Sign in with
Sign up | Sign in
Your question

Hexus.net benchmarks Nehalem

Last response: in CPUs
Share
August 19, 2008 1:04:08 AM

http://www.hexus.net/content/item.php?item=15015&page=1

Check out the memory bandwith, insane! :ouch: 

Application performance ain't too shabby either, with Nehalem @ 2.93GHz generally beating the higher clocked QX9770 @ 3.2GHz, sometimes by huge margins. Gaming performance is poor, but can be attributed to improperly installed drivers apparently.
Anonymous
a b à CPUs
a b } Memory
August 19, 2008 1:17:05 AM

nice

either way a highly oced core 2 system will survive nehalem :D 

its good cause i just got a q9550 and i'm hoping to reach upwards of 3.8 ghz with it
a b à CPUs
August 19, 2008 1:27:15 AM

idk if I should wait for Nehalem.....i just like to play some games when I have time lol .....I will propably go with q6600 and oc it :) 
Related resources
Can't find your answer ? Ask !
August 19, 2008 1:40:48 AM

or e8500, that should do a bit better with games
a b à CPUs
a b } Memory
August 19, 2008 2:02:25 AM

Quote:
nice

either way a highly oced core 2 system will survive nehalem :D 

its good cause i just got a q9550 and i'm hoping to reach upwards of 3.8 ghz with it

We'll have to see how well Nehalem OCs. I have high hopes :D 
August 19, 2008 2:40:52 AM

Damn, they're taking AMD's only hope away, artificial memory benchmarks.

What are all the fanboys going to say now!?
August 19, 2008 3:05:03 AM

Say about what?? You mean you're actually impressed? Sure it's an improvement, but not by as much as the Intel fanboys were frothing at the mouth about. We're talking about 8 threads here. The way this chip was hyped, we should be seeing enormous gains in multithreaded situations.
August 19, 2008 3:45:11 AM

piesquared said:
The way this chip was hyped, we should be seeing enormous gains in multithreaded situations.


So up to 35% faster clock-for-clock vs Intel's current generation in highly threaded applications and absolutely destroying AMD's current chips isn't good enough?

What exactly were you expecting?
Anonymous
a b à CPUs
a b } Memory
August 19, 2008 4:18:35 AM

500 % increase in speed :D 

that way stuff that already doesn't need the full potential of a core 2 duo system can fly at near break-neck speeds :p 
a b à CPUs
August 19, 2008 4:20:26 AM

^+1
not to mention it was on immature hardware and drivers

Edit: this was at mark g
August 19, 2008 4:23:08 AM

2 key phrases there: "up to 35%" and "highly threaded applications". There was only one test where Nehalem was 33% faster, povray. And that was 64 bit where Core 2 is crippled, losing 10% in transition. So it's a <20% improvement. However it is a good marketing tactic, i'll give them that.

Cinebench show's decent improvements from 32 bit to 64 bit. However this is a meger 7% in 32 bit and 15% in 64 bit. And this is a multithreaded benchmark, 8 threads vs 4 threads.

Let's move on.

8% slower in Hexus wav.

14% faster in Divx (enhanced multithreading)

Winrar shows a very nice improvement.


Gaming. Well, there seems to be a software issue. Strange though that Intel would set up a demo of their latest and greatest and intentionaly cripple it.
Or do we actually believe that Intel had no knowledge of Hexus getting their hands on this thing. I find that HIGHLY unlikely.

So yeah, I was expecting much more. Can't say i'm dissapointed though, sorry.
a b à CPUs
August 19, 2008 4:27:48 AM

how is Core2 crippled in 64-bit. Link?
a b à CPUs
a b } Memory
August 19, 2008 4:34:53 AM

piesquared said:
2 key phrases there: "up to 35%" and "highly threaded applications". There was only one test where Nehalem was 33% faster, povray. And that was 64 bit where Core 2 is crippled, losing 10% in transition. So it's a <20% improvement. However it is a good marketing tactic, i'll give them that.

Cinebench show's decent improvements from 32 bit to 64 bit. However this is a meger 7% in 32 bit and 15% in 64 bit. And this is a multithreaded benchmark, 8 threads vs 4 threads.

Let's move on.

8% slower in Hexus wav.

14% faster in Divx (enhanced multithreading)

Winrar shows a very nice improvement.


Gaming. Well, there seems to be a software issue. Strange though that Intel would set up a demo of their latest and greatest and intentionaly cripple it.
Or do we actually believe that Intel had no knowledge of Hexus getting their hands on this thing. I find that HIGHLY unlikely.

So yeah, I was expecting much more. Can't say i'm dissapointed though, sorry.


Keep in mind that all of those numbers you are quoting are a speed comparison between a 3.2GHz QX9770 and a 2.93GHz Nehalem. It isn't slower than the QX6800 (a closer match in clock speed) in anything - it at least matches the current offerings clock for clock in every way (and keep in mind, the 2.93GHz model that was tested is the midrange version - there will be an extreme edition at 3.2GHz).
August 19, 2008 4:56:59 AM

BadTrip said:
how is Core2 crippled in 64-bit. Link?


There are a bunch of optimizations for 32-bit code in Core2 which aren't implemented for 64-bit code; but which are in Nehalem.

For example, if I remember correctly a 32-bit compare followed by a conditional jump is treated as a single instruction inside the Core2 CPU, but a 64-bit compare followed by a conditional jump is treated as two instructions; Nehalem treats both cases the same.

This is precisely why 64-bit benchmarks are interesting to see, because they're where Nehalem will show the greatest increase over Core2 and Phenom.
August 19, 2008 5:05:14 AM

BadTrip said:
how is Core2 crippled in 64-bit. Link?


Well, crippled might be to strong of word. I'm refering to macro-op fusion.

Related or not, the point is c2d dropped from 2878 to 2617 for 64-bit, while Nehelam's drop was insignificant. Which is why the results look more impressive than they are in that particular bench.
August 19, 2008 5:08:36 AM

piesquared said:
Which is why the results look more impressive than they are in that particular bench.


Are you claiming that comparing a 64-bit Core-2 to a 64-bit Nehalem is somehow a defective benchmark? 64-bit is where the majority of software will be going over the next couple of years, so it's far more representative of the difference between the CPUs than 32-bit code.
a c 127 à CPUs
a b } Memory
August 19, 2008 5:15:30 AM

No you have to remember he is primarily a AMD fan and even a nice boost like Nehalem is showing seems insignificant.

AMD fans used to rely on memory bandwidth for server apps but if it is DOUBLING the bandwidth for a Phenom 9950 then its going to do some hurting in the server arena.

Heck People were saying that QPI is fast but that fast....dude thats INSANE bandwidth right there.... and its using only 1066MHz DDR3.....I wounder what 1333 or even 1600MHz DDR3 would provide.....
August 19, 2008 5:29:06 AM

MarkG said:
Are you claiming that comparing a 64-bit Core-2 to a 64-bit Nehalem is somehow a defective benchmark? 64-bit is where the majority of software will be going over the next couple of years, so it's far more representative of the difference between the CPUs than 32-bit code.


Nope, what i'm saying is, although Nehelam appears to show a significant improvement moving from 32-bit to 64-bit in povray, that increase is entirely due to C2E's decrease in said benchmark.
a b à CPUs
a b } Memory
August 19, 2008 5:52:25 AM

Actually, I would say it tramples everything in both 32 and 64 bit. No need to specify which - it wins quite handily in both cases.
August 19, 2008 6:49:39 AM

piesquared said:
2 key phrases there: "up to 35%" and "highly threaded applications". There was only one test where Nehalem was 33% faster, povray. And that was 64 bit where Core 2 is crippled, losing 10% in transition. So it's a <20% improvement. However it is a good marketing tactic, i'll give them that.

Cinebench show's decent improvements from 32 bit to 64 bit. However this is a meger 7% in 32 bit and 15% in 64 bit. And this is a multithreaded benchmark, 8 threads vs 4 threads.

Let's move on.

8% slower in Hexus wav.

14% faster in Divx (enhanced multithreading)

Winrar shows a very nice improvement.


Gaming. Well, there seems to be a software issue. Strange though that Intel would set up a demo of their latest and greatest and intentionaly cripple it.
Or do we actually believe that Intel had no knowledge of Hexus getting their hands on this thing. I find that HIGHLY unlikely.

So yeah, I was expecting much more. Can't say i'm dissapointed though, sorry.


I'm not sure what you were expecting (miracles? cure for cancer? :p ) but keep in mind this is the $530 2.93GHz model trouncing the currently $1500 3.2GHz QX9770 in MT, whilst holding its own in ST. The 3.2GHz Nehalem EE will 'only' cost $999 (hey, its all relative right?) and boost performance a further ~9% over the 2.93GHz model tested here. So on a clock for clock basis it looks like a 5 - 10% jump in ST, and 20 - 40% jump in MT.

Finally, its not final production silicon and thus most likely not fully optimised at this point, the reviewer seems confident we'll see even higher numbers by release, at the very least for gaming, well I hope so anyway! ;) 
August 19, 2008 7:31:38 AM

When it comes to apps and such, it also comes to OS. Whats Windows 7? Exclusively 64bit? Its good to see theyre pushing ahead in 64bit, but until its used, the same 32bit increase would be nice to see. I think thats his point. I somewhat agree, tho it is progress in the right direction
August 19, 2008 7:44:53 AM

piesquared said:
Nope, what i'm saying is, although Nehelam appears to show a significant improvement moving from 32-bit to 64-bit in povray, that increase is entirely due to C2E's decrease in said benchmark.


Are you saying that it does not matter that nehelem is faster because core2 was slower. I dont get it. Only thing that matters is that it is faster.
August 19, 2008 7:55:22 AM

gallag said:
Are you saying that it does not matter that nehelem is faster because core2 was slower. I dont get it. Only thing that matters is that it is faster.


piesquared sure has an uncanny knack of turning a positive into a negative. ;) 

You'd think Nehalem improving on 64bit performance would be greatly welcomed, especially since its the way of the future. But instead he focuses on the fact that Core 2 lacks macro-ops fusion in 64bit and somehow turns it into an 'excuse' why Nehalem is substantially faster in 64bit! Amazing.

August 19, 2008 8:02:36 AM

Also, they say : "Nehalem won't matter much if you play games that are limited by the graphics subsystem, usually at higher resolutions and image-quality settings, but it's always nice to have extra power under the hood, we suppose." But looking at those numbers, its under the prefered 60 fps http://www.hexus.net/content/item.php?item=15015&page=8 So basically their statement and observations are invalid . Its well known that current cpus become a slowdown even at higher res with todays graphics cards, and this is a poor showing by Nehalem

August 19, 2008 8:20:13 AM

JAYDEEJOHN said:
Also, they say : "Nehalem won't matter much if you play games that are limited by the graphics subsystem, usually at higher resolutions and image-quality settings, but it's always nice to have extra power under the hood, we suppose." But looking at those numbers, its under the prefered 60 fps http://www.hexus.net/content/item.php?item=15015&page=8 So basically their statement and observations are invalid . Its well known that current cpus become a slowdown even at higher res with todays graphics cards, and this is a poor showing by Nehalem


Selective reading much? It's clearly a driver issue. The numbers for gaming are meaningless at this stage.

Quote:
We alluded to the fact that it wasn't all roses in the Nehalem garden, as far as the test box was concerned. One look in Device Manager showed that not all the correct drivers had been installed, which did little to hinder 2D performance, but played a part in sub-optimal 3D results.

A table has been included to highlight the results we observed, but it is abundantly clear that something was awry in the test box.


August 19, 2008 8:30:22 AM

We will see, its getting close to release, and if it doesnt almost double its gaming performance, itll easily be the weakpoint for i7
August 19, 2008 8:36:03 AM

What Im saying is, at 80% more, youll get around the increases we see in the other tests. So, not being selective, and not overly optimistic either, thats about what it needs.
August 19, 2008 8:39:08 AM

What concerns me is, using the AA and AF modes shows a huge drop off. And the claims that higher res doesnt effect cpus is bull. These new cards are made for those higher res'. No one plays with no AA and AF with these sorts of setups. Anyone like to take a poke as to why adding AA and AF has such a dramatic effect?
August 19, 2008 8:41:29 AM

JAYDEEJOHN said:
We will see, its getting close to release, and if it doesnt almost double its gaming performance, itll easily be the weakpoint for i7


What are you talking about? Double its gaming performance over what? Again, you can't base anything on those numbers as the drivers aren't correctly installed.

FWIW I don't expect Nehalem to be a lot faster than Penryn in gaming, unless the game was heavily multithreaded such as FSX. I guess you could call that a 'weak point' but GPU limitation is unavoidable on a lot of games at high res + details, and a faster CPU is not going to help there.

JAYDEEJOHN said:
What concerns me is, using the AA and AF modes shows a huge drop off. And the claims that higher res doesnt effect cpus is bull. These new cards are made for those higher res'. No one plays with no AA and AF with these sorts of setups. Anyone like to take a poke as to why adding AA and AF has such a dramatic effect?

AA/AF only increases the workload of the GPU. The workload on the CPU is the same whether it is 640 x 480 or 2560 x 1200.
August 19, 2008 8:46:01 AM

So its a skewed test. OK, I hope it turns out different in the end. Already we are seeing cpus needing to be oceed in some games for optimal fps, and I was hoping Nehalem would elimanate this, but it doesnt look like it.
August 19, 2008 8:49:33 AM

Look to many benchmarks in games, especially at higher res. Upping the cpu speed helps alot on certain games. Its a fact. Just like any other app, ocing helps, its just that you have min frame rates and other demands in gaming where the cpu does play a prominent role. Oceed, you get better performance.
August 19, 2008 8:56:59 AM

JAYDEEJOHN said:
So its a skewed test. OK, I hope it turns out different in the end. Already we are seeing cpus needing to be oceed in some games for optimal fps, and I was hoping Nehalem would elimanate this, but it doesnt look like it.


What games would they be? FSX? Sup Com? WiC? In any CPU limited game Nehalem will probably be a bit faster, but I don't think its realistic to be expecting the same 20 - 40% gains in rendering and video encoding, where we see the advantage of SMT. Not many games can take advantage of 8 threads (FSX can, but thats an exception).
August 19, 2008 9:03:07 AM

OK, dont believe me, maybe you will believe this guy "Until recently, the 280 and the 4870 X2 are showing CPU limitations even at high resolutions... up to 1920x1200, 2560x1600 the only exception." Thats a quote from JJ. Im not making it up, nor am I slamming Intel. Try Crysis, AoC to name a few, and theres others
August 19, 2008 9:10:15 AM

JAYDEEJOHN said:
Look to many benchmarks in games, especially at higher res. Upping the cpu speed helps alot on certain games. Its a fact. Just like any other app, ocing helps, its just that you have min frame rates and other demands in gaming where the cpu does play a prominent role. Oceed, you get better performance.


You've got it the wrong way around. The benefits of a faster CPU taper off at higher resolutions because the GPU becomes the bottleneck.

http://www.guru3d.com/article/cpu-scaling-in-games-with...

As I said earlier, in gaming the workload for the CPU is the same regardless of resolution. Increasing resolution/AA/AF only taxes the GPU, not the CPU.
August 19, 2008 9:14:03 AM

Been gaming much? Why in some benches, some cards are actually SLOWER in lower res?
August 19, 2008 9:16:37 AM

When the gpu asks more from the cpu than it can deliver, it causes problems in the gpu delivery. When the gpu is challenged, that allows for the cpu to catch up. It didnt used to be this way. Times have changed
a b à CPUs
a b } Memory
August 19, 2008 12:30:24 PM

Reads article...

Compares the pre-production Bloomfield to the Phenom BE sampled...


Walks away.
August 19, 2008 12:44:35 PM

Good for crunching numbers in multithreaded apps, so so in gaming walks away disappointed
August 19, 2008 12:58:20 PM

Im hoping that it oces well, or comes in ata higher stock speed. Too many trends currently in gaming where a heavily oceed cpu is very helpful, and sometimes needed. The trend in faster gpus isnt changing, Im just hoping that Nehalem will keep pace
Anonymous
a b à CPUs
a b } Memory
August 19, 2008 1:15:19 PM

to be honest, i'm not actually that pleased with nehalem... sure I wasn't looking for stellar performance, but it really wasn't all that it was hyped to be... it was hyped to be the next core 2.... and its not.... its like... idk 3/4 of a step or something
a c 127 à CPUs
a b } Memory
August 19, 2008 1:21:26 PM

JAYDEEJOHN said:
Been gaming much? Why in some benches, some cards are actually SLOWER in lower res?


Did you just ask this? I don't want any part of this argument because I see it as useless BUT at lower res the game does become CPU limited. Thats a known fact hence why most of these games would make sense if the drivers for the CPU are not installed correctly.

At res such as 1280x1024 and the equivalent widscreen res, the GPU starts to take over as the main limitor.

Now yes if the drivers were not installed correctly the games FPS would be affected since its probably not getting the speed required to the GPU. But lets wait until we have a final system and final drivers before making a judgement that we all know is not possible to make without everything being set.
August 19, 2008 1:25:07 PM

Good you included main limitor, but, look at the 8800GTX, its 20 months old now. How does it compare to the 4870x2? Maybe 50%? And in another 18 months? Get where Im going? Nehalem needs to rock to keep up, no nickle and dime 5-12% increases seen over 6 months time
August 19, 2008 1:30:53 PM

4 Months ago, I posted a thread on here about this. Everyone said, no, a cpu today is just fine. At that time, I said, cpus are losing to gpus in output, and was hoping Nehalem would stem this trend. I dont see it. Will Nehalem oc better? or come in at higher stock speeds? And dont go off on the multi thread bunk, as we just havnt seen it. Im not picking on you jimmysmitty, but you of all people know, theres several games out currently thatre cpu diminished unless theyre oceed. Im looking towards Nehalem to change this. Am I wrong to hope?
a c 127 à CPUs
a b } Memory
August 19, 2008 1:34:56 PM

Well with the bandwidth that QPI is putting out on a UNFINISHED product I would think thats plenty of bandwidth to keep a GPU happy.

Of course we wtill have to wait till the final drivers/chip comes out for a full judgement call. But thus far it looks very decent for just adding a IMC and also a bit of improvement core wise. In fact if you calculate it the performance compared to the 3.2GHz chip on a clock per clock basis it still does better.

I expect the finalized product to increase single threaded persormance by about 9-10% but I still will withold judgement till I see a final product.
August 19, 2008 1:45:55 PM

How about oc? or stock speeds?
a c 127 à CPUs
a b } Memory
August 19, 2008 2:05:24 PM

I am guessing that the NDA is not fully released yet to allow it to be shown.

I have read a few sites that have been messing with the OCing state its quite good so we will have to wait and see.
August 19, 2008 2:10:18 PM

The general feeling Im getting is, unless these numbers are way off, and it doesnt come in with a faster clock, some people wont upgrade to it, or alot wont, just depends on how it finishes. Im hoping it kicks arse, but right now, this hexus "preview" to me isnt either doing it any favors, or is way off
a c 99 à CPUs
August 19, 2008 2:29:38 PM

jimmysmitty said:
Well with the bandwidth that QPI is putting out on a UNFINISHED product I would think thats plenty of bandwidth to keep a GPU happy.


A GPU doesn't have much to do with QuickPath/HyperTransport speed unless it is an IGP. In the case of an IGP, the QPI/HT link speed is important as it's the GPU's memory I/O link. There isn't a whole lot of bandwidth used between an IMC-containing CPU and the northbridge if there isn't an IGP sitting in that northbridge. Case in point: an X2 6400+ with a mere 2000 MT/sec on its HT link outperforms some Phenoms, despite the Phenoms having 3600-4000 MT/sec on their HT links. If the HT link speed were that important, then you'd expect to see a huge increase in GPU performance with the Phenoms.

Quote:
I expect the finalized product to increase single threaded persormance by about 9-10% but I still will withold judgement till I see a final product.


What do you base this on? The Core i7 has already gone through at least a few steppings already. Further steppings at this point are to correct bugs and increase yields and perhaps clock speed and power consumption, not to add new features. Getting a 9-10% increase in IPC between now and release would indicate there is some very significant problem with the implementation of the processor that is fixable with just a respin of the silicon. That's possible but unlikely as it would take a big problem like the L3 not working or the IMC having horrible latency and clock speed issues. The tests run thus far do not suggest anything like that, so I think that what you see for raw CPU performance now is what you'll get on launch day. A platform bug could hurt something badly, but it would be specific, such as a bad PCIe controller refusing to activate all links, choking the GPU's bandwidth or a badly-tuned VRM. Such a problem would manifest itself in only certain tests, such as high-resolution gaming benchmarks being crap or extremely high idle power draw. But it wouldn't affect CPU-based performance such as encoding one bit.

The other thing that could perhaps net the Core i7 that kind of performance later is a lack of specific Nehalem compiler optimization options in gcc, icc, or MSVCC today. But most software is compiled for a generic class of CPU (such as -march=i686, -mSSE on 32-bit OSes or generic -O optimizations on x86_64 OSes) and not specific models, so these kinds of optimizations would be basically only seen in "one-off" kinds of programs and not in general usage. Otherwise, any other CPU will either fail to execute the program (such as an older A64 single-core in a program with SSE3) or potentially run it with poor performance.
August 19, 2008 2:35:52 PM

Maybe as to this article, a 10% perf increase would be seen in gaming , due to complications, as said in the article?
August 19, 2008 2:37:38 PM

Reading now. And commentating.

Quote:
As a summary, we know that the monolithic, 45nm-based Nehalem microarchitecture has an integrated memory controller - supporting tri-channel DDR3-1,600 RAM for the desktop - QuickPath Interconnect, simultaneous multithreading, and a three-level cache hierarchy with a large pool shared amongst all cores.


As a AMD fan, well, this part made me giggle like a little girl.



I loved were the Blue Block of "IGraphics" fits in this....picture.

Sorry two pages of power point presentations doesn't clear, nothing. Like Larrabee, until i see a Nehalem on the test bed (or somebodys home) my predictions stand. Nehalem might flop due to software. Getting a 4 cores with HT ( 8 logical cores) is nice. Most apps only take advantage of 2 cores. HT might help, but why the hell you want 8 logical cores if apps just don't use them ?

Lets see if im right or whatnot. The next graph made me crack a smile. Quoting hexus once more:

Quote:



HEXUS' PiFast test calculates the constant Pi to 10m places, using a brute-strength approach. What's interesting is that the single-threaded test is almost as fast on the 2.93GHz Nehalem as on the 3.2GHz QX9770, suggesting that memory bandwidth is coming into serious play.

AMD's fastest consumer CPU, the 2.6GHz-clocked Phenom X4 9950 BE, is some significant way behind.


This just made me pop a vein in my forehead. Why ? Newegg prices:

AMD Phenom 9950: 235$
Intel QX9770 : 1459$

Nice comparison !!! It is really...stupid actually. Like comparing a Peugeot 206 1.4 HDI to my Lancia. My Lancia does 240 in less than 40 secs. The Peugeot will get there when George Bush will self proclaim as a Gay man. Both things wont happen.
But, ill keep it cool, and ill keep reading.

Another silly quote:

Quote:
Nehalem's performance is born from taking the Core 2 architecture as a base and adding sensible, performance-enhancing additions such as an integrated memory controller, QuickPath interconnect, tiered cache, and tri-channel memory. Last but not least, SMT (simultaneous multithreading) provides a healthy boost, too.


Every other Multi-Core/Multi-thread is based or has SMT instructions on it. Simultaneous Multithreading was first researched by IBM in 1968. It is like Ford announcing his "new" F150 with a 4 cylinder engine and a turbo. None of the latter are cutting edge innovations. They were, very long time ago.

In all honestly, that review showed the expected. Multi-thread apps shine, some others just don't add up. Normal benchmarks at this scale. Multi-Core was the future is now the present. A Many-Core approach will probably fail, by the same reason SunSPARC CPUs aren't used mainstream.
What i did not like was the sensationalist tone the reviewer gave to it. And by the way, i like to game, doing Winrar is just a hobbie. Sorry about the sarcastic tone but this review was kinda biased. And was short of amazing really with all this hype i was expecting much more.

Sorry about the sarcasm, but the interpretations of the results from THAT review is what i call crap.
!