Nehalem Performance Preview @ AnandTech

All about them.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
krille
Posts: 357
Joined: Thu Mar 23, 2006 4:56 am
Location: Sweden

Nehalem Performance Preview @ AnandTech

Post by krille » Thu Jun 05, 2008 1:01 pm

Amazing performance (as expected).
Today, Intel is without competition at the high end but that hasn't stopped it from topping its performance by 20 - 50% with its brand new architecture. Intel's Nehalem processor is due out at the end of this year but we previewed its performance today and were very impressed; if you thought Penryn was fast, just wait until you see Nehalem.
And for us SPCR'ers:
Built on the same 45nm process as Penryn, we expect Nehalem to have higher power consumption than Penryn but given Intel's target of a 1% increase in performance for no more than a 1% increase in power consumption per microarchitectural change - the results should be reasonable
Still getting better it seems. Although not very impressive power figures, I guess undervolting and/or underclocking should make up for it and more!

Get it while it's hot, start loading the URL today and you might have it loaded within a day or less. Seriously, they should look into getting more bandwidth.
http://www.anandtech.com/cpuchipsets/in ... spx?i=3326

fri2219
Posts: 222
Joined: Mon Feb 05, 2007 4:14 pm
Location: Forkbomb, New South Wales

Re: Nehalem Performance Preview @ AnandTech

Post by fri2219 » Thu Jun 05, 2008 1:07 pm

krille wrote: Still getting better it seems. Although not very impressive power figures
According to one article I read, it has 135W TDP- I hope that's an error.

jaganath
Posts: 5085
Joined: Tue Sep 20, 2005 6:55 am
Location: UK

Post by jaganath » Thu Jun 05, 2008 1:16 pm

Intel's target of a 1% increase in performance for no more than a 1% increase in power consumption
so in other words performance per watt is completely unchanged. :(

krille
Posts: 357
Joined: Thu Mar 23, 2006 4:56 am
Location: Sweden

Post by krille » Thu Jun 05, 2008 1:59 pm

jaganath wrote:
Intel's target of a 1% increase in performance for no more than a 1% increase in power consumption
so in other words performance per watt is completely unchanged. :(
That would be the worst case scenario, according to the preview.

Airshark
Posts: 44
Joined: Wed Aug 23, 2006 7:38 am

Post by Airshark » Thu Jun 05, 2008 2:19 pm

That 1%-for-1% is an obvious typo. It says later in the article that the performance is 20-50% better for only about 10% more power than Penryn.

tehfire
Posts: 530
Joined: Mon Jan 01, 2007 9:57 am
Location: US

Post by tehfire » Thu Jun 05, 2008 2:59 pm

What Anand's referring to is back in the development day Intel always had the motto "if feature x causes a 2% performance increase for no more than a 1% power increase, then we'll put the feature in". They're not saying that it's always the case, but they will (try to) never implement a feature that disproportunately causes an increase in power consumption. As Airshark said, Anandtech saw only a 10% increase in power consumption.

Keep in mind that since this is system-wide power consumption, the actual chip may take less than 10% more energy then Penryn. After all, this new system is using triple-channel DDR3.

jaganath
Posts: 5085
Joined: Tue Sep 20, 2005 6:55 am
Location: UK

Post by jaganath » Thu Jun 05, 2008 3:03 pm

also, you would expect to see a slight power increase anyway, as Nehalem has integrated mem-controller.

smilingcrow
*Lifetime Patron*
Posts: 1809
Joined: Sat Apr 24, 2004 1:45 am
Location: At Home

Re: Nehalem Performance Preview @ AnandTech

Post by smilingcrow » Thu Jun 05, 2008 5:23 pm

The performance per watt under load is impressive apart from the fact that they don’t state which application they were running when they measured the data.
The idle load is increased by 10W which is very poor and I’m sure the triple channel memory doesn’t help.
But it’s way too early to conclude too much although the raw performance does look very promising.
fri2219 wrote:According to one article I read, it has 135W TDP- I hope that's an error.
Intel’s TDP figures these days are fairly meaningless in determining actual power consumption apart from those used in mobile CPUs so I wouldn’t be concerned about the TDP. When you look at the TDPs for the 45nm chips and then read the real-world data you’ll see that there’s nothing to be alarmed about.

krille
Posts: 357
Joined: Thu Mar 23, 2006 4:56 am
Location: Sweden

Post by krille » Fri Jun 06, 2008 2:57 am

Airshark wrote:That 1%-for-1% is an obvious typo. It says later in the article that the performance is 20-50% better for only about 10% more power than Penryn.
No, you obviously just skimmed the article. It says "1% increase in performance for no more than a 1% increase in power". It means for any design change to be implemented, that change must either (a) be as power efficient as earlier or (b) more power efficient. Add all the (a)s and (b)s together and you end up with a (b) processor. jaganath only seems to have made a very cynical comment, because while all (a)s is possible, it's not likely.

For servers, SMT could theoretically improve power efficiency a lot. As for single threaded performance increase, the only such test that was made showed a clock-per-clock performance gain of 26% for the aforementioned system power increase of 10%.

The motherboards used should not yet have been fully optimized for power efficiency and power saving and as such things could get better yet.

smilingcrow
*Lifetime Patron*
Posts: 1809
Joined: Sat Apr 24, 2004 1:45 am
Location: At Home

Post by smilingcrow » Fri Jun 06, 2008 4:08 am

krille wrote:As for single threaded performance increase, the only such test that was made showed a clock-per-clock performance gain of 26% for the aforementioned system power increase of 10%.
The original single-threaded Penryn Cinebench numbers were incorrect and when corrected the single threaded Cinebench benchmarks shows only a 2% increase in core-to-core performance from Penryn to Nehalem at the same clock speed.
They didn’t show the power consumption but at this stage with this immature platform the performance per watt for applications that can utilise only 1 or 2 threads will likely be worse than for Penryn.

This version of Nehalem is aimed at high end desktops & workstations and will be priced accordingly. It’ll be more interesting to see how the mainstream version with dual memory channels pans out.

FartingBob
Patron of SPCR
Posts: 744
Joined: Tue Mar 04, 2008 4:05 am
Location: London
Contact:

Post by FartingBob » Fri Jun 06, 2008 4:12 am

Considering that the board was really just a production prototype (as was the chip itself, unless Intel are already making them 6 months ahead to prepare for big demand...but i doubt that alot), the performance is incredible.
I hope the TDP comes down a little, 135w will be hard to cool quietly. Around 100w would be much more manageable.

Matija
Posts: 780
Joined: Sat Mar 17, 2007 3:17 am
Location: Croatia

Post by Matija » Fri Jun 06, 2008 4:24 am

Four cores with hyperthreading... Dunno. Might not be so power-efficient after all.

jaganath
Posts: 5085
Joined: Tue Sep 20, 2005 6:55 am
Location: UK

Post by jaganath » Fri Jun 06, 2008 4:45 am

jaganath only seems to have made a very cynical comment,
skeptical, perhaps is a term with fewer negative connotations. Intel's interest in power efficiency is only a relatively recent phenomenon, painful memories of Prescott and to a lesser extent Northwood still remain. it wouldn't be the first time Intel has "bought" increased performance at the "cost" of extra power consumption.

krille
Posts: 357
Joined: Thu Mar 23, 2006 4:56 am
Location: Sweden

Post by krille » Fri Jun 06, 2008 5:06 am

jaganath wrote:
jaganath only seems to have made a very cynical comment,
as a native English speaker I took the "1% for 1%" at face value, its exact meaning. that's not cynicism, you may need a new English dictionary. Intel's performance per watt credentials are only recently improved, painful memories of Prescott and to a lesser extent Northwood still remain. it wouldn't be the first time Intel has "bought" increased performance at the "cost" of extra power consumption.
Being a native English speaker doesn't justify you extrapolating things that aren't there. If your power efficiency predictions are right ("1% for 1%"), Intel is right. If Intel is right, you are not necessarily right however. The prediction of yours was indeed cynical ("they may not lie, but they certainly won't do any better than their worst guarantee", sort of). Now what's the need of discussing this.

In mathematical terms, Intel says: [Nehalem's Performance Per Watt]/[Penryn's "PPW"]>=1
You say: [Nehalem's PPW]/[/Penryn's PPW]=1
It doesn't take a rocket scientist to notice the difference.

Besides, as a native English speaker you should try to spell properly. As of now the way you write is actually deeply miscrediting your whole native English speaker argument. So don't go around flaunting your first language like that if you're going to be sloppy, it doesn't help you.

smilingcrow > Darn. That's bad news. If single-threaded performance isn't improving, there probably will be little reason to upgrade. Hopefully proper triple-channel memory as well as an optimized motherboard/chipset, BIOS and drivers may help. Let's hope!


edit:
jaganath wrote:
jaganath only seems to have made a very cynical comment,
skeptical, perhaps is a term with fewer negative connotations. Intel's interest in power efficiency is only a relatively recent phenomenon, painful memories of Prescott and to a lesser extent Northwood still remain. it wouldn't be the first time Intel has "bought" increased performance at the "cost" of extra power consumption.
Err... it appears you edited your post slightly before I posted mine. Now it's a much less hostile version, that's good. :)

Anyway, I agree skeptic would work as well.

Tzupy
*Lifetime Patron*
Posts: 1561
Joined: Wed Jan 12, 2005 10:47 am
Location: Bucharest, Romania

Post by Tzupy » Fri Jun 06, 2008 7:40 am

I don't understand why people complain about the 135W TDP. For such computing power it's acceptable IMO, considering the 45 nm process.
The triple-channel DDR3 controller could use upto 15W of power, so it's going to be a bit more difficult to cool the CPU quietly, compared with a Penryn.
No one forces you to buy the quad-core, I bet the dual-core is going to be good enough for most people and use much less power.
Or, if you must have the quad-core and want to cool it very quietly, wait for the 32 nm process (or for AMD's Bulldozer).

krille
Posts: 357
Joined: Thu Mar 23, 2006 4:56 am
Location: Sweden

Post by krille » Fri Jun 06, 2008 8:04 am

Tzupy wrote:I don't understand why people complain about the 135W TDP. For such computing power it's acceptable IMO, considering the 45 nm process.
The triple-channel DDR3 controller could use upto 15W of power, so it's going to be a bit more difficult to cool the CPU quietly, compared with a Penryn.
No one forces you to buy the quad-core, I bet the dual-core is going to be good enough for most people and use much less power.
Or, if you must have the quad-core and want to cool it very quietly, wait for the 32 nm process (or for AMD's Bulldozer).
I for one prefer an OC'ed dual-core to a quad-core any time. (And as all SPCR'ers now this is a valid comparison as for us it's all about PPW.) And with the reintroduction of HT a dual-core might be the sweet spot (seeing as newer games and application often can take slight advantage of a quad-core over a dual-core). I am hopeful! :D

smilingcrow
*Lifetime Patron*
Posts: 1809
Joined: Sat Apr 24, 2004 1:45 am
Location: At Home

Post by smilingcrow » Fri Jun 06, 2008 8:12 am

FartingBob wrote:I hope the TDP comes down a little, 135w will be hard to cool quietly. Around 100w would be much more manageable.
Intel already has desktop Penryn Quad cores @ 3.2GHz rated at 136W & 150W depending on the socket type and they do pretty well in real-world power tests so I don’t see that the TDP is an issue. The better performance per watt with heavily multi-threaded applications is a plus over Penryn but the extra power draw of the IMC is a minus with regard to CPU cooling.

smilingcrow
*Lifetime Patron*
Posts: 1809
Joined: Sat Apr 24, 2004 1:45 am
Location: At Home

Post by smilingcrow » Fri Jun 06, 2008 8:17 am

krille wrote:I for one prefer an OC'ed dual-core to a quad-core any time. (And as all SPCR'ers now this is a valid comparison as for us it's all about PPW.) And with the reintroduction of HT a dual-core might be the sweet spot (seeing as newer games and application often can take slight advantage of a quad-core over a dual-core). I am hopeful! :D
The raison d'être of Nehalem seems to be for workstation and server loads so I’m not sure that a dual core version is going to be a big step forward over Penryn.

krille
Posts: 357
Joined: Thu Mar 23, 2006 4:56 am
Location: Sweden

Post by krille » Fri Jun 06, 2008 9:21 am

smilingcrow wrote:The raison d'être of Nehalem seems to be for workstation and server loads so I’m not sure that a dual core version is going to be a big step forward over Penryn.
Well it should be, seeing as it's Intel's "tock". It's quite possible there won't be too much reason to upgrade to it, though, until its "tick", considering it's DDR3 and all.

Post Reply