Intel's Conroe Previewed and Benchmarked

DrFishstik writes "Anandtech has a few preliminary benchmarks on Intel's new Conroe architecture. From the article: 'As far as we could tell, there was nothing fishy going on with the benchmarks or the install. Both systems [AMD 2.8Ghz OC and Conroe] were clean and used the latest versions of all of the drivers.'"

Comments

Wait and see

[xming]

As pointed out by Ars http://arstechnica.com/news.ars/post/20060307-6334 .html I think we should wait and see for the more objective benchmarks. Anyway 2006 will be a good year for CPUs

Re:Wait and see

[hattig]

The K8 core is as similar to the K7 core as the conroe core is to a PIII core.

I.e., at first glance there are similarities which can lead to the obvious thought that the K8 core is just a K7 core with memory controller, but actually they're completely revamped, overhauled, enhanced and redone.

I agree that it is time for AMD to get a "K9" out of the door as the K8 as it is won't compete against Intel's offerings unless AMD somehow get 3.6GHz out of 65nm at launch (which is extremely unlikely). Of course, K8L will probably put AMD back into the lead in terms of floating point anyway, but integer is going to be very weak.

Unless AMD is sandbagging - but that's a faint hope for even the most ardent AMD fanboy. I think they miscalculated Intel this time around.

Which of AMD or Intel has the most fangirls?

Re:Wait and see

[mooingyak]

ars technica != Anandtech

Good summary of the Anandtech article though.

Re:Wait and see

[imsabbel]

Well, thats certainly not correct.
The internal differences between p6 and k7 are enormous.
From microops sheduling (k7 using packed microops, in some kind of on the fly VLIW ) to the execution units (fully piplelined and superscalar FPU, for example, compared the non-fully piplelined scalar one), virtually the only thing thats the same is the fact it eats x86 opcode.

But the fact is that the changes between p6 and the new p-m derivates are VERY much larger than the change from 99s k7 to the latest k8.
Just look at a current die-photo of a k8... back in 99, the core transistor count was at the edge of what was possible economically, with l2 cache externally implemented. Nowadays, the nearly unchanged core is just a small lump on the side of the large and not very dense l2 cache-array...

Re:The Conclusion

[Kjella]

by the time this new intel is out, AMD should already be well and truly released. probably also embedding themselves more in Dell's good books and taking more than 80% of the market. Intel are fighting the loosing battle.

1) AMD has something like 20% of the processor market, including OEMs. They couldn't deliver 80% of the market in many years even if the market wanted it.
2) AMD has no major process/architecture shifts between now and Conroe's release.
3) The AMD chip was already overclocked (but then again, they may have gotten a golden sample from Intel).
4) It's losing, not loosing.

Re:AMD Processor Model Unknown

[iainl]

The AMD was overclocked to the timings of the one that won't officially be released until June - unsurprisingly, AMD won't let them have a pre-production chip to demonstrate how their one is even faster.

Re:Shock news.

[timeOday]

Also, the forthcoming AMD processors are a new core architecture and will support faster RAM with an onboard memory controller.

The review did address that, as best they could:

While we're still comparing to Socket-939 and only using RD480, it does seem very unlikely that AMD would be able to make up this much of a deficit with Socket-AM2 and RD580. Especially looking at titles like F.E.A.R. where Conroe's performance advantage averages over 40%, it looks like Intel's confidence has been well placed.

As for your assertion that MHz don't mean anything, that's just wrong. Within a single architecture, speed is nearly proportional to MHz. For a 2.66 GHz Intel to crush a 2.8 GHz AMD so convincingly, does not mean good things for AMD if the Intel can easily reach 3 GHz. It means AMD would have to be at about 3.8 GHz to keep pace: 2.8*(3/2.66)*1.2 = 3.7895 assuming these benchmarks show a 20% lead for Intel.

The real hope for AMD here is that these results won't hold to other benchmarks in general. Apparently this set of benchmarks was handpicked by Intel, so that's almost certainly the case to some degree.

Your questions answered

[Namarrgon]

the number of concurrent threads

If you're referring to "Hyperthreading", Conroe has none that I'm aware of. One thread at a time, in hardware (whatever you like in software of course).

the power consumption and with that the heat output

Conroe is supposed to have a Thermal Design Power of only 65W. Compare this to the current 3.6GHz P4's TDP of 115W. AMD rate the Athlon FX60's TDP at 110W; however AMD quote the maximum possible thermal dissipation while Intel quotes "typical", usually 75% of maximum (which would make the FX60 about 82W by Intel's reckoning) .

of course the expense of both the processor and the ram it needs

The 2.4GHz and 2.6GHz Conroes are expected to sell for US$316 and US$530 respectively, in 1000-unit quantities (the FX60 was released at US$1031). RAM is harder; reportedly Conroe chipsets will use DDR2, but possibly packaged as new FB-DIMMs. I don't have pricing for those yet, but they'll probably cost more. Consumer motherboards may just use standard DDR2 DIMMs.

Re:Shock news.

[fitten]

The things that AMD has said that they have are F, G, and H revisions of the K8 core (the core that the Athlon64, Turion, Sempron64s, and Opterons are based on) which, other than DDR2 support, not much more information is available. There is another revision called the K8L which will supposedly have 2x the FPU units for about a 50% gain in FPU performance. These will most likely be HPC blade Opterons or some such.

DDR2-800 support, which is the known upgrade, basically adds bandwidth to a chip that isn't bandwidth starved as it is. Current speculation is that the new DDR2-800 Athlon64s will show up to a 10% performance increase on extreme bandwidth benchmarks (synthetics and HPC crunchers, for example).

THe simple fact remains that intel needed to do these tests at all, side by side. That's an admission on their part that AMD is beating them and beating them hard.

Intel has publicly stated (admitted) this already. This demo is to show that the chips they have planned for Q3'06 release (speculation is that they will be delivering machines based on it in July which is the very beginning of Q3, which is only 4 months away) perform well.

By the way, if speculation is that machines will be selling in July, this would imply that the chips are in manufacturing even as we speak. This means that Apple is most likely to announce availability of the new Intel based Power Macs around this time, as well and the various benchmark sites to have their hands on 'pre-production' machines in two to three months tops. We'll be able to see the real story then.

The only announced things from AMD even remotely in this time frame (specifically July and Q3'06) are the AM2 socket for DDR2-800 and a speed bump of the FX-62 to 2.8GHz (which is the equivalent of the overclocked part in the demo). Given that DDR2-800 is expected to be a 10% speed bump at most in most cases and that Conroe will be available at 3GHz (if not higher as rumored - 3.33GHz), I predict (a rather easy prediction to make) that AMD will be playing catch-up for once in the past few years.

Re:Shock news.

[fitten]

Oh... another thing that I forgot was that somewhere in the F, G, and H revisions (and probably the L), HT was supposed to be bumped up to 333MHz (1.333GHz effective) from the current 200MHz (1GHz effective). Given that tests have already shown that 800MHz effective HT performance is statistically equal to 1000MHz effective HT performance, boosting HT speed will probably give a small (1% to 3%) performance increase at best. In actuality, the HT speed increase is required for DDR2-800 to run at its best so the performance gain for it is probably inclusive to any gains shown by DDR2-800 adoption.