noob confused about i5/i7

[plion]

From what I understand the i5 will have onboard pcie controller?

So how much impact will it have on gaming? Should I sell my i7 920 and wait?

All I have right now is the i7-920 and a cosmos s case, haven't bought anything else yet

 

[Lonyo]

i7 is supposed to be the high end.
It would be weird if i5 managed to outperform the i7 platform, and a bit of a backwards move from Intel.

 

[Imager]

I don't know much about the differences, but if you have a i7 now, why would you sell it? This is probably a better/more fair question for someone that is looking to buy new, but not already has bought. I wouldn't resell an i7 i've already purchased.

 

[OCGuy]

No, you should not sell what you currently have.

 

[plion]

thanks guys, your right I'm just being paranoid. I'll keep this baby. Now Which motherboard to go for now? lol ; )

 

[Idontcare]

Originally posted by: Lonyo
i7 is supposed to be the high end.
It would be weird if i5 managed to outperform the i7 platform, and a bit of a backwards move from Intel.

He's asking specifically in regards to gaming, and the question is quite relevant because the i5 has integrated PCIe whereas the i7 requires a "hop" over the QPI to the x58 chipset in order to access the PCIe and then onwards to the GPU.

It stands to reason to expect gaming on an i5 to be better than an i7 of equivalent clocks, memory config, etc. But how much better?

And is the delta worth worrying about? 103 fps on an i5 versus 101 fps on an i7 isn't likely to be worth contemplating. 45fps on an i5 vs 40fps on an i7 would be.

I'd recommend the OP sell his i7 920 NOW and wait until the i5's debut to then decide based on data whether to invest in an LGA1156 system or in an LGA1366 system.

(given that the OP is not currently using the i7, nor does he have the mobo, etc, for it yet)

Its not like the i7 920's are going to get more expensive between now and when the i5's debut, so having one sitting in your closet isn't the same as squirreling away gold or platinum bullion.

 

[plion]

thanks for the reply. I'll need to give it some more thought it seems

 

[ilkhan]

IDC: isn't it still a hop over QPI? Its just a physically shorter connection. Actually, how is that gonna work? The lynnfields shouldn't have the second die, but they'll still have the PCI-E...
plion: Unless you are running high end SLI (needing x16/x16 vs x8/x8, the i5 might not be a bad way to go. Single GTX285 is gonna run fine on a single x16 connection (i5 single x16 or x8/x8, i7 allows x16/x16.) I agree with IDC, unless you are planning on getting the board/etc now, sell the i7 until you are ready to build.

 

[VirtualLarry]

Core i7 920 is down to $200 at Microcenter, in-store only.

I would sell, before prices on these things drop any lower.

 

[taltamir]

well... yes the i5 would have lower latency due to direct interface PCIe for video card... but its limited to one 16x slot or two 8x... so a mobo getting two 16x using a northbridge has higher bandwidth with worse latency. combine this with the i7 having triple chanel ram and you have one little thing going for the i5 and many many things going for the i7.

For multi GPU (multiple slots that is) the i7 should faster, for single GPU the i7's advantages should compensate for its disadvantages and it should still be faster overall.

 

[Idontcare]

Originally posted by: ilkhan
IDC: isn't it still a hop over QPI? Its just a physically shorter connection. Actually, how is that gonna work? The lynnfields shouldn't have the second die, but they'll still have the PCI-E...
plion: Unless you are running high end SLI (needing x16/x16 vs x8/x8, the i5 might not be a bad way to go. Single GTX285 is gonna run fine on a single x16 connection (i5 single x16 or x8/x8, i7 allows x16/x16.) I agree with IDC, unless you are planning on getting the board/etc now, sell the i7 until you are ready to build.

Oh yeah, you just jogged my memory, we had this conversation before already...that even though PCIe is brought "under the IHS" it would still require a communication interface between the chip and the PCIe logic and we reasoned it would make no sense for Intel to create an entirely new intrachip protocol for this so they are likely to still use QPI to communicate with the PCIe even in i5.

:thumbsup: My memory is going down hill quick!

 

[ilkhan]

thats what I recall as well. Did we ever reason out if that was for clarkdale/arrandale only, or did the QPI->PCI-E scheme work for the lynnfields too? There's no separate die for lynnfield, of course.

 

[IntelUser2000]

Originally posted by: ilkhan
thats what I recall as well. Did we ever reason out if that was for clarkdale/arrandale only, or did the QPI->PCI-E scheme work for the lynnfields too? There's no separate die for lynnfield, of course.

Here's the wafer for Lynnfield: http://img257.imageshack.us/my...age=lynnfieldwafer.jpg

and Bloomfield: http://img521.imageshack.us/my...ge=bloomfieldwafer.jpg

Do you notice a difference? On the shot for Lynnfield, there's something between L3/Core/IMC and the QPI connection, that isn't there in the Bloomfield shot. If you aren't sure about which section represents what, look here for a reference shot of Bloomfield: http://chip-architect.com/news/Shanghai_Nehalem.jpg

I bet that's the PCI Express controller.

I'd think that's conclusive enough evidence that PCI Express controller is integrated into Lynnfield.

To the OP:

Here's a good comparison of low and high res gaming performance. The problem with i7 wasn't merely that it wasn't much faster than the Core 2 Quads, but rather being even slower in some cases.

Low Res Gaming: http://www.techspot.com/review...20-940-965/page12.html

High Res Gaming: http://www.techspot.com/review...20-940-965/page11.html

There's something up with i7 940 and 920 with UT3 and Far Cry 2. While they manage to be ahead of every non-i7 CPUs in the low resolution tests, they fall behind and even end up in the last place with high resolution tests!

We can rule out memory bandwidth being a problem, so its unlikely dual channels on Lynnfield makes it slower : http://www.techspot.com/articl...erformance/page14.html

What's different with Core i7 920/940 and the i7 965EE?? On the non-EE models, the Uncore(QPI/IMC) clocks at 2.13GHz while on the 965EE the Uncore clocks at 2.67GHz. Since we ruled out the possibility of the memory performance being the problem, it might have to do with having an extra hop.

The difference is probably enough to make it at least not slower than the Core 2 Quads in single GPU, high detail configs, which current Bloomfields sometimes fail to do.

 

[phillyman36]

What if you arent really a gamer? Would i7 be a better choice for video editing than a i5?

 

[Idontcare]

Originally posted by: IntelUser2000

Originally posted by: ilkhan
thats what I recall as well. Did we ever reason out if that was for clarkdale/arrandale only, or did the QPI->PCI-E scheme work for the lynnfields too? There's no separate die for lynnfield, of course.

Here's the wafer for Lynnfield: http://img257.imageshack.us/my...age=lynnfieldwafer.jpg

and Bloomfield: http://img521.imageshack.us/my...ge=bloomfieldwafer.jpg

Do you notice a difference? On the shot for Lynnfield, there's something between L3/Core/IMC and the QPI connection, that isn't there in the Bloomfield shot. If you aren't sure about which section represents what, look here for a reference shot of Bloomfield: http://chip-architect.com/news/Shanghai_Nehalem.jpg

I bet that's the PCI Express controller.

I'd think that's conclusive enough evidence that PCI Express controller is integrated into Lynnfield.

If it hasn't been said before, let me be the first to say "Damn, you're good!" :thumbsup:

Awesome post, value/word density is very high.

Here I rotated, cropped, and scaled the Lynnfield chip to that of Hans' annotated Nehalam shot:

http://i272.photobucket.com/al...ldversusBloomfield.jpg

As you say, the differences are self-evident and conclusive. Good find!

 

[ilkhan]

Originally posted by: Idontcare

Originally posted by: IntelUser2000

Originally posted by: ilkhan
thats what I recall as well. Did we ever reason out if that was for clarkdale/arrandale only, or did the QPI->PCI-E scheme work for the lynnfields too? There's no separate die for lynnfield, of course.

Here's the wafer for Lynnfield: http://img257.imageshack.us/my...age=lynnfieldwafer.jpg

and Bloomfield: http://img521.imageshack.us/my...ge=bloomfieldwafer.jpg

Do you notice a difference? On the shot for Lynnfield, there's something between L3/Core/IMC and the QPI connection, that isn't there in the Bloomfield shot. If you aren't sure about which section represents what, look here for a reference shot of Bloomfield: http://chip-architect.com/news/Shanghai_Nehalem.jpg

I bet that's the PCI Express controller.

I'd think that's conclusive enough evidence that PCI Express controller is integrated into Lynnfield.

If it hasn't been said before, let me be the first to say "Damn, you're good!" :thumbsup:

Awesome post, value/word density is very high.

Here I rotated, cropped, and scaled the Lynnfield chip to that of Hans' annotated Nehalam shot:

http://i272.photobucket.com/al...ldversusBloomfield.jpg

As you say, the differences are self-evident and conclusive. Good find!

very good find. Useful info, both of you.
Does the section labeled "Bridge to 2nd Die?" look like another QPI pair to you guys?

Originally posted by: phillyman36
What if you arent really a gamer? Would i7 be a better choice for video editing than a i5?

i7 (s1366) will be getting a 6 core gulftown option. Nor will the extra memory bandwidth hurt. So definitely. We're still not sure it'll be drop-in compatible, but it'll be s1366.

 

[ihyagp]

I'm curious about how PCIe latency affects framerates too. Anyone here have a board with a single PCIe x16 on the northbridge, and another on the southbridge, and is willing to do benchmarks comparing a (modern) GPU in one vs the other? That would shed some light on this. Obviously the difference in latency won't be as big as moving it fron the northbridge to the CPU die, but its the closest thing we have now.

 

[Idontcare]

Originally posted by: ihyagp
I'm curious about how PCIe latency affects framerates too. Anyone here have a board with a single PCIe x16 on the northbridge, and another on the southbridge, and is willing to do benchmarks comparing a (modern) GPU in one vs the other? That would shed some light on this. Obviously the difference in latency won't be as big as moving it fron the northbridge to the CPU die, but its the closest thing we have now.

Would the comparisons between SLI rigs that have NF200 versus one that do not have NF200 for SLI provide equivalent insight into the order-of-magnitude impact of adding latency/hops to the PCIe topology?

3-way SLI & FarCry 2: X58 vs. X58 + NF200

NF200 "True" 3-Way SLI Preliminary Results

If we take that to be an example of "how bad can we make it?" then it should give us some insight into "how much room for improvement is there when we try and make it even better than current x58?"

The answer would appear to be "a few fps" or a couple percentage points.

 

[ihyagp]

Not sure. Seems like a single gpu would be the best case for testing this. In fact I'm pretty sure the interconnect between X58 and NF200 is just PCIe x16, so there really isn't any added bandwidth here.

Originally posted by: Idontcare

Originally posted by: ihyagp
I'm curious about how PCIe latency affects framerates too. Anyone here have a board with a single PCIe x16 on the northbridge, and another on the southbridge, and is willing to do benchmarks comparing a (modern) GPU in one vs the other? That would shed some light on this. Obviously the difference in latency won't be as big as moving it fron the northbridge to the CPU die, but its the closest thing we have now.

Would the comparisons between SLI rigs that have NF200 versus one that do not have NF200 for SLI provide equivalent insight into the order-of-magnitude impact of adding latency/hops to the PCIe topology?

3-way SLI & FarCry 2: X58 vs. X58 + NF200

NF200 "True" 3-Way SLI Preliminary Results

If we take that to be an example of "how bad can we make it?" then it should give us some insight into "how much room for improvement is there when we try and make it even better than current x58?"

The answer would appear to be "a few fps" or a couple percentage points.

 

[Denithor]

Originally posted by: IntelUser2000
Here's a good comparison of low and high res gaming performance. The problem with i7 wasn't merely that it wasn't much faster than the Core 2 Quads, but rather being even slower in some cases.

Take a look at these two reviews: SingleGPU on i7 and MultiGPU on i7. Two things to note: i7 does not provide much (if anything) to single card setups versus previous quads (or even duals, in many cases) but the results with multiGPU setups is dramatic, to say the least. From previous discussion we've held here this boost is generally attributed to the much more efficient distribution of data among the cores (CPU and GPU) thanks to the QPI.

So the i5 with an onboard PCIe controller (which shortens this pathway even further) is actually highly interesting from a gaming viewpoint. As long as the dual x8 lanes don't constrict the GPUs i5 may actually beat i7.

Originally posted by: ihyagp
Not sure. Seems like a single gpu would be the best case for testing this. In fact I'm pretty sure the interconnect between X58 and NF200 is just PCIe x16, so there really isn't any added bandwidth here.

In fact, it looks to me like the NF200 chip simply adds overhead, restricting bandwidth slightly and hurting performance.

So much for "True SLI" - nV should stick with making GPUs and leave chipsets to the chip producers.

 

[Idontcare]

Originally posted by: Denithor

Originally posted by: ihyagp
Not sure. Seems like a single gpu would be the best case for testing this. In fact I'm pretty sure the interconnect between X58 and NF200 is just PCIe x16, so there really isn't any added bandwidth here.

In fact, it looks to me like the NF200 chip simply adds overhead, restricting bandwidth slightly and hurting performance.

So much for "True SLI" - nV should stick with making GPUs and leave chipsets to the chip producers.

My point was simply that we have an example where we can see the relative magnitude of adding more latency to the PCIe topology from CPU to GPU...ala NF200...and the impact isn't all that significant.

This suggests that going the other direction and removing a plausibly equal or less amount of latency is not likely to have a postive influence beyond 2-3% either. We aren't talking about things that double the latency or cut it in half, we are down into the "give or take 2-3 fps" territory.

While that's not likely to then become the underpinnings of next-gen GPU technology, it might be advantageous for other metrics of relevance such as GPGPU processes or some such.

And while i7 did not bring an improvement in the IPC for gaming over the IPC delivered by Penryn architecture in gaming, it did bring substantial IPC/watt (or more to the point, fps/watt) improvements over penryn.

Originally posted by: Idontcare
Thanks, I appreciate the sanity check, so I'm not entirely off-base on this I guess.

Still though I like the new power numbers that Anand published. It's actually quite a nice showing for Nehalem.

Not sure why nobody actually crunches the data into performance/watt metrics anymore, guess its not sexy enough anymore. It's so 2007.

I went ahead and crunched Anand's data to convert it to performance/watt:

CPU...................................QX9770 (3.2GHz)..........Core i7-965 (3.2GHz).............Improvement
POV-Ray..............................11.4 PPS/Watt..............17.5 PPS/Watt......................53%
Cinebench (1 thread)............20.3 CBMarks/Watt.......26.6 CBMarks/Watt...............31%
Cinebench (max threads)......61.8 CBMarks/Watt.......81.5 CBMarks/Watt...............32%
3dsmax 9 SPECapc CPU........0.060 /Watt..................0.084 /Watt..........................41%
x264 HD Encode Test............0.32 fps/Watt................0.44 fps/Watt.......................38%
DivX 6.8.3............................2.61 Watts...................1.84 Watts............................29%
Windows Media Encoder........2.01 Watts....................1.34 Watts............................33%
Age of Conan.......................0.35 fps/Watt................0.46 fps/Watt........................31%
Race Driver GRID.................0.30 fps/Watt...............0.34 fps/Watt........................15%
Crysis..................................0.14 fps/Watt...............0.16 fps/Watt........................15%
FarCry 2..............................0.32 fps/Watt................0.42 fps/Watt........................34%
Fallout 3...............................0.25 fps/Watt...............0.37 fps/Watt........................45%

Unless I made a mistake in the math the i7 beat the QX9770 in every test. The average percent power consumption reduction per unit of work being done is 33% for the i7 over yorkfield.

Now I am finally seeing the 30-40% power consumption reduction numbers I was expecting once performance is normalized Me much happier now!

 

[cbn]

Are the first Core i5s going to be 32nm? Or do those come later?

 

[Idontcare]

Originally posted by: Just learning
Are the first Core i5s going to be 32nm? Or do those come later?

45nm first. Then shortly thereafter (months) the 32nm ones will.

The difference, well there are many, but the zeroth order difference will be that the 45nm ones are quadcores whereas the 32nm ones will be dual-cores.

http://www.anandtech.com/cpuch...howdoc.aspx?i=3513&p=4

For reasons not fully explained (or at least not fully understood), a resident board expert here and industry insider (Aigo) is violently opposed to the proposed reality communicated in Anand's article I linked above. Just full disclaimer to say not everyone agrees to the timeline, so you may get alternate opinions on this if you ask around.