Computex finally coughs up Nehalem benchmarks

Robert Hallock 5 Jun, 2008 at 11:21pm ET Article published in Tech

If you’re a silicon nerd like me, you’ve spent the last few days rolling your eyes at the deluge of sub-notebooks, off-brand heatsinks and blurry RV770/G280 shots. What you were looking for was a glimpse of the elusive Nehalem being toyed with in the wild, free from the manacles of Intel’s PR department.

You knew it would be fast, I knew it would be fast. You hungered for charts, graphs and cold, hard numbers. Pleasantly, Anandtech has recently coughed up the goods and delivered the benchmarks straight to us.

What we have learned so far is that even on a board that did not have working PCIe, nor working multi-channel memory, it:

Is ~40% faster than a Penryn clock for clock
Consumes 10% more power than a Yorkfield
Uses a new socket
Has four physical cores, and four virtual cores due to Hyperthreading (It’s baaaaa-aaaack!)
Is as fast as a 3.2GHz skulltrail system at an artificially-limited speed of 2.66GHz
Uses DDR3
Has an IMC known as QuickPath Interconnect (QPI)

And more!

Ladies and gentlemen, this is the 411 we’ve all been waiting for, and it does not fail to impress. When we stop to realize that the chip was factory-locked, running single channel instead of triple channel memory, and had a broken PCIe implementation on the board, it becomes astonishing.

The Nehalem is truly a marvel, handily slaughtering their own chips as if they were a competitor. Between the Nehalem and the Tegra, we’ll have months of delicious silicon to speculate on before the end of 2008, when we can all own a shiny Nehalem of our very own.

Comments

5 Jun 2008 ~ 11:38pm BuddyJ Provided we can afford the new processors, motherboard, and DDR3 by then.
6 Jun 2008 ~ 7:52am Zuntar Fast, but affordable? Can you say price premium?
6 Jun 2008 ~ 8:06am lemonlime Been waiting for this for a long time

Intel has finally implemented the only remaining features that AMD had on its list of advantages--IMC, hyper transport, three level cache, native quad etc. Those features alone didn't serve AMD well in comparison to Intel's Core2, but combine them with Intel's efficient cores and you've got a recipe for success.

As if I wasn't concerned enough about AMD after Phenom's release..
6 Jun 2008 ~ 8:08am Zuntar Amen ^^^

for Intel

:Pwned: for AMD
6 Jun 2008 ~ 9:13am jokerz4fun It sounds killer!!!
So should I start selling drugs so I can afford it?
6 Jun 2008 ~ 9:29am BuddyJ I hear there are some empty corners on 10 Mile.
6 Jun 2008 ~ 9:49am Linc

Buddy J wrote:

I hear there are some empty corners on 10 Mile.

That was a lie. Keebs' territory runs deep.
6 Jun 2008 ~ 9:50am jokerz4fun Sweet! The suburbs pay better then the city
6 Jun 2008 ~ 9:51am jokerz4fun

Keebler wrote:

That was a lie. Keebs' territory runs deep.

Can we share? I promise I wont steel any clients of yours!
6 Jun 2008 ~ 1:52pm Thrax
6 Jun 2008 ~ 1:59pm jokerz4fun
7 Jun 2008 ~ 3:28pm Qeldroma I'm not an AMD fanboy and my latest buys were Intel. I've also had to focus on my job and haven't been able to ferret out all the details- so bear with me here. But I really question one feature here.

I'm not sure why Intel insists on hyperthreading. Wasn't it one of the Netbust hamstrings that AMD countered by turning out a true dual core and kicked their buts? Wasn't it one of the "features" that was dropped in favor of a smaller, higher performance pipe and more power efficent CPU that made Conroe such a sucess? If they're going to an 8 core anyhow- I'd rather have the cores than the Hyperthreads. It looks like dropping HT would save some watts as well. This would make sense since they have to use up a chunk of silicon to support it.

They couldn't disable the HT on this sample and I'm not familiar enough with many of the test apps to say how much that mattered in some of the results. But if that's the case- then maybe I'm less impressed than I should be.

Hyperthreading just seems like some Intel inside old-boy agenda. It may also turn out to be more of a handicap than a worthwhile feature.
7 Jun 2008 ~ 3:37pm Thrax Hyperthreading was a compensation to keep Netburst's retardedly long pipeline filled. It was a 30 or 40 stage pipe by all accounts, which is probably the worst design decision in x86/x87 history. It's also an amazing technology... It allows one execution engine to process two threads simultaneously. Even without another set of cores, this can improve the throughput of a chip by up to 30%, all while keeping the footprint of the chip down and keeping the dissipation/consumption way down.

Hyperthreading was dropped because Netburst sucked. HT was never a hamstring, it was perhaps one of the only brilliant ideas to have come from Netburst, and certainly one of the scant few design elements that kept the P4 from sinking like a brick.

Now we come to Nehalem which, I theorize, has a slightly longer pipe than Conroe/Kentsfield or Wolfdale/Yorkfield. As the benchmarks indicate, the single 2.66GHz Nehalem is faster than an SMP quad skulltrail system at 3.2GHz. Obviously the 4 physical/4 virtual is working to take the speed crown in such a matchup, suggesting that cores ain't everything. All of that's with single channel memory hamstringing performance. As a quad, ignoring HT, it's a quantum leap in horsepower (Again). Adding HT in is icing on the cake.
7 Jun 2008 ~ 4:17pm Qeldroma Thanks, T. That makes a bit more sense. Perhaps it was my experience with it folding that has left me less than impressed and questioning its value. We should know more as time goes on.

Yeah, I agree- the reduced L2 cache latency and an apparently very effective L3 looks to have really improved CPU/memory interaction- especially with a single channel. Impressive as that is- it bodes well to hopefully scaling well with added channels. This is a huge hurdle they've overcome with such a breakthrough. AMDs HT 3.0 looks to be left wanting in comparison.

Which leaves me with a possible sad realization: We may soon only have one consumer CPU provider. This scenario never bodes well.