz?

 

[wetcat007]

For people who are complaining about the possible MSRP for the chip, it will most likly be available for a bit less, and it will be kicking Intel's ass after all, and take a look at the lowest priced top end intel chip $613 Pentium 4 3.2GHz 800, it beats that and has 64 bit on it, with less heat production and a integrated northbridge. AMD is for teh first time in a while, really looking like a serious problem to Intel, this bad boy is ready for release, Intel's new chip has problems and more setbacks, and no 64bit, AMD will most likly be winning this round by a lot.

 

[Regs]

Damn 450 is just for the 3200 ! I didn't even see the 850 price tag of the FX! 850 dollars? Forget that sh*t.

 

[digitalsm]

Originally posted by: Regs
Damn 450 is just for the 3200 ! I didn't even see the 850 price tag of the FX! 850 dollars? Forget that sh*t.

I dont see how a 2Ghz A64 FX will cost $850. I mean the 2Ghz Opteron 246 cost $800-900 and should have a price premium on a SP chip. I seriously doubt the 2Ghz FX will be $850, more likely it will be $650.

 

[charlie21]

It's exciting to finally see a new architecture. When was the last time you remember seeing a CPU benchmark that really made you sit back and go, wow, now that's some improvement!

 

[mechBgon]

Originally posted by: digitalsm

Originally posted by: Regs
Damn 450 is just for the 3200 ! I didn't even see the 850 price tag of the FX! 850 dollars? Forget that sh*t.

I dont see how a 2Ghz A64 FX will cost $850. I mean the 2Ghz Opteron 246 cost $800-900 and should have a price premium on a SP chip. I seriously doubt the 2Ghz FX will be $850, more likely it will be $650.

That makes sense. Those who get their undergarments in a bundle at a mention of The Inquirer need not visit the following link: Athlon 64 pricing predictions

 

[vash]

Hey, I have a question for you all. Anyone remember when the first Athlon 500s appeared? Fast and furious these things were AND they were more expensive when initially introduced. Funny how that is happening again.

This is a new chip, that needs new boards and they are going to be a wee bit expensive. Over time, the prices will drop (maybe later than some people want) and they will become competitive. Remember, there is a lot of AthlonXP left in the market. How would OEMs feel if A64s came and flooded the market while the AthlonXPs were sitting on shelves? These OEMs need to clear their shelves of existing AthlonXPs before A64s fill up space.

vash

 

[draggoon01]

Originally posted by: charlie21
It's exciting to finally see a new architecture. When was the last time you remember seeing a CPU benchmark that really made you sit back and go, wow, now that's some improvement!

toms h-ware, but when they do it, it's laughed at...

 

[Pariah]

Originally posted by: Evan Lieb

Originally posted by: Pariah
Overclocking a CPU to attain a clock speed that already exists (Opteron 246) in an attempt to predict the performance of a CPU that is not officially released and whose specs aren't even known yet? I thought Anandtech was supposed to be above such tabloid material.

Is it somehow unbelievable to think that we know the A64 specs in advance and so are able to publish a preview that will show almost exactly how the high-end A64 will perform at 2.0GHz?

It's pretty close to unbelievable after reading the intro and conclusion which both quote the Inquirer as a source and make numerous vague statements that don't indicate any level of certainty.

"The impact of Dual-Channel memory is a little harder to estimate in our tests. Athlon64 has been widely reported to be single-channel, where Opteron is Dual-Channel. Again, we expect our results reported here to be in the ballpark."

That's a statement with some real conviction behind it. What does in the ballpark mean? 2%, 5%, 20%? Calling the article an Athlon64 preview is a blatant lie. It's not. It's a review of an overclocked Opteron which from all reports contains architectural differences whose impact cannot be estimated except by guessing at this point. I would guess an Athlon64 will likely perform within 10% or so of an equally clocked Opteron, but that's all it is, a guess. Slapping Athlon64 on the name the article doesn't give any of us a better estimation of A64 performance than what he have with the already available Opteron 2GHz benchmarks. And in some respects, due to the slight OC, these benchmarks are likely even more inaccurate than what was already available.

 

[mechBgon]

Wesley laid out his methods and his reasoning, he reported his results, and he made some comments about the results. He feels they give an indication of Athlon 64 performance, and he appropriately qualified his statements by pointing out that there are variables yet to be determined.

Since Athlon 64 FX is Opteron in essence, it's hard to argue that the results aren't relevant at all. Now, what the effect will be of trading dual-channel R-ECC for single-channel low-latency DDR400 is, yes... that remains to be seen. So does the effect of 64-bit optomization.

Personally I appreciated the article. I'm not going to cover my eyes and plug my ears just because it wasn't a "real" Athlon 64 at stock clock. To each his own...

 

[Alkaline5]

Originally posted by: mechBgon
...what the effect will be of trading dual-channel R-ECC for single-channel low-latency DDR400 ... remains to be seen

Is everyone just ignoring me or do you not trust ocworkbench? Cause Socket 754 single channel performance is exactly what this shows in great detail.