Early 45nm Deneb Spotted

[RallyMaster]

My stock E7200 does 1M SuperPi in 20 seconds. I think that pretty much says everything.

 

[bryanW1995]

Originally posted by: Ocguy31
I dont have to go to a chinese website to spot a 45nm....I just need to open my computer case....like 7 months ago.

lol

 

[bryanW1995]

Originally posted by: RallyMaster
My stock E7200 does 1M SuperPi in 20 seconds. I think that pretty much says everything.

well, actually, it says that your e7200 does 1m superpi in 20 seconds. amd has not done well EVER in superpi, even when they dominated intel during the skt 939 heyday.

 

[Toadster]

hmm - my rig does SuperPi in 16 seconds (1M) - what's the big deal ???

 

[Dadofamunky]

Originally posted by: aigomorla

Ive seen Neha, ive played, touched, drooled.
After that wonderful experience, i gave away erinyes to setup a new bench. This is how much of a WOW i said after i came back.

So does your NDA extend to not telling us how many cores this Neha had and what OS it was running on???? Clock speed, perhaps?

I'm sorry, but can't you open the kimono just a little bit more?

 

[zach0624]

Originally posted by: Toadster
hmm - my rig does SuperPi in 16 seconds (1M) - what's the big deal ???

It may not seem like much ( I assume you run an intel rig) but for AMD this is a smoking super pi time.



I really don't like super pi for comparing rigs across any different types of cpus (company, family, cache size) because it can vary greatly depending on many system variables like I bet for many apps this cpu would beat my 4300 at 3ghz at stock speeds but my 4300 could get sub 20 times in its sleep.

 

[lopri]

Surely Intel wouldn't tell everything to the real world. They have to somehow metamorphose a PCIe switch to a 'north bridge' and give it a l33t name like 'X58' so that they can keep selling chipsets. (and lawsuits) QPI my a**.

 

[myocardia]

Originally posted by: Extelleron
Nehalem is a server CPU at heart IMO. Every improvement in Nehalem is more important in the server space than on the desktop. HyperTransport has given AMD an advantage in servers for years now, but it hasn't done much for them on the desktop. I don't think QPI will do much for Intel on the desktop either. SMT makes sense for servers as I said earlier, but not for desktops at this point.

Umm, you must not know (or at least not remember) that the Athlon 64 was designed from the ground up as a server CPU, as was the Phenom.

 

[myocardia]

Originally posted by: Somniferum
The funny thing is, I feel the same way about my 3800+ @2.53GHz. Could it be that CPUs haven't really advanced all that much in the past couple of years, when it comes to real world performance?

Actually, CPU speeds have come a long way since the X2's. Now, whether or not you personally need anymore CPU speed is another matter altogether. Of course, I see that as the problem that both AMD & Intel face: they both already have many, many times more speed than either Internet Explorer or M$ Office requires. And if you stop and think about it, besides us few gamers, that's what ~99% of the world buys a computer to do.

Sure, SuperPi might be fearing for its life, but last time I checked, pi was a constant.

I love this quote. Mind if I use it some other time (here, at anandtech), as long as I give you credit?

 

[Martimus]

I don't understand the comparison of this score to a Yorkfield and saying that a Nehalem would smoke it because of the comparison. Since the Nehalem has slower cache and less cache to a Yorkfield, it will probably have a worse score to a yorkfield in SuperPI as well. That doesn't mean that it won't be much faster in most applications. The same goes for this Deneb. We can only tell that it does better than the current generation Phenom at this benchmark. This could be because it has more L3 cache, or maybe the L2 cache is running quicker, or something else is true.

Aigo, why don't you ask your friend to run SuperPI and downclock your QX9650 to the clock speed of the Nehalem and run SuperPI, and see which one has a better score. You don't have to tell us what the score is, but I would expect your chip to win.

 

[Extelleron]

Originally posted by: myocardia

Originally posted by: Extelleron
Nehalem is a server CPU at heart IMO. Every improvement in Nehalem is more important in the server space than on the desktop. HyperTransport has given AMD an advantage in servers for years now, but it hasn't done much for them on the desktop. I don't think QPI will do much for Intel on the desktop either. SMT makes sense for servers as I said earlier, but not for desktops at this point.

Umm, you must not know (or at least not remember) that the Athlon 64 was designed from the ground up as a server CPU, as was the Phenom.

What does that have to do with what I said about Nehalem?

A64 was indeed designed with many attributes that made it very attractive for server use, most notably HyperTransport. But these ended up making a big deal on the desktop as well. K8 also incorporated design changes beyond HT/IMC as compared to K7.

I think that K7 needed HyperTransport/IMC more than C2D needs QPI/IMC though. Remember that K7 had a dual-pumped FSB, meaning the max FSB it ever reached was 333-400MHz. P4 at that time was using a 533-800MHz FSB depending on the model, with 800MHz becoming into 2003 and certainly with Prescott-based P4s. The bandwidth that HyperTransport offered was quite important to K7 with its 333-400MHz FSB.

Core 2 is a bit of a different story. It has a quad pumped FSB of 1333MHz. Sure, you see gains moving to a higher FSB with C2D, so there is a need for more bandwidth and that is where QPI will come into play. But the gains are very minimal for the most part. Without a doubt moving to an interface like QPI is necessary for future apps, but for desktop performance, I don't think you will see much difference. On the server side of course....definitely it is important right now.

And as for the IMC.... K7 had at max a 512KB L2. AMD is best suited with an IMC because they cannot fit as much cache on die as Intel can. AMD's cache is larger and they cannot afford the die space either. With an IMC, AMD doesn't need as much cache on board so they could stick with 512KB-1MB per core, meanwhile we see Intel with 2MB-3MB per core with the latest C2D's. Current C2Q's have 12MB L2... moving to Nehalem will bring gains from faster memory access, but remember it has less cache than Yorkfield.

Nehalem is going to be a great server CPU, just like K8 was. But I think its new features and tweaks may not lend itself as well to desktop apps as K8 did (for its time). It will be an improvement over Core 2 around the board, but you will notice the differences most in server apps.

 

[Idontcare]

Originally posted by: Extelleron
Nehalem is going to be a great server CPU, just like K8 was. But I think its design may not lend itself as well to desktop apps as K8 did (for its time).

You lost me with this closing statement.

What is it about Nehalem that could possibly not lend itself well to today's desktop applications inasmuch as any other processor currently lends itself?

What might any other processor entail in order to better lend itself to desktop apps more so than Nehalem?

The only thing I see that makes releasing Nehalem today on the desktop any less more exciting than the K8's release 3 years ago is that 8-thread 4GHz processors with triple-channel DDR3 and super-low latency IMC's is just ridiculous overkill for the desktop market in every sense of the word.

If even 1% of Nehalem users actually peg their 8 threads at 100% utilization doing something for >1hr per day I would be utterly astounded. Web surfing and emailing is hardly challenging for any modern CPU.

 

[harpoon84]

Originally posted by: Extelleron
Nehalem is going to be a great server CPU, just like K8 was. But I think its design may not lend itself as well to desktop apps as K8 did (for its time).

What 'desktop' apps are you talking about exactly? Most general purpose desktop productivity apps are not that CPU bound by nature, the most demanding (and time consuming) apps for desktop users would probably be video encoding and 3D rendering, both of which Nehalem appears to excel at:

http://www.anandtech.com/cpuch...howdoc.aspx?i=3326&p=6
http://www.anandtech.com/cpuch...howdoc.aspx?i=3326&p=7

 

[Extelleron]

Originally posted by: Idontcare

Originally posted by: Extelleron
Nehalem is going to be a great server CPU, just like K8 was. But I think its design may not lend itself as well to desktop apps as K8 did (for its time).

You lost me with this closing statement.

What is it about Nehalem that could possibly not lend itself well to today's desktop applications inasmuch as any other processor currently lends itself?

What might any other processor entail in order to better lend itself to desktop apps more so than Nehalem?

The only thing I see that makes releasing Nehalem today on the desktop any less more exciting than the K8's release 3 years ago is that 8-thread 4GHz processors with triple-channel DDR3 and super-low latency IMC's is just ridiculous overkill for the desktop market in every sense of the word.

If even 1% of Nehalem users actually peg their 8 threads at 100% utilization doing something for >1hr per day I would be utterly astounded. Web surfing and emailing is hardly challenging for any modern CPU.

You misunderstood what I was trying to say (I didn't word it very well).

Nehalem is going to be better than C2D for everything imaginable (the only thing I could possibly think of is a contrived situation where an app would fit into C2Q's 12MB cache but not the 8MB L3 + 1MB L2 cache of Nehalem, and even that is iffy). But the additional features in Nehalem are not going to lend themselves as well to desktop apps as K8's additional features did or certainly what we see going from PD->C2D.

Nehalem will be significantly better in certain apps on the desktop and it should dominate the server market. Stuff like web browsing, email, word processing.... as you said, that is no longer CPU-bound on even the slowest modern CPUs. But what about stuff like gaming or general desktop apps that do not scale to 8 threads? That is where I see Nehalem being an incremental upgrade to Core 2. It seems to be that a lot of the performance improvement we have seen documented comes from apps that will use all 8 threads that Nehalem's SMT allows for.

Look at Cinebench for example, Anandtech's review. Cinebench is a perfect benchmark for something like this because it shows both single and multi-threaded performance. Nehalem provides only a ~3% single threaded improvement over Core 2, but a 21% improvement in the multi-threaded test.

In 3dsmax, a clearly multi-threaded test, we see a huge (~40%) performance improvement over Core 2, but remember that part of that comes from the outlier test CBALLS2. Most of the tests show around 20-30% improvement, with as low as 10% in one test.

Nehalem is going to be impressive, I'd sure like to own one when it comes out and so would every other enthusiast. But I am saying that I think if you expect to see a Core 2-type improvement here, then you will be disappointed. If you do rendering that uses 8 cores or run a server, Nehalem is up your alley. But unlike Core 2, Nehalem is not going to provide revolutionary performance in every single app that one could possibly run. Compared to P4, Core 2 was completely new (obviously it was evolved from P3 but that is another story).... Nehalem is a tweaked Core 2 with much more selective performance improvements.

 

[OCGuy]

As much as I think Nehalem will be absolutely amazing.....

I think people sitting back with P4 or older are insane not to jump on the current gravy train.

There is something for everyone right now, and it is extremely inexpensive.

4ghz capable 45nm CPU - $130

4GB DDR2 800- As low as $40?

ATI sweetened up the GPU market prices, so no matter which way you go its a win.


I think we will look back on 2007-2009 as a great time to build a well above-average system.

 

[harpoon84]

I see what you are trying to say - Core 2 made massive improvements across the board over P4, whereas Nehalem will only make marginal improvements in single threaded performance, but hefty improvements in multithreaded performance. This should come as no surprise though - single threaded performance is already very strong on Core 2 - Nehalem is addressing the only remaining 'weakness' in Core 2, which is multithreaded performance, especially on the multi socket level, but even on desktop apps this is sometimes evident eg. in Cinebench Core 2 gets about a 3.6x speedup with 4 threads, Phenom gets about a 3.95x speedup. Nehalem exceeds 4x speedup thanks to SMT, so in essense its a Core 2 on multithreading steroids.

 

[Somniferum]

Originally posted by: myocardia

Originally posted by: Somniferum
The funny thing is, I feel the same way about my 3800+ @2.53GHz. Could it be that CPUs haven't really advanced all that much in the past couple of years, when it comes to real world performance?

Actually, CPU speeds have come a long way since the X2's. Now, whether or not you personally need anymore CPU speed is another matter altogether. Of course, I see that as the problem that both AMD & Intel face: they both already have many, many times more speed than either Internet Explorer or M$ Office requires. And if you stop and think about it, besides us few gamers, that's what ~99% of the world buys a computer to do.

Well, I did say "when it comes to real world performance." I'm a gamer as well, and SuperPi speeds notwithstanding, where is the killer app that would convince me to upgrade to C2D or Quad core right now? I can play Crysis with no issues, it just requires a little tweaking. Assassin's Creed runs beautifully, Grid even more so, and this is with 16xQ AA and 16x AF. Maybe if I was an RTS fan I'd feel differently?

Sure, SuperPi might be fearing for its life, but last time I checked, pi was a constant.

I love this quote. Mind if I use it some other time (here, at anandtech), as long as I give you credit?

I would be honored.

 

[harpoon84]

Originally posted by: Somniferum
Well, I did say "when it comes to real world performance." I'm a gamer as well, and SuperPi speeds notwithstanding, where is the killer app that would convince me to upgrade to C2D or Quad core right now? I can play Crysis with no issues, it just requires a little tweaking. Assassin's Creed runs beautifully, Grid even more so, and this is with 16xQ AA and 16x AF. Maybe if I was an RTS fan I'd feel differently?

An overclocked C2D @ 4GHz would probably allow you to play such games at higher or max details (no 'tweaking') whilst giving you a much higher framerate at the same time. Whether you 'need' such performance is another matter altogether.

 

[Idontcare]

Originally posted by: Extelleron
You misunderstood what I was trying to say (I didn't word it very well).

Nehalem is going to be better than C2D for everything imaginable (the only thing I could possibly think of is a contrived situation where an app would fit into C2Q's 12MB cache but not the 8MB L3 + 1MB L2 cache of Nehalem, and even that is iffy). But the additional features in Nehalem are not going to lend themselves as well to desktop apps as K8's additional features did or certainly what we see going from PD->C2D.

Nehalem will be significantly better in certain apps on the desktop and it should dominate the server market. Stuff like web browsing, email, word processing.... as you said, that is no longer CPU-bound on even the slowest modern CPUs. But what about stuff like gaming or general desktop apps that do not scale to 8 threads? That is where I see Nehalem being an incremental upgrade to Core 2. It seems to be that a lot of the performance improvement we have seen documented comes from apps that will use all 8 threads that Nehalem's SMT allows for.

Look at Cinebench for example, Anandtech's review. Cinebench is a perfect benchmark for something like this because it shows both single and multi-threaded performance. Nehalem provides only a ~3% single threaded improvement over Core 2, but a 21% improvement in the multi-threaded test.

In 3dsmax, a clearly multi-threaded test, we see a huge (~40%) performance improvement over Core 2, but remember that part of that comes from the outlier test CBALLS2. Most of the tests show around 20-30% improvement, with as low as 10% in one test.

Nehalem is going to be impressive, I'd sure like to own one when it comes out and so would every other enthusiast. But I am saying that I think if you expect to see a Core 2-type improvement here, then you will be disappointed. If you do rendering that uses 8 cores or run a server, Nehalem is up your alley. But unlike Core 2, Nehalem is not going to provide revolutionary performance in every single app that one could possibly run. Compared to P4, Core 2 was completely new (obviously it was evolved from P3 but that is another story).... Nehalem is a tweaked Core 2 with much more selective performance improvements.

Ah I see now, excellent, I agree to the fullest with your statements.

To put it another way we could say that there isn't much in the current penryn design which critically bottlenecks it's IPC with current desktop apps...save for those corner cases involving unaligned cache steps and things that saturate a quad-pumped FSB (bandwidth limitations).

For those corner cases on the desktop segment the Nehalem architecture will markedly triumph over Penryn at similiar clockspeeds as these weaknesses in Penryn are expected to be handidly dealt with by Nehalem's improvements.

For everything else on the desktop (your single-threaded and perhaps up to quad threaded), Nehalem will need to bring more clockspeed to the table as IPC by itself is not devastingly improved upon unless you walk out the thread axis to something >4.

It does give Deneb a shot at effectively catching up to the performance numbers that desktop users will actually use...does nothing to ASP's though if Intel really does hawk their 2.66GHz bloomfield for $284 as Amberclad reported.

 

[VirtualLarry]

Originally posted by: Idontcare
Ah I see now, excellent, I agree to the fullest with your statements.

To put it another way we could say that there isn't much in the current penryn design which critically bottlenecks it's IPC with current desktop apps...save for those corner cases involving unaligned cache steps and things that saturate a quad-pumped FSB (bandwidth limitations).

For those corner cases on the desktop segment the Nehalem architecture will markedly triumph over Penryn at similiar clockspeeds as these weaknesses in Penryn are expected to be handidly dealt with by Nehalem's improvements.

For everything else on the desktop (your single-threaded and perhaps up to quad threaded), Nehalem will need to bring more clockspeed to the table as IPC by itself is not devastingly improved upon unless you walk out the thread axis to something >4.

It does give Deneb a shot at effectively catching up to the performance numbers that desktop users will actually use...does nothing to ASP's though if Intel really does hawk their 2.66GHz bloomfield for $284 as Amberclad reported.

Well said. That makes a lot of sense. So in other words, Nehalem is more evolutionary than revolutionary in terms of IPC gains. So overall, for most desktop tasks, it's hardy faster than a Penryn, unless you are massively multitasking.

In real-world terms, running DVDShrink can take up to two threads, which it generally pegs at close to 100%, sometimes less. With an 8-thread Nehalem, I could run four of those tasks at one time, but then I would be *severely* I/O bound. I would need a RAID array or several work HDs to use at once, in order to be able to load the CPU to 100%.

So exploitation of the performance gains given by Nehalem, will require re-thinking the optimization of everything else that goes into the system.

Personally, I still think that Microsoft is missing the boat on not offering a multi-User version of Vista, one in which you can use a multi-headed video card, and some extra USB keyboards and mice, and allow multiple users to log in and use the PC at one time, effectively turning a single PC into multiple PCs. Computers are becoming powerful enough to do that, even to be able to play 3D games amongst each other on the same CPU! So why isn't Microsoft jumping on this high-end bandwagon?

 

[uclaLabrat]

Originally posted by: VirtualLarry


Personally, I still think that Microsoft is missing the boat on not offering a multi-User version of Vista, one in which you can use a multi-headed video card, and some extra USB keyboards and mice, and allow multiple users to log in and use the PC at one time, effectively turning a single PC into multiple PCs. Computers are becoming powerful enough to do that, even to be able to play 3D games amongst each other on the same CPU! So why isn't Microsoft jumping on this high-end bandwagon?

Because microsoft would then want to charge people 8 times for one product license? That wouldn't fly :-D

 

[VirtualLarry]

Originally posted by: uclaLabrat
Because microsoft would then want to charge people 8 times for one product license? That wouldn't fly :-D

Charge for CALs, just like server OSes. The effect is the same, the user logs into the computer and uses the applications. What does it matter if they are using RDP protocol, or directly logging in to the machine.

 

[Idontcare]

Originally posted by: VirtualLarry

Originally posted by: uclaLabrat
Because microsoft would then want to charge people 8 times for one product license? That wouldn't fly :-D

Charge for CALs, just like server OSes. The effect is the same, the user logs into the computer and uses the applications. What does it matter if they are using RDP protocol, or directly logging in to the machine.

A possibility here is that it (multi-core & multi-thread/core) is occuring on a far faster timeline than Microsoft is able to change strategies and capitalize on.

Even the internet caught microsoft by total surprise and there efforts to jump on the bandwagon took quite a looong time. Netscape and AOL had a good run for a while.

Right now Microsoft is still blinded by their desire to dominate the web, ala there pre-occupation with google and yahoo.

This hardware cadence is just way too rapid for Microsoft to lock too. Just look at how disconnected the GPU and game programmers are at times and they are all far smaller and far nimbler business entities.

 

[Martimus]

Originally posted by: Idontcare
A possibility here is that it (multi-core & multi-thread/core) is occuring on a far faster timeline than Microsoft is able to change strategies and capitalize on.

Even the internet caught microsoft by total surprise and there efforts to jump on the bandwagon took quite a looong time. Netscape and AOL had a good run for a while.

Right now Microsoft is still blinded by their desire to dominate the web, ala there pre-occupation with google and yahoo.

This hardware cadence is just way too rapid for Microsoft to lock too. Just look at how disconnected the GPU and game programmers are at times and they are all far smaller and far nimbler business entities.

If we are lucky, it will give another software company an opening to build a competing Operating System to satisfy the new market possibility. It would be nice to get competition with Operating Systems again.