backtop


Print 36 comment(s) - last by The Sword 88.. on Apr 19 at 9:30 PM

Intel benchmarks its dual and quad-core "Penryn" processors

Intel is set to introduce its Penryn-family later this year beginning with its Xeon processor family. Penryn is a die shrink of the current Conroe-family to 45nm. Intel previously confirmed Penryn taped out in January and fully capable of booting Windows Vista, XP, Linux and Mac OS X operating systems.

In addition to the shrunken fabrication process, Penryn features new SSE instructions and more L2 cache. New to Penryn are SSE4 instructions for more efficient executions of SSE, SSE2 and SSE3 instructions and new enhancements for image, video processing and vectorization.

Intel plans to upgrade the cache on dual and quad-core Penryn-family products. Dual-core Yorkfield processors gain an additional 2MB of L2 cache for a total of 6MB. Quad-core Yorkfield processors, which are essentially two Wolfdales sandwiched together, feature 12MB of L2 cache. Intel’s current Kentsfield quad-core processor only has 8MB of L2 cache.

Intel has designed its Penryn-family to operate in speeds excess of 3.0 GHz. The current dual and quad-core samples demonstrated at Spring IDF in Beijing, China operate at 3.33 GHz. These early Penryn samples also operate on a 1333 MHz front-side bus.

The early Penryn dual and quad-core processors are operational as well. Intel has benchmarked its dual and quad-core in various applications and games to salivate the public. The Intel test system consists of its D975XBX2 BadAxe 2 motherboard with 2 GB of memory in dual-channel and a GeForce 8800 GTX running Microsoft’s Windows Vista Ultimate 32-bit. Intel installs a pre-production dual-core and quad-core Penryn processors clocked at 3.33 GHz with 1333 MHz front-side buses in its test system. Intel also benchmarks a Core 2 Extreme QX6800 for reference.

Applications
Model
Dual-core
Penryn
Quad-core
Penryn
C2E
QX6800
Mainconcept H.264 Encoder
(seconds, lower is better)
119
73
89
DivX 6.6 Alpha with VirtualDub 1.7.1
(seconds, lower is better)
22
18 38
Cinebench R9.5 (CPU benchmark)
(higher is better)
1134
1935
1549
Cinebench R10 beta (CPU benchmark)
(higher is better)
7045   13068  10416

Video encoding shows an 18 percent improvement with the new Penryn-family when it comes to H.264 encoding. DivX encoding shows a 52 percent improvement with the quad-core Penryn over the previous Core 2 Extreme QX6800. The dual-core Penryn processor is able to beat out the quad-core Core 2 Extreme QX6800 by 42 percent when it comes to DivX encoding. Cinebench R9.5 and R10 beta reveals performance increases of 25 percent when comparing the quad-core Penryn and Core 2 Extreme QX6800 processors.

3D Applications
Model
Dual-core
Penryn
Quad-core
Penryn
C2E
QX6800
3DMark06 v1.1.0 Pro - CPU
(higher is better)
3061
4957
4070
3DMark06 v1.1.0 Pro - Overall
(higher is better)
11015
11963
11123
Half Life 2: Lost Coast
(frames per second, higher is better)
210
210
153

Half Life 2: Lost Coast shows gains of 37 percent between the Penryn and Core 2 Extreme QX6800. The dual-core Penryn has no troubles keeping up with its quad-core counterpart. Synthetic 3D benchmarks such as 3DMark06’s CPU benchmark shows 22 percent gains with Penryn.

Expect Intel to release Penryn later this year beginning with its Xeon products, desktop Core 2 products should follow shortly.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

3.3GHz vs 2.9GHz
By psychobriggsy on 4/18/2007 8:25:07 AM , Rating: 4
I hate it when benchmarks are given with completely different clock speed CPUs and presented as if they're clock-for-clock improvements (i.e., they don't mention the clock speed difference is part of the reason for the faster performance).

3333MHz is nearly 14% faster than 2933MHz, which is the QX6800. The latter is a chip that many people claim could easily be sold at 3333MHz today if Intel desired to.

It seems that the main improvements are primarily coming from two things: SSE4 (significant for the video encoding improvements), and the faster FSB for bandwidth reliant apps like Cinebench (as QX6800 is quite crippled here by its 1066MHz FSB, which is 266MHz per core, consider a single core P4 was bottlenecked at 2GHz with a 400MHz bus, and here we have a core that's probably equivalent to a 5GHz P4 albeit with larger caches and a better prefetcher). It'd be nice to see how Penryn compares with an overclocked QX6800 at 3333MHz when the review is done.

Not putting down Penryn on course, it looks like it is going to be an awesome chip and put a lot of pressure on AMD to ramp up Barcelona clock speeds or try and sell them cheaper than they'd like. It seems that Intel continue to perform excellently. When Intel integrates the memory controllers in Nehalem AMD will be in for yet another big world of hurt.




RE: 3.3GHz vs 2.9GHz
By Moishe on 4/18/2007 8:50:24 AM , Rating: 1
yeah... which is why we will truly see the performance when someone makes a proper evaluation of the chip and compares it to a comparable AMD chip. This is interesting but it really doesn't say much.


RE: 3.3GHz vs 2.9GHz
By Mitch101 on 4/18/2007 9:13:30 AM , Rating: 1
I agree this doesnt look much different than an overclocked chip with some SSE4 enhanced benchmarks.

I would like to see the benchmarks copare the chips at the same clock speed to determine what true perfomance increases there are architechturally.

In the end this seems like a publicity stunt to take eyes off people thinking about Barcelona.


RE: 3.3GHz vs 2.9GHz
By Mitch101 on 4/18/2007 10:39:50 AM , Rating: 1
Just read anantech's short info and sadly Wolfdale compared to Core 2 conroe are about the same.

If you add 10% overclock to conroe and add 10% in the benchmarks they are about identical. I know the one benchmarks shows 19% but if you overclocked a Conroe via the FSB I would bet those numbers are a lot closer than 19% difference as well as that was only at a lower resolution the upper resolution showed 10.6%.

I have a feeling Penryn might not be enough to beat Barcelona. It may come down to who's chips overclock the best and Intel at 45nm should have an advantage over AMD at 65nm. Then it may come down to bang for the buck.


RE: 3.3GHz vs 2.9GHz
By deeznuts on 4/18/2007 12:53:49 PM , Rating: 1
Wait so add the 10% overclock and then add in the advantage of the faster cpu (here 10%)? With that type of formula every cpu ever made will be identical to any other cpu ever made in the history of CPUs.

Of course they used the lower resolution, they need to isolate the cpu from the effects of the video card.

Dailytech, always good for a daily laugh reading the comments.


RE: 3.3GHz vs 2.9GHz
By SmokeRngs on 4/19/2007 3:07:34 PM , Rating: 2
So, a dual core P4 running at 3.33Ghz runs everything with the same performance as a Conroe at 3.33Ghz and a Penryn at 3.33Ghz. Not to mention an Athlon X2 at 3.33Ghz has the same performance.

I am guessing you do not know anything at all about architecture differences since you're saying every CPU every made would perform the same if it was running the same clock speed.

The point about the "test" CPUs is valid. The older CPU is running a slower clock speed so it's very likely it will not have the same performance since there are not major architecture changes being made. It's a die shrink and a refresh. To have a valid comparison you would need both CPUs to be running similar clock speeds. I don't remember if you can use half multipliers for Intel CPUs but if so, all you would need to do is raise the multiplier on the older CPU to 12.5 to match clock speeds.

From the results of the benchmarks, it looks like Penryn isn't going to bring a whole lot in performance improvements over Conroe at this time except for certain applications. Then again, this isn't a major architecture change and any increase in efficiency is welcomed.

Just remember, these numbers are from Intel and the normal cherrypicking of benchmarks is certain to have happened here. To me, this just says that Penryn from a stock speed performance standpoint won't be very enticing to me right now. There should be other advantages such as lower power but I usually don't worry about that too much as long as my current CPU isn't a furnace. Then again, I'm looking forward to the Intel quad cores in the third quarter if Intel drops the prices to what it's speculated they are going to be. Perfect drop-in replacement to my C2D.


RE: 3.3GHz vs 2.9GHz
By Oregonian2 on 4/18/2007 1:33:45 PM , Rating: 2
Benchmarks at different clock speeds is fine. Would make no sense whatsoever, for instance, to only compare Conroe and Pentium IV parts that are rated the same clock speed. Just as one didn't compare AMD vs Pentium chips based on the same clock speed.

More important is to compare at the same price-point. Harder to do prior to release, but really what is important, I think.

An alternative good comparison might be at the same power-point (if one can somehow equate the different ways that is spec'd).

A third point might be 'the fastest available' to see where the top end is.

But clock-for-clock throughput is mostly interesting only for the designers of the chips I think, those talking about pipeline strategy tradeoffs and the like (closest I get to that nowadays would be in FPGA designs).


RE: 3.3GHz vs 2.9GHz
By caqde on 4/18/2007 3:26:01 PM , Rating: 2
The idea behind the same clock speeds is to find out what the architectural changes actually do. Otherwise you can't really tell which architecture is better. Each type of benchmark has it's own merits the one your talking about is good for released products. But in this case it would have been nice if Intel put a 2.93Ghz Penryn in the mix for comparison.


RE: 3.3GHz vs 2.9GHz
By Ringold on 4/18/2007 8:58:21 PM , Rating: 2
I also appreciate clock-for-clock comparisons because, for an overclocker, it provides a rough mental guide for expected performance gains when looking at all kinds of different processors one can buy at various price points and how people have managed to OC them. When people report OC's and get it added to OC databases they don't typically run a full suite of benchmarks for everyones benefit, so having clock-for-clock comparison information in the back of ones mind can just be of some use.

Of course, it's no replacement for any other benchmark, and I agree the best general comparison is chip at a price point rather than at a clock speed. Clock-for-clock is a compliment.


RE: 3.3GHz vs 2.9GHz
By The Sword 88 on 4/19/2007 9:30:45 PM , Rating: 2
I thought it was die shrink not an architectural change


RE: 3.3GHz vs 2.9GHz
By spartan014 on 4/18/2007 11:42:25 PM , Rating: 2
To know the exact performance benefit, they should have taken a Conroe overclocked to 3.33GHz and 1333 MHz FSB for comparison. Then we could see what improvement does the 45nm die shrink and additional 2 MB cache brings...


RE: 3.3GHz vs 2.9GHz
By SmokeRngs on 4/19/2007 3:12:15 PM , Rating: 2
Actually, it would be better to keep the Conroe at 1066 FSB. That's what it is at stock and it should stay that way for the comparisons. You can also add in Conroe at 1333 bus for an extra comparison but I would like to see the differences at stock FSB but equal clock speeds. The Conroe at 1333 FSB would be to determine the differences the increased FSB makes.

Then again, my Conroe is running 1600FSB already so it doesn't matter to me. Too bad I can't do comparisons. I only have an E6400 and with the stock cooling can't get it stable over 3.2.


can't wait
By Moishe on 4/18/2007 8:12:06 AM , Rating: 2
Can't wait to see the real/thorough benchmark numbers from Anandtech... looks like these Penryn cores will be very fast.

Based on the fact that the 2core was the same as the 4 core in the HL2 benchmark it looks like maybe the CPU was being held back by something else. That's cool considering lately people complain about the high end video cards waiting on the CPU.




RE: can't wait
By Bladen on 4/18/2007 8:19:37 AM , Rating: 2
I think the idea of these tests with Penryn was to try to steal the limelight from Barcelona.

Otherwise, why wouldn't they do a clock to clock comparison if not but to amplify the advantages?

Still, though, any positive development is good. So is competition though, so I hope Barcelona is somewhere within 5-10% of Penryn clock for clock.


RE: can't wait
By Proteusza on 4/18/2007 8:24:49 AM , Rating: 2
quote:
Based on the fact that the 2core was the same as the 4 core in the HL2 benchmark it looks like maybe the CPU was being held back by something else. That's cool considering lately people complain about the high end video cards waiting on the CPU.


It might also be possible that the source engine doesnt take advantage of more than 2 cores. Come to think of it, I dont know if Source is even multi threaded.


RE: can't wait
By AnnihilatorX on 4/18/2007 2:09:33 PM , Rating: 2
quote:
Based on the fact that the 2core was the same as the 4 core in the HL2 benchmark it looks like maybe the CPU was being held back by something else


I don't think HL2 is optimized for quadcore. May be dual, but definitely not quad.
The only game I know that do optimize for quads is Supreme Commander.


codenames?
By webdawg77 on 4/18/2007 10:15:15 AM , Rating: 2
I thought Wolfdale was the dual-core, and Yorkfield was the quad-core?




RE: codenames?
By coldpower27 on 4/18/2007 11:56:10 AM , Rating: 2
They are but DailyTech is speaking about the family as a whole. The "Penryn Family" which includes all 45nm based Core 2's.


RE: codenames?
By webdawg77 on 4/18/2007 2:10:25 PM , Rating: 2
quote:
Dual-core Yorkfield processors gain an additional 2MB of L2 cache for a total of 6MB. Quad-core Wolfdale processors, which are essentially two Yorkfield’s sandwiched together, feature 12MB of L2 cache.


RE: codenames?
By TechLuster on 4/19/2007 1:04:03 AM , Rating: 2
quote:
I thought Wolfdale was the dual-core, and Yorkfield was the quad-core?

You're correct--DT got it wrong. Wolfdale is in fact the dual-core Conroe successor, while Yorkfield's the 45nm Kentsfield replacement.


AMD Barcelona
By DannyH246 on 4/18/2007 8:27:08 AM , Rating: 2
I am a self confessed AMD Fanboy, but hats off to Intel they manage to show benchmarks for unreleased/upcoming products, what on earth are AMD doing?

Where the F* is Barcelona???????




RE: AMD Barcelona
By Chadder007 on 4/18/2007 9:13:28 AM , Rating: 2
AMDs response to this may simply be.... "Aww crap" but I hope they are hiding some other secret improvements to their processors. Intel just showed their cards, now its time for AMD to do the same.


RE: AMD Barcelona
By ObscureCaucasian on 4/18/2007 9:35:43 AM , Rating: 2
Hopefully we'll hear something from AMD before they launch the R600 cards.


Overclockability?
By JamRockaz on 4/18/2007 10:24:32 AM , Rating: 2
I would love to see how far these new penryn overclocks. I bet they can easily touch the 5Ghz mark on air alone. I wouldnt be suprised if they reach 8 or 9Ghz on water.




RE: Overclockability?
By Mitch101 on 4/18/2007 10:47:45 AM , Rating: 2
I think it will top out around 4.0-4.2ghz on air.

Overclocking Conroe you mainly overclocked the FSB with these the FSB is already set higher possibly leaving less room for overclocking. Getting good ram to overclock it much further will be a premium. Certainly 45nm will help but your overclock is only as good as its weakest point. Motherboards will be an overclock issue and I believe you might need a new chipset (bearlake?) to get any real high overclocks with the new chips.


RE: Overclockability?
By Scimitar on 4/18/2007 12:15:06 PM , Rating: 2
9 GHz on water? ROFLMAO. Do they walk on water too?

Most Conroe X6800s do about 3.5GHz to 4.1 GHz on air and about the same or very slightly higher on water. Since these 45nm chips are starting at 400 MHz higher I would guess 4GHz to 4.5+ GHz on air is a fair guess. Probably 6GHz+ on LN2.


correct numbers?
By webdawg77 on 4/18/2007 8:26:53 AM , Rating: 1
Model
Dual-core Penryn / C2E QX6800
Mainconcept H.264 Encoder(seconds, lower is better)
119 / 89

Cinebench R9.5 (CPU benchmark)(higher is better)
1134 / 1549

Cinebench R10 beta (CPU benchmark)(higher is better)
7045 / 10416

3DMark06 v1.1.0 Pro - CPU (higher is better)
3061 / 4070

3DMark06 v1.1.0 Pro - Overall(higher is better)
11015 / 11123

Is it just me, or does the new dual core Penryn acutally under-perform the current C2E? The Penryn has a higher clock 3.33GHz vs 2.93GHz, a higher FSB 1333MHz vs 1066 MHz, and more cache 6MB L2 vs 4MB L2, not to mention the smaller die process 45nm vs 65nm.

Shouldn't these numbers be higher or at least roughly the same?




RE: correct numbers?
By Bladen on 4/18/2007 8:38:13 AM , Rating: 2
The Core 2 Extreme is of a quad core variety, the Penryn on the left is of a dual core variety.

The tests highlighted are assumably fairly multi-threaded.


RE: correct numbers?
By webdawg77 on 4/18/2007 8:39:49 AM , Rating: 2
You are correct. I didn't even noticed the "Q" in the model number. My bad :).


Scaling
By psychobriggsy on 4/18/2007 9:41:51 AM , Rating: 2
It's interesting to see the scaling from two cores to four.

For instance, in video encoding, a task which I presume was running on all 4 cores, we get a 1.63x speedup and a 1.22x speedup. These are not good figures, even ignoring the second result as being poorly optimised for quad-core.

The Cinebench results scale far better, suggesting that the caches are helping greatly for Penryn. 1.71x and 1.86x for v9 and v10 respectively. I guess that's where 12MB of L2 comes in handy. Can someone verify whether the quad-core has a bus between the two cores to offload the FSB coherency traffic?

And finally 3DMark. The CPU test scales by 1.62x, not that great, but it's 3DMark. The most pertinent for gamers is the 1.09x scaling from dual-core to quad-core. If you're a single tasking gamer, then quad-core Penryn seems rather pointless - until game engines start supporting quad-cores efficiently. Let's not talk about HL2.




RE: Scaling
By sdsdv10 on 4/18/2007 10:10:39 AM , Rating: 2
quote:
If you're a single tasking gamer, then quad-core Penryn seems rather pointless - until game engines start supporting quad-cores efficiently. Let's not talk about HL2.


How about Supreme Commander then?
http://forums.anandtech.com/messageview.aspx?catid...


Oh yuck
By rcc on 4/18/2007 11:55:06 AM , Rating: 3
quote:
Intel has benchmarked its dual and quad-core in various applications and games to salivate the public.


C'mon guys. It's ok to make the the public salivate. It's not ok to drool on them.




.
By Hyperlite on 4/18/2007 8:06:26 AM , Rating: 2
O.O

Looks good!




Shadest preformance preview I've ever seen
By Daven on 4/18/07, Rating: 0
By Andypro on 4/18/2007 12:46:07 PM , Rating: 2
The Penryn family was never intended to be much more than a simple die shrink of conroe. It has been in Intel's timeline for well over a year now:

2k6: New arch. (Conroe)
2k7: Die shrink of last year's arch. (Penryn)
2k8: New arch (Nehalem)
2k9: Die shrink of last year's arch..
etc.

Their testbed seemed very fair to me. Very similar to the preliminary Conroe testbed early last year. They aren't trying to deceive people or preempt AMD as far as showcasing their product imo. It's their IDF and they have these chips ready, so why not. If they wanted to overinflate their improvement % numbers they would have added an AMD chip into their testbed for comparison. But they didn't.

I'm actually quite impressed that they did add a couple of architectural improvements like Radix16 divider and SSE4 instructions. As for the "true clock for clock" comparison, I'm all for that to see architectural improvements, but since there are very few in this case I think it's fine for them to use the speeds at which they'll be selling these chips. After all, they have the headroom to clock them however they like. This is a tangible benefit of the Core architecture over AMD's line. Will Barcelona change this? Perhaps. But we won't know until they decide to tell us.

In the mean time, I'm going to be checking forums to see if Penryn engineering samples start showing up soon.


Performance
By Setsunayaki on 4/19/2007 12:52:16 AM , Rating: 2
This really doesn't matter much for the following reasons

1) I am currently running an overclocked e6300 that has reached 3.5ghz on air cooling. I keep it at 2.4ghz on a multiplier of 6 because it can run on 1.068v and that is important for heat. Secondly. Once I optimized windows, I scored higher in CPUMark2 than the X6800. This means I can run a graphic card to its potential power at 400 x 6. My PC is overclocked and its stable and made with parts to be a silent PC to the point I got scared thinking the PC was broken when I turned it on because I didn't hear a wind tunnel powering up like I was used to hearing. Sure, I can run on 2.8ghz, 3.0ghz and 3.5ghz....but that doesn't really do much for me unless a processor requires such speed in order to unlock the maximum power of a graphic card.

I already have 400 - 500mhz FSB....why do I need 333mhz BUS? If my processor fries....then I will buy a 6320 that has 2MB extra Cache to match the higher end processors and I can get 2 - 5% performance in gaming due to the cache. I could also buy a Quad core...or even a Penryn...

2) Most motherboards in existence that allow for 1333mhz FSB force you to overclock to reach that speed. This means a real motherboard being released that is 1333mhz stock will cost more. The processor will cost more as well since the Conroes are floating around as well. Finally....In order for true 1 : 1 synchronization to be reached, we need 1333mhz DDR-2 meaning the latencies would be shot.

3) This benchmark is flawed because of the different BUS speeds. If they truly wanted to make a REAL BENCHMARK. They could have launched the Penryn at 266 x 10 for 2.66ghz and take the X6800, Drop its multiplier by 1, for 266 x 10. Then they could have benchmarked both processors and we could have seen a real difference.

This is a test of Process type....65nm vs 45nm. What we may see is lower heat as long as the Voltages can be reduced. If you have a 45nm process, and the voltage can not be reduced, its not much for heat reduction.I guarantee you however that we won't see higher performance, that in fact performance will be the same under the SAME BUS. I do however bet the the voltage can be lower and that it has less heat and uses less power.

Why do I need a FSB of 333mhz as tested above...if I already can reach 400mhz (1600mhz) along with 500mhz (2ghz)?

Right now it costs $400 to get performance higher than stock penryn and x6800 conroe speeds. That is how much I paid for my MBDs, RAM and Processor....then 150 more for cooling and case + power supply. 100 for the HDD and Optical drive...I already owned the graphic card (X850XT PE....which I can overclock it from 540/590 to 625/625) and gives me max framerate in Command and Conquer 3 and I can really play in Quake 4. Only drawback is it doesnt have HDR for Oblivion....

Think about what I have said....We are all waiting for "performance" when in truth performance beyond what we are waiting for is already out there at a price that is cheaper and not 1st generation technology. ^_^ We just have to risk it and overclock and if we succeed, we don't have to wait. ^_^




"Paying an extra $500 for a computer in this environment -- same piece of hardware -- paying $500 more to get a logo on it? I think that's a more challenging proposition for the average person than it used to be." -- Steve Ballmer

Related Articles













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki