THG (P)reviews "Core 2 Quadro" - aka Kentsfield!

[MU_Engineer]

Well Jack, the FX-62 is a dual-core CPU at 2.8 GHz and the Kentsfield is a quad-core CPU at 3.33 GHz. Also, each core has the same number of FPUs, so you have twice the amount of FPUs on the Kentsfield than the K8. I know what you're trying to get across (that the integer performance of the Core 2 arch is good) but when one takes into context the differences in the setup, the results are less outrageous, but still impressive, mostly due to the 4-issue core and the single-cycle FPU than anything else.

The Kentsfield at 3.33 GHz is 20% faster than the FX-62 in clock speed. The Kentsfield is a 4-issue chip and the FX-62 is a 3-issue chip. So you would expect the spread to be (3.33/2.80) * (4/3) * 2 = 317%. The Core 2 is roughly 10% more efficient clock-for-clock than the K8, so that brings the integer performance up to 349%. The integer unit in the Kentsfield takes a single cycle to do 128-bit calculations and the K8 takes two as its 80-bit unit splits 128-bit numbers up into 2 64-bit parts to take 2 cycles IIRC. So voila- there's your 700%. BUT that 700% speedup will likely never get seen "in the wild" as getting a consistent 4 IPC will be pretty much impossible for the CPU- 2-2.5 is as good as it gets from what I have heard. So that 4-issue core giving a 33% boost will be negated for the most part.

 

[Dante_Jose_Cuervo]

Wow... I'm really impressed... but I'll wait for the true quad-cores for the DP systems and then upgrade from my woodcrests.

 

[Artmic]

Hmm, those 4 X 4 sound more like some type of lumber product.

What is AMD thinking with that $ x $ crap? What is stopping somone from getting a Tyan board with dual sockets and putting 2 dual core cpus on them?
They will still be slower than a woodcrest system.

Oh well, too bad about the software support for quad cores, i doubt we will see games with quad core support anytime soon, maybe 18 months from now. It is one thing to do media encoding, and another to use 4 cores on a game.

 

[MU_Engineer]

I'll probably end up getting a pair of 8-core CPUs in 3 years. By then they should be at a good price- $1000 or so with the board. That should provide a nice jump over my X2 4200+ when doing compiles. MAKEOPTS="-j31" just screams fast...

 

[SuperFly03]

Wow... I'm really impressed... but I'll wait for the true quad-cores for the DP systems and then upgrade from my woodcrests.

I concur, I am very impressed with Kentsfield. It is nice to see a system limited by its graphics card instead of its CPU (like my system). I am still batting from the single core dugout so Kentsfield is looking like an excellent buy. Imagine that increase... going from a measely 2.75GHz Venice to a 3GHz+ Kentsfield, thats nuts.

I am heading towards peltier cooling and Kentsfield gives me damn good reason to go that way too. A TDP in excess of 100w, plus 3.33GHz w/o a voltage bump? Oh I definately see 4GHz on the horizon if I can get the temp down low enough. I may have to go insane and get a 407w Qmax peltier to cope with the 4 cores, but we shall see.

From what I see, anything having to do with video or audio encoding gets a huge bump from 4 cores. Oddly enough that may be the one thing I do least, maybe its because I can't ever get the freaking audio and video synchronized. I blame my current rig, although I don't acctually know the problem.

What I want to do is encode my current DVD's into Divx, XVid, or WMV HD (obviously 480P to match DVD quailty) so I can store them all on my home server and then just stream them to any of my TV's. Kentsfield may provide the raw horsepower that I need to encode a few hundred DVD's in a reasonable time frame. I don't even want to deal with HD DVD or BluRay and their insanely overpriced marketing schemes lol.

Point is, props to Intel for really putting AMD to shame. I have and AMD and still like them, but I have to say it: AMD you woke a sleeping giant and are now under the wrath of god (of the silicon industry that is ).

4x4 = Desperation from AMD (sorry to say it... I doubt it will do much for them)

 

[SuperFly03]

Hmm, those 4 X 4 sound more like some type of lumber product.

What is AMD thinking with that $ x $ crap? What is stopping somone from getting a Tyan board with dual sockets and putting 2 dual core cpus on them?
They will still be slower than a woodcrest system.

Oh well, too bad about the software support for quad cores, i doubt we will see games with quad core support anytime soon, maybe 18 months from now. It is one thing to do media encoding, and another to use 4 cores on a game.

Unless AMD has commissioined a gaming company to write some 'AMD optimized 4x4 code', Kentsfield is showing us roughly how gaming will interact in this environment. COD2 obviously was poorly coded for multi-core environments, it won't even run on anything >2 core. Any more umph that a midraged C2D or a high end AMD X2 and you simply run into GPU limitations, so I suspect 4x4 is not really going to amount to much in gaming. Even the quad SLI is buggier than heck, and only eaks out a few more FPS than a good high end SLI solution.

The cost on this will be prohibitive for now gain ---- Baron will eat it up though, he seems to be in a charitable mood.

Yeah I am looking forward to being GPU limited, its an easier fish to fry. Just pop out the old GPU and pop in a new one. CPU's usually come with a motherboard, RAM (depending on timign of the switch) and god knows what else. I am planning for a Vista rig because, honestly, I am looking forward to Halo 2 Vista damn it . I've been on both sides of the coin (CPU limited v. GPU limited) and I personaly think its easier being GPU limited.

Gotta love this "transitional era" into multicore architecture, what a freaking pain for programmers. Oh well... where is my quad-core.

Edit: 4x4 is a pathetic marketing attempt to drive dual opterons mainstream imo. Instead of focusing on gimics, AMD should bring somethign to market worth our money...

 

[MU_Engineer]

I am wondering why nobody put the single-cycle integer units in chips already. It did not seem too terribly difficult for Intel to do so. I guess AMD rested on its laurels with the K7 and K8s' 3 fully-pipelined FPUs beating the PIII's single FPU and NetBurst's 1 fully-pipelined unit and 1 partially-pipelined unit and stuck with the standard 80-bit integer unit like everybody else did at the time.

AMD said that the K8L will have single-cycle integer units and be a 4-issue core like the Kentsfield. It will be much wider on the inside than the K8 is and will be a native quad-core unlike Kentsfield. So it will be a very interesting 2007 for sure. Nobody is sure on what AMD is really up to and the ETA of their products as they have only told the press about their 65nm transition after they'd been shipping wafers for two weeks and the quad-core tapeout only after it had happened also. I guess they have to be vague to try to head Intel off at the pass. Too bad for us though 🙁 The closest we get to knowing what they might be up to is The Inq

 

[Artmic]

Hmm, those 4 X 4 sound more like some type of lumber product.

What is AMD thinking with that $ x $ crap? What is stopping somone from getting a Tyan board with dual sockets and putting 2 dual core cpus on them?
They will still be slower than a woodcrest system.

Oh well, too bad about the software support for quad cores, i doubt we will see games with quad core support anytime soon, maybe 18 months from now. It is one thing to do media encoding, and another to use 4 cores on a game.

Unless AMD has commissioined a gaming company to write some 'AMD optimized 4x4 code', Kentsfield is showing us roughly how gaming will interact in this environment. COD2 obviously was poorly coded for multi-core environments, it won't even run on anything >2 core. Any more umph that a midraged C2D or a high end AMD X2 and you simply run into GPU limitations, so I suspect 4x4 is not really going to amount to much in gaming. Even the quad SLI is buggier than heck, and only eaks out a few more FPS than a good high end SLI solution.

The cost on this will be prohibitive for now gain ---- Baron will eat it up though, he seems to be in a charitable mood.

Indeed, the Quad SLI stuff is just marketing crap IMO, i got the nVidia 7950 card, and don't plan any quad crap any time soon. lol ALthought i do think the 7950 was worth every penny so far. lol

I hope they release more dual-gpu based boards in the future, i really hate putting 2 separate vid cards into a computer.

 

[SuperFly03]

I hope they release more dual-gpu based boards in the future, i really hate putting 2 separate vid cards into a computer.

I am all for that statement. I've had 7900 GTX's in SLI and I've had one by itself... I can definately agree with you on that one. Even if it's only akin to 7900GT's in SLI, it is better to have that in one card then have two 7900 GT's seperate or two 7900 GTX's unless you plan to liquid/peltier cool them and go nuts with volt modding, but thats a very rare case.

I wish ATI would counter with something akin to the 7950... although they don't really need to performance wise, but you get the idea...

 

[ltcommander_data]

Unless AMD has commissioined a gaming company to write some 'AMD optimized 4x4 code', Kentsfield is showing us roughly how gaming will interact in this environment. COD2 obviously was poorly coded for multi-core environments, it won't even run on anything >2 core.

I have no doubt that Intel is going to be offering developers support to multithread. It was my understanding that Quake 4 and COD2 are as multithreaded as they are now because Intel offered direct support. (Intel just released support for better mobile game development, http://www.theinquirer.net/default.aspx?article=34207, which is nice). Obviously though most "multi"-threaded applications are actually barely "dual"-threaded. In general I think we've pretty saturated the performance benefit of more cores at 4 for the time being. Which is pretty quick considering they were just introducing dual core not long ago. I wonder how long it's going to take for developers to catch up? It's still questionable whether it's more beneficial to push multithreading first or to push 64-bit first.

Any opinions on whether Kentsfield is a 1067MHz or 1333MHz part from my earlier post?

 

[Artmic]

I hope they release more dual-gpu based boards in the future, i really hate putting 2 separate vid cards into a computer.

I am all for that statement. I've had 7900 GTX's in SLI and I've had one by itself... I can definately agree with you on that one. Even if it's only akin to 7900GT's in SLI, it is better to have that in one card then have two 7900 GT's seperate or two 7900 GTX's unless you plan to liquid/peltier cool them and go nuts with volt modding, but thats a very rare case.

I wish ATI would counter with something akin to the 7950... although they don't really need to performance wise, but you get the idea...

Indeed i was hoping ATI would get some dual GPU's out but nothing so far, i am very interested in the DX10 part from ATI, i might just get that if it blows the 7950 out of the water, even in single gpu mode lol

 

[sepheronx]

this is funny, I remember before C2D came out, everyone here that was AMD biased said "oh, AMD will win, C2D is just paperwork and blah blah blah. I wonder what ever happened to MadModMike?

lol, now Quadcore showing its better, and people say "oh 4x4 will do great" even though it will be way over priced, pluse coding for the information of the 2 processors to even communicate is gonna make it a tad bit slower then real dual core, making the quad core kentfield a more eye catching object.

 

[SuperFly03]

Direct X 10 could prove just as interesting as the C2D and C2Q have proven. I understand they are adding a third shader and moving towards a unified shader architecture to allow dynamic allocation of shader units so that shaders can be allocated based on need rather then arbitrary production decisions.

Also, I am in the process of testing a GPU OC theory conflict: Which is worth more in GPU OC'ing the core or memory MHz for MHz. I contend that the Core is more worthwhile, while I have had a longtime veteran contend that the memory is more worthwhile (given his test is 3-4 years old).

Prelim data suggests the core is better, but I will have to go through time demos in some games and given my short supply of them it should be interesting. Today I took the time to put 3DMark06 through the paces, next I will move onto Oblivion, Quake 4, and maybe HL2 if I can get my hands on those games w/o paying.

--Jumping

I would suspect M$ has taken some strides to provide good multi-core support, even beyond dual core. If they haven't taken quad core into account, I would expect a patch of some nature pretty damn quick. For all the complaints people lodge against M$ they do a decent job at responding to the industry (note I say decent, it depends on your perspective and experience).

Vista could be amazing or it could be a trojan horse designed to steal our money. I have played with it and its nice, but I haven't taken the time to really toture test it and find out where all the power user tricks hide.

 

[ltcommander_data]

Kentsfield may become the hot ticket item when Vista hits as I suspect Vista will be nicely tuned for multicore support, at least at the consumer level. Not sure here --- do you have any opinions?

Yeah, the hope is that Vista will bring improved multicore support. However, I'm thinking that between the extra eye candy and the need for developers to optimize for Vista's new environment, we won't be seeing the benefits in the near term. In theory, XP's multicore support should have been pretty good considering it's descended from Windows 2000 which already support multiprocessor systems.

In regards to why it took so long to get single cycle SSE processing it was because SSE was always kind of a tacked on feature. It's really the major difference between the PII and PIII. Allowing single cycle 128-bit processing would have required considerable redesign to the PII with widening the units and ports. It probably wouldn't be that complicated as in you just need to widen existing things, but it still takes work to lay everything out again and limit the bottlenecks that the increased processing potential creates. So the PIII just used the existing 64-bit structure of the PII to do SSE and the PIV tried to avoid the problem by using frequency scaling. It's only now that Intel finally had to take a step back that they took the time to widen everything, which they seemed to do excessively although bottlenecks still do exist like the 16-byte predecode width.

 

[clue69less]

Holy moly this review is completely crazy!! The Core 2 Quadro will be so future proof.

Yea, prolly make it till next Tuesday at least.

 

[Artmic]

Unless AMD has commissioined a gaming company to write some 'AMD optimized 4x4 code', Kentsfield is showing us roughly how gaming will interact in this environment. COD2 obviously was poorly coded for multi-core environments, it won't even run on anything >2 core.

I have no doubt that Intel is going to be offering developers support to multithread. It was my understanding that Quake 4 and COD2 are as multithreaded as they are now because Intel offered direct support. (Intel just released support for better mobile game development, http://www.theinquirer.net/default.aspx?article=34207, which is nice). Obviously though most "multi"-threaded applications are actually barely "dual"-threaded. In general I think we've pretty saturated the performance benefit of more cores at 4 for the time being. Which is pretty quick considering they were just introducing dual core not long ago. I wonder how long it's going to take for developers to catch up? It's still questionable whether it's more beneficial to push multithreading first or to push 64-bit first.

Any opinions on whether Kentsfield is a 1067MHz or 1333MHz part from my earlier post?

Hmmm, i wonder if Intel would really spend time with helping developers with Multithreaded code.

I doubt the developers are stupid, they know what they are doing, the question is how they do it. meaning how they divide the workload between the cores.

if Core A has a 100 percent load and Core B is at 20 percent load due to concurrency dependencies god knows what else, no amount of tweaking will speed up the process. How do you run Gaming code in parallel?
I tried to tweak some code so that it would utilize a second core, and it takes a long time to switch something made for one core into something that will take advantage of the second core, damn I wish there was a compiler flag that would do this work for ya 😉 lol

 

[SuperFly03]

Holy moly this review is completely crazy!! The Core 2 Quadro will be so future proof.

Yea, prolly make it till next Tuesday at least.

Brilliant remarks as always 😀

Glad someone has a sense of humor.

Artmic--

It is to Intel's advantage to help developers utilize multi-core tech, imo. Honestly, why would I buy a octcore proc if a quad core proc does everything I need to and doesn't stutter? However, if they can get the code through effectively, Intel may be able to sell the idea of an Octcore CPU.

I realise its a very high level and simplistic arguement, just like the arguement about single-core vs. dual-core, but I think most people can get onboard with a dual core proc these days.

 

[ltcommander_data]

SSE in 1999 was actually are response to 3DNow!, which launched in 1998. Some people say that 3DNow! was an example of AMD leading the way although 3DNow! in itself was a response to Intel's 1997 MMX. It's just good competition. It's true that the best way would have been for single cycle operation to begin with, but Intel probably didn't have the time and needed to respond quickly to 3DNow! so they just tacked it on to the existing PII and relabeled it the PIII. Realistically, the latter Coppermine PIIIs were probably a bigger redesign than the transition from Deschutes PII to Katmai PIII.

 

[SuperFly03]

I hate you all. I was only about 10 when that crap started happening. Hell I didn't even have a computer back then, didn't get one till I was in high school really. Damn you people and your long term experience with this crap!

In any case, I want to see what effect single execution SSE has on such programs such as Folding@Home which are driven by SSE extentions. So, I assume C2D was the first architecture with single exectuion SSE?

EDIT: Post above and mine went at the same time, see --- SuperFly03 knows the name of the game too ....

Skills baby! lol

 

[corvetteguy]

This is very impressive, except for the apps that perfomed worse on the QC but like you said, poor programming.

I think this will be great, but again i think AMD has a better platform and a better future. I figured out that 4x4 will be compatible will all future processors and will have specially made ones until 65nm comes out. It'll be a great system when you get two native quad cores, quad sli and 8 gigs of RAM. But i am pretty sure it can also be priced down to compete with "normal" systems like kenstfield. The only expensive part that can't be toned down will be the RAM but hey, you still get more RAM.

I donno, again i think Intel will pwn until 07' 1h and then around next summer AMD will come back.

I'm gonna laugh when i get flammed for holding out hope for AMD

 

[SuperFly03]

Ok give me a little more credit than that, I am not silly enough to believe it to be a 128-bit cpu lol.

Although with all with all these damn bits flying around it is easy to get confused, 128 bits for dual channel, 32 bit for OS, 256 bit GPU memory interace, now 64 bit OS with 32 bit support, DAMN IT MAKE UP YOUR MIND!

Corvetteguy--

You won't get flamed by respectible people (of which there are few of these days). As jumping said, as long as your objective about the brand you like then you can say whatever you want. The key is in acknowledging the faults if there are some. That is the diffrence between a fanboy and a brand loyalist in my book: the ability to admit when the competition has a better product and why.

I love my 7900GTX (esp since I got it on evga step up program) but I know that the X1950XTX beats it hands down and I can admit that, but I am still perfectly happy with it.

Edit: I would expect Intel to be unchanllenged until H1 2007, and probably still bring both guns through 2007. Hopefully AMD can stay competitive

 

[SuperFly03]

Ok give me a little more credit than that, I am not silly enough to believe it to be a 128-bit cpu lol.

Apologies, please keep in mind that when I do things like this I do it because you are not the only person reading the post --- I do it for the casual reader who may not even be a member.

As I said --- you certainly know the name of the game.

Jack

Good point, I forget sometimes there are people who read threads and never post, who might get confused if they were quickly scanning.

I have a tendency to forget that not everyone thinks the way I do, gotten me into alot of problems 8O .

 

[apache_lives]

heh at first AMD fanboys were saying the QDR fsb will cap conroe's performance, then the quad core with the same BS story - P4 WAS SLOW IF ANYTHING, NOT THE FSB (P4 did what is was supposed to do EVENTUALLY - it beat the K7 ), and besides, if the "slow" fsb is capping the performance, then what happens when they replace it intel still has that ace up there sleve.

AMD's 4x4 design (aka server dual socket for a desktop machine) is as half ass'd as a P4 EE and Smithfield (two wrongs dont make a right, twice the threads wont do anythin at all, two prescotts wont make a decent chip, twice the cache wont do any better and twice the sockets sure as hell aint goin to do much better either), and the FX-62 (2.8ghz) can barly match an E6600 (2.4ghz), let alone a quad core at 2.66ghz (then theres the small issue of head room - 3.3+ghz vs 3ghz, plus the small matter of clock for clock performance), they cant clock there chips any higher (nice one IBM), and if they want to compete with a Intel Quad Core (currently what.... $999?) there going to have to half there prices of the top chip (FX-62) to ~$500 each, then theres the boards (well lack of concidering the price and complexity thanks to 940 per socket and the low demand -> high price), and then theres the lower performance, high power/thermal demands and not to mention the lower performance, things sure are lookin crud for AMD here.

I should also mention that if AMD does bring dual socket boards to the desktop... theres absolutly NOTHING AMD can do to stop intel doing the same, and i must say OUCH if Intel does cause 2 x 4 core chips are goin to be a major pain in the ass for AMD, and WTF is 4x4 - ITS 2 x 2 core cpus, what is it? marketing BS?

K8L better be bloody fast at a good price and they better hurry up, things aint lookin good.

 

[ltcommander_data]

WTF is 4x4 - ITS 2 x 2 core cpus, what is it? marketing BS?

The first 4 is supposed to be 4 cores and the 2nd 4 is supposed to be Quad SLI. It was originally and AMD-nVidia partnership which is why the ATI deal seems funny although it may be to satisfy nVidia.

I don't think it's in Intel's best interest to respond with a 2 processor desktop system. The dual FSB would be quite expensive for them to implement and they'll need chipset redesign to get rid of the FB-DIMM MC in their server chipsets. It'll also dilute their server market which is the last thing they need since they are just starting to regain ground with decent technology.

 

[corvetteguy]

Ya i know you and some others don't flame people------except baron. But many do so i thought i'd make a prediction to make them look like idiots.

You should hear what people call me. Fanboy, loyalist, noob, brainwashed consumer whore, traitor, enemy of the state

@Apache, i just saw him post

4x4 is not, "half-assed" and even if it doesn't beat kenstfield it will be a powerful platform when some better cpu's are put in it. Like i said, 2H 07' will be a golden time.