Athlon64 Preview at xbitlabs

[imgod2u]

They got their hands on a 1.6GHz model. They've put up the <A HREF="http://www.xbitlabs.com/articles/cpu/display/athlon64.html" target="_new">preview</A> (someone should really cache this as the server is taking a huge hit). Seems to do pretty well. One wonders what frequency it'll be released at though as the 1.6 GHz "2800+" model doesn't look to be fairing too well against the 2.8C P4 (FSB800, HT enabled) and often looses to the T-Bred "2800+" (understandable since the T-Bred is running at 2.25 GHz). One can only wonder what an actual 2.0 or 2.2 GHz model could perform like upon release. Hopefully it'll be more deserving of the PR model than this is since it is lagging pretty far behind the 2.8C.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

 

[Mephistopheles]

Just checked that preview. (took a reeeally long time, BTW)

Quite interesting numbers. Looks good, but not so good after all. Of course, there are no 64-bit apps in there, but even by the end of the year, the 32-bit performance will probably still be the most important one. Also, this is a sample, and A64 won´t compete with "2.xC"s or "3.xC"s but with up to 4Ghz of Prescott... Hope they can do that.

We´ll get a better impression of x86-64 technology´s potential after Opteron is released, I guess, but I´m slightly disappointed at those numbers. Hopefully they´ll get those PR ratings right!

The IPC is still excellent, by the way. I wonder how big a boost 64-bit apps will really get in pratice...

 

[mac404]

Just remember this is still with beta and engineering sample stuff, not to mention at a lower speed than what it should ship at in September (I've heard about 2ghz, maybe even 2.2 or 2.4). They still have hope.

You can help. Just send money to the bad motherboard relief fund, for those suckered into a KT166 and SDRAM with their new Athlon...

 

[Mephistopheles]

You´re right, of course...

But I still hope they don´t label a 2.0Ghz A64 a 4000+... it won´t perform that good! That 2800+ didn´t quite live up to its rating...

But I guess we should still keep an open mind. Their PR rating is bound to cause difficulties when architectural changes as big as those change the performance of processors across various tasks in different ways...

 

[reever2]

Blegh, i would much rather see a preview with the nvidia or amd chipset

 

[mac404]

I should have added that it was with the VIA chipset, too. Way to put it in about the worst possible light, X-bit. It's latency is very good then for a VIA chipset.

You can help. Just send money to the bad motherboard relief fund, for those suckered into a KT166 and SDRAM with their new Athlon...<P ID="edit"><FONT SIZE=-1><EM>Edited by mac404 on 04/19/03 00:42 AM.</EM></FONT></P>

 

[reever2]

Actually if you think about it, the A64 is already looking successful. After all many people who will buy the A64 will most likely be gaming, which it seems this processor owns at. I guess the inquirers benchmarks were right, it did hit 16k easily even at such a low clockspeed.

 

[Schmide]

It took me forever to get through that one. That is one busy site.

Those SSE numbers from SiSoft Sandra show that without a much faster memory system, no CPU is going to keep up with the 875. If you normalize the Intel CPU's integer numbers the 2.53ghz = 5,118 per ghz, while the 2.8 = 7,369 per ghz. The Athlon 64 1.6ghz = 5,725 per ghz. Since the 2.53 and the Athlon 64 have a similar front side bus throughput, and come out with similar marks per ghz while the Intel systems have similar ghz yet differ in marks and front side bus. We must assume the SSE tests are very dependant on front side bus throughput. We will only know this when a dual memory controller Opteron is tested. Since it operates with a similar front side bus throughput as the 875.

Dichromatic for your viewing plesure...

 

[Twitch]

Looks like it really will be a kick-ass processor for games though. I think especially once it goes to 1.8 or 2.0 gig...


<-----Insert witty sig line here.

 

[imgod2u]

Yes, SIMD can be very taxing on the memory subsystem due to the sheer amount of data neccessary. But keep in mind to accomplish the same throughput using scalar (such as x87) instructions would take even more memory bandwidth. It's not surprising that we see the SSE/SSE2 component not provide a significant boost compared to the x87 component. The K7/K8's x87 component performs very well. We are, however, seeing that with SSE2, the clock-normalized performance of the P4 becomes much closer to that of the K8/K7.
Another thing to mention is that since the memory controller is on the CPU itself, the chipset xbitlabs chose is relatively irrelevent. Memory performance relies solely on how optimized the integrated memory controller is.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.

 

[Schmide]

After looking around, I'm starting to think it's the p4 2.8ghz's hyper-threading that makes its score so high in SSE.

Dichromatic for your viewing plesure...

 

[baldurga]

Sorry, but I try to read the article but seems unavailable. Is it a general issue? Maybe they take out because NDA?

Not sure AMD want that info goes public ...


Still looking for a <b>good online retailer</b> in Spain :frown:

 

[mr_gobbledegook]

Update: Website seems to be working fine 13:45(GMT) although a tad slow. (I am however using a 56K connection)

<font color=purple>Ladies and Gentlemen, its...Hammer Time !</font color=purple>

 

[Mephistopheles]

Hm... According to the benchmarks, the 2800+ at 1.6Ghz only handily defeats the 2.8C P4 in UT2K3. The AXP could already do that... and the 2800+ A64 doesn´t win in any other gaming benchmark. So I don´t think there´s nothing new here, right?... You can´t possibly represent gaming performance only with UT2K3 - not only because that would be too limited in scope, but because that would paint too good a picture for the Athlons. Like saying a 2.2Ghz P4 on DDR266 beats an Athlon 2800+ on nForce2 with Dual-DDR333 in mp3 audio-encoding - it <A HREF="http://www.tomshardware.com/cpu/20030217/cpu_charts-25.html" target="_new">does!</A> That´s a bad representation of the 2.2Ghz, as an example. Not because it´s not true, but because the 2.2Ghz <i>doesn´t</i> beat the 2800+ in any other scenario at all (obviously, right?)

Once it scales to 2Ghz, though, it might be an excellent performer (considering perfect scaling). The Northwood won´t keep up with that because of design and thermal issues, but Prescott will be very capable of scaling beyond 4Ghz... That´s a 43% scale from current specs (Northwood/2.8Ghz to Prescott/4.0Ghz+), with an architectural improvement about which we have little info (more cache, improved branch prediction (supposedly), improved HT, and so on). A64 is expected to be capable of 2.0-2.25Ghz (at least), which is also about 41% scaling over that preview we saw... And while that was probably an up-to-date design, it might still be subject to change. It looks as if AMD will be able to compete well with Intel, I´d say, but it´s still hard to say anything else at this point... What do you guys think?

Hm... I was still a bit disappointed about the SSE2-implementation on A64. I thought that would give it more of a boost... So much for it, then. But it´ll still be interesting to see a 64-bit capable OS running an A64. The x86-64 implementation was the more exotic of the two anyway.

 

[bikeman]

Yup. I think the Athlon64 is a nice CPU. If it weren't for the irritating PR-rating, I think it could grow to be a nice success. Why the PR-rating bothers me? First of all, this '2800+' is generally not comparable to the 'old' 2800+, and two its performance in comparison to the AXP varies to much. I am getting more and more convinced AMD should really drop its PR-rating. Though that is a hard subject on this forum, if I recall correctly ...

Et voilà, my two cents ...

Greetz,
Bikeman

<i>Then again, that's just my opinion</i>

 

[eden]

My personal view on this:

-Now I've had it, I feel like I wanna punch AMD so hard for that 2800+ rating.

Now, I have to say, the per-clock jump, if some noticed, in the gaming department, rose significantly, from 20-35%! Amazing jump, very powerful. I wouldn't attribute more than 5% to the cache, because we've already seen the Barton's doubling of cache performance.
The integrated memory controller has proven itself worthy however. It proves that no matter how much you improve on a normal Northbridge, you will never reach the level of an integrated memory controller. I'd like to see Slvr's take on this.
The boost was tremendous per clock.
Sure there were those applications who rely on simple pure clock speed, those simply prove how monkey programmers can't even properly make some parallelism out of x86.

Additionally, who in the world is stopping AMD from bumping up the clock? Xbit reports that the extra two integer stages will help in clocking, so why is it so hard for AMD to bump the clock?

I was indeed surprised by the SSE2 performance. It is confusing to be honest. One must wonder if you used Dual Channel chipsets in the future, like the nForce, for the K8, will you get better results?
Also, no one commented on the ridiculously powerful bandwidth output of the Athlon 64. At 96%, you guarantee the near maximum possible. A P4 with Canterwood can't even reach 80%!

All I can say is that the CPU, if clocked at 2.2GHZ, WILL become the undisputed gaming machine. If the current system yeilded over 20% more performance per clock in games, imagine a higher clocked one, with the future workarounds AMD will do before releasing the Athlon 64.
All I hope is they don't do a suicide mission with the PR rating. It is ridiculous now, seriously, and I wish I could just punch the guy who is making them. It is truly getting me personally.

--
This post is brought to you by Eden, on a Via Eden, in the garden of Eden.

 

[mr_gobbledegook]

Some important facts from the artical:

- Athlon 64 supports DDR400
- Memory latency = 96ns !! (Athlon XP = 165ns, P4 = 260ns)
- ALU performance in Athlon is 8% higher than that of the Athlon XP ALU
- Mp3 encoding with lame Athlon 64 suffers a complete failure becuase of it low clock frequency.
- Unreal Tournament benchmark: Athlon 64 (210fps),Athlon XP (197fps)

I reckon the Athlon 64 shows a lot of potential. Why are people getting upset with the PR rating ??!! THIS IS NOT THE FINAL RELEASE ! I'm sure AMD will adjust the PR rating 100 times before the final release anyway. The 1.6Ghz Athlon 64 in the review will most likely be marketed as a low end budget processor (probably replacing the Duron). AMD have hinted that Athlon 64 will be released around 2Ghz. This boost in clock frequency and couple of months of tweaking before the final release could make it a strong contender. But then again Prescott could blow it out the water ! I think I'll wait till Sep to get the <b>real</b> picture.

<font color=purple>Ladies and Gentlemen, its...Hammer Time !</font color=purple>

 

[eden]

Considering most of Prescott's enhancements have been already used on the NW, now remains the 1MB cache & the internal enhancements.
So overall, I don't expect Prescott to be insanely more powerful in IPC, however its clock speed will likely be killer.

AMD needs to realize, that clock speed is indeed, you can only go so far without upping the clock speed. Intel may have been right in the high clockrate.

--
This post is brought to you by Eden, on a Via Eden, in the garden of Eden.

 

[Mephistopheles]

Yes, this will become quite a machine if the clock is upped to 2.0 or 2.25 Ghz. But once you think that prescott will have better (not devastating, but better, no doubt) IPC and will be very scalable (90nm) up to 4Ghz or more, then this dispute will become interesting...

Considering most of Prescott's enhancements have been already used on the NW

This I don´t quite understand. When you say "prescott enhancements", I naturally assume things that are NOT on Northwood... like the things you mentioned. So what is it that has already been used? The FSB, I guess?

 

[mr_gobbledegook]

We do not know much about Prescott at this time anyway. Only the following facts are definate about Prescott right now:

- 800Mhz bus
- 1MB L2 cache
- SSE3 (13 new instructions)
- 0.09 micron process

We do not know the precise details of the core modifications that Intel will make to the CPU. With the extra space they save using the new .09 process Intel will have lot more extra die space to play with which they will put to good use.

I have a lot of faith when <A HREF="http://www.anandtech.com/cpu/showdoc.html?i=1810&p=23" target="_new">Anandtech</A> say:

<i>"The true benefits of both the 800MHz FSB and Hyper Threading will be realized with Prescott later this year as the processor will push higher frequencies and contain Hyper Threading specific architectural improvements that should improve performance even more."</i>


<font color=purple>Ladies and Gentlemen, its...Hammer Time !</font color=purple>

 

[eden]

Yup, these are the enhancements majorly. SSE3 however (dunno if that is what it should be called, extra extensions for HT) is not gonna be a big deal like SSE2.
I do recall also some cache architecture enhancements or size, for the Trace Cache, and perhaps some other things.

--
This post is brought to you by Eden, on a Via Eden, in the garden of Eden.

 

[martinrichards23]

Studying the results it looks as though the A64 1.6ghz would need to be 15-20% faster in clock to keep up with the p4 2.8ghz, AMD said the A64 would be 15% faster in 64 bit mode so this pr rating is probably the final one it is just that we need a 64bit o.s. to see the A64 is all its glory. Also the 1.6ghz will certainly be the slowest model released (if released at all), the fastest being 2/2.2 ghz (600mhz faster than xbit model), whereas intel will have 3.2ghz p4 at best(400mhz faster than xbit model), thus the performance gap will close and AMD will almost certainly gain performance crown (as long as roadmaps dont change), especially if msoft release 64bit windows.

 

[fireflayer]

I was just curious... I read the HardOCP article about Cantwood and on the gaming benchmarks with the 9700 Pro the video card was the bottleneck... can anyone explain to me why that isn't the case in the X-Bit article? Both were both @ 640x800.

 

[reever2]

Lower resolutions put less stress on the video card and more stress on the cpu, thats why he did it at 640x480, he didnt want the video card having anything to do with performance

 

[fireflayer]

I know that... but HardOCP said they did that as well.