Sign in with
Sign up | Sign in
Your question

AMD and Intel Upgrade Path (And noise)

Tags:
Last response: in CPUs
Share
September 29, 2007 9:19:30 PM

I am in the market for a new machine now, and the situation can not wait (dead CPU). I have a fairly loose budget, but am not a bleeding edge kind of fellow. What platform, at this time, has the best upgrade path? I want and easy upgrade path to the new gen/rev of processors that will be coming out soon (Penryn or Phenom).

My current (dead) setup is:
AMD X2 4200+ Manchester
nForce4 Ultra 939
nVidia 7800GT
OCZ 2GB DDR PC3200
Zalman 7500-CU

Things I do:
WoW
Supreme Commander
HL2 and derivatives
TFC (when it comes out)
Software Development (physics engine, very CPU intensive ATM)
PVR

Which, other than Supreme Commander and probably TFC is more than enough for my needs, but I can't justify fixing the current rig (buying obsolete hardware just doesn't jive with me).

Right now I am considering one of the following:

C2D E66xx or E67xx
P35/775 based board
2GB DDR2 800
Same vid
Zalman 9700

or

X2 6000+
nForce 570 or 590 AM2 board
2GB DDR2 800
Same vid
Zalman 9700

Which has the best upgrade path? Optimally, I would want to be able to do a drop in replacement of the CPU in about 6 months with a Phenom or an Intel quad of some sort. Is that possible, or am i stuck with replacing the MB and RAM again too? Do I need an AM2+ to later drop in a Phenom? Am I likely to get a board that a manufacturer will provide support for Phenom? Will Penryn likely be supported on the board above?

Also, which of the above has the best thermal and noise characteristics? I can't find any direct comparisons on the net.
September 29, 2007 10:33:17 PM

If you want the best price/performance i would buy this
C2D E2160 or an E6600 if you don't know how to o/c
A-P35-DS3L
Ram: Patriot Extreme Performance 2GB (2 x 1GB) 240-Pin DDR2 SDRAM
Same vid
Tuniq Tower 120/ or Thermalright ultra 120 extreme

You should spend about $300-$320 on this upgrade and you can easily overclock the e2160 to 3.0-3.2ghz (will smoke the x2 6000+) and you can just drop a quad core Penryn when they come out in January.
September 30, 2007 12:02:33 AM

Get the Q6600 now and forget upgrading in 6months.
The Intel Chips coming up in 6months will only be a modest upgrade.
The Chips coming out in late '08 are going to be the next gen ones.

So you are better off waiting about a year for your upgrade.

The AMD's put off alot more heat than the AMD chips.

(Note: Current Mobos will accept both Phenom and Pennryn)
Related resources
a c 479 à CPUs
September 30, 2007 12:21:44 AM

From a strictly "long term" point of view. Going with AMD may be the better choice.

AMD will be releasing socket AM3 CPUs in 2008 since they plan on switching over to DDR3 RAM. AM2 and AM3 are pin compatible so AM3 should last at least until 2009. I'm not sure what socket the next generation of AMD CPU will use (after Phenom). While socket AM3 CPUs can be used on socket AM2 motherboards, AM2 CPUs will not work on socket AM3 motherboards because of the lack of DDR2 support.

As for Intel, their next generation of CPUs (Nehalem - 2nd half of 2008) will require a new motherboard since Intel is changing sockets again. I think LGA775 had a 3 or 4 year run. That means Penryn CPUs will be your final upgrade path if you build a system around the Core 2 Duo right now.

As for performance, I think Intel's continuing line of Core 2 Duos will outperform AMD's Phenom CPUs especially since the initial Phenoms are having some manufacturing problems which is why they will be "Tri-Core" CPUs until AMD can improve their manufacturing process.

Hypothetically speaking, if you want to upgrade a PC you build today in late 2009 then you should be able to do so if you go down the AMD path. You should be able to drop in any socket AM3 CPU into a current generation socket AM2 motherboard. you will only lack support for DDR3. At this point in time DDR3 doesn't really offer much improvements over DDR2, the that should change by 2009.

Upgrading an Intel platform in 2009 means you will be stuck with a Penryn CPU. That's not inherently bad, but whatever new generation of CPU AMD releases in 2009 (if they release one) may outperform the Penryn.

I intend on upgrading my PC when Penryn is released next year, but if I see some preliminary benchmarks of Nehalem's performance, then I'll probably wait for Nehalem.
September 30, 2007 7:42:36 AM

I agree with jaguarskx, that going with AMD will be more suitable for you in "long terms". The future AM3 CPU can easily drop into current AM2 Motherboards (if AMD keeps their promise). However, if you want gaming performance, Phenom will likely not live up to the hype. Since it is a server chip, it is designed to handle server applications, but not desktop applications. If you only game, Core 2 will be your best bet.

As a result, I would suggest you going with P35 board, and an E6x50 CPUs. E6x50s are less expensive than their E6x00 counterparts, and run at the same speed. If you're not into OCing, you should seriously consider E6x50 series.

If you want to use this PC for the next 3~4 years, you can easily upgrade your computer to Penryn, which will be relatively cheaper by then. However, because Intel's newer generation CPU codenamed Nehalem will come out in 2H08, and they feature a complete different socket, you will not be able to upgrade to 8 cores.

It all comes back to whether the gaming softwares will greatly take advantage of multi-core by 2009. Chances are games won't be fully optimized for multi-core by then, so having a quad-core system should still be relatively safe. I'm sure having a 3.4~3.6Ghz quad core processor can still hold its place in terms of performance.

I know I'm getting a little wordy here, but I promise this is my last point :D . Pouring your money into graphic cards will be better off than pouring money into processors. You'll have a better gaming experience.
October 1, 2007 4:17:13 AM

So, I decided on the following:

Abit IP35 Pro
C2D E6750 (don't care to overclock, would rather not deal with the issues)
2x1GB DDR2 800 4-4-4-12
430W ATX12V 2.0
Zalman 9500 (I like a quiet box, plus might OC the chip once it's past its prime in performance)

I don't care to overclock, so I went with the 6750. It's dirt cheap comparatively though, since I bought the 4200+ for $400 and the E6750 today is only $200, lol.

I'm not much of a 3d gamer, so my 7800GT is doing just fine for me right now. The E6750 should greatly improve my Supreme Commander performance, which is my only real performance issue with my current rig.

Thanks for the thoughts!
October 1, 2007 6:22:08 AM

get the p35 forget the amd fanatics - if that is not the lamest amd argument - i have heard yet

the new intel chips in 08 will last 5 years with 4 fast cores

skip the abit go for asus p5k- e wifi

overclock is the only way to go - if you can not bare to do it yourself the mobo has auto oc
October 1, 2007 7:46:48 AM

dragonsprayer said:
get the p35 forget the amd fanatics - if that is not the lamest amd argument - i have heard yet

the new intel chips in 08 will last 5 years with 4 fast cores

skip the abit go for asus p5k- e wifi

overclock is the only way to go - if you can not bare to do it yourself the mobo has auto oc


Not everyone wants to touch overclocking. Whether its because they have no experience, don't want to risk their processors, or simply don't have the equipment to do it, you have to respect their choice.

Please don't associate Core 2 with overclocking. I know they overclock very, very well, but that doesn't mean everyone who gets a Core 2 should overclock.
a b à CPUs
October 1, 2007 8:37:55 AM

Fedaykin311 said:
So, I decided on the following:

Abit IP35 Pro
C2D E6750 (don't care to overclock, would rather not deal with the issues)
2x1GB DDR2 800 4-4-4-12
430W ATX12V 2.0
Zalman 9500 (I like a quiet box, plus might OC the chip once it's past its prime in performance)

I don't care to overclock, so I went with the 6750. It's dirt cheap comparatively though, since I bought the 4200+ for $400 and the E6750 today is only $200, lol.

I'm not much of a 3d gamer, so my 7800GT is doing just fine for me right now. The E6750 should greatly improve my Supreme Commander performance, which is my only real performance issue with my current rig.

Thanks for the thoughts!

I'm hoping that your 430w PSU is of decent quality? If you buying this with your other parts, we can help suggest some decent PSU's, if you don't already own it. Did you have a budget for all of your parts?
October 1, 2007 12:03:39 PM

yomamafor1:
I think you're a little mixed up. Phenom is not a server chip, Barcy is.
October 1, 2007 5:08:03 PM

I was comparing the architecture of Phenom and Core 2. I apologize if my wording caused confusion.

Phenom is a derivation of the server chip Barcelona, so its architecture is more suitable for server applications, like heavy floating point math, and scalability.

In desktop environment, you rarely need scalability (unless you want to be FASN8ed), floating point math, and fast interconnect.

On the other hand, Core 2 is designed from Mobile, so its a lot more relevant to everyday computing, including gaming.

This is why you see Core 2 trouncing Phenom in game FPS, and load speed.
October 1, 2007 5:31:11 PM

Hmmmm, I did not know that AM3 CPUs would be compatible with AM2 boards. Where have I been? Still messing around with socket 939 I guess. Shame on me!
October 1, 2007 5:38:03 PM

You can still get GOOD 939 opterons cpus.
October 1, 2007 5:48:10 PM

errr no phenom gaming specs have been released, if you can post links i would like to see them. only benchies that have been released have been barcelona bench's in games, and barcelona is the server chip, on a server motherboard with ECC ram at 667.....phenom will be better for gaming then barcelona consider it will have htt3.0 and ddr800/ddr 1066 (standard memory not server memory) that and from what ive read, k10 will be very similar if not better to penryn clock for clock. do your research dont compair apples and oranges(server to consumer) chips. as soon at we get some phenom benchs we'll see how it stacks up with penryn(imo i think it will be as good clock for clock as penryn maybe a little less) also k10 has samples running at 3.7ghz so later revisions of the silicone will render much higher clocks, and hopefully great oc's :-D. im not a fanboy just want to get the facts straight. if i am wrong in any of this info send me a link to where it says im wrong and i will admit im wrong(we all make mistakes) id say if you want upgradeability, go with an AMD setup, if you want the best system now, go with intel. its the end of the line in mid 08 though because of the socket change and intels IMC with quickpath, then things should get interesting :-D. i just realised how much i just typed and im going to shut up now.
October 1, 2007 5:57:58 PM

Quote:
k10 has samples running at 3.7ghz


I am interested on where you got this info? I havn't heard of them going anywhere near this.
October 1, 2007 6:06:33 PM

if i had any idea what site i read that on i would post it :(  but i do rember reading that(im fairly sure i didnt dream it up....) ill look around i visit so many sites its hard to keep track of what i read where, but im preatty sure phenom will be launching on B3 or B2 revision. ill be sure to look around tonight for where i read all this.
October 1, 2007 6:09:18 PM

Yes, now, if you had to choose between an AMD Athlon 64 X2 4200+ Toledo 2.2GHz 2 x 512KB L2 Cache or an AMD Opteron 165 Denmark 1.8GHz 2 x 1MB L2 Cache for video encoding purposes, logic would seem to dictate the Opteron, right?
October 1, 2007 6:11:09 PM

im runnin a opteron 170@ 2.8 right now. the new steppings of all the 939 opterons oc like mad:p . some 170's have been seen at 3.0+ on air.
October 1, 2007 6:38:50 PM

bwdsmart said:
errr no phenom gaming specs have been released, if you can post links i would like to see them. only benchies that have been released have been barcelona bench's in games, and barcelona is the server chip, on a server motherboard with ECC ram at 667.....

Correct. The other "demo" we saw was at AMD's Analysis Day, which was a 3.0Ghz Phenom running Stranglehold w/ Tri-fire. There's no clear way to judge the performance of the rig.

Quote:

phenom will be better for gaming then barcelona consider it will have htt3.0 and ddr800/ddr 1066 (standard memory not server memory) that and from what ive read, k10 will be very similar if not better to penryn clock for clock.

Ok, wrong, and wrong.
Phenom will indeed have HT3.0, but it will not benefit in gaming. If bandwidth is so important to gaming, why is Intel's Core 2 trouncing K8, when FSB is clearly inferior than HT?

Secondly, according to Anandtech, K10 only has about 15% IPC improvement over K8, which is still another 15~20% short of Core 2. Penryn is about 5~10% superior than Core 2 in IPC. You do the math.
Quote:

In other words, how much faster can we expect Phenom to be vs. the Athlon 64 X2? To put it succinctly, it looks like around 15% clock for clock


http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3092&p=5

Quote:

do your research dont compair apples and oranges(server to consumer) chips. as soon at we get some phenom benchs we'll see how it stacks up with penryn(imo i think it will be as good clock for clock as penryn maybe a little less)

I would say, with all due respect, I did a lot more research than you do. True, I shouldn't compare apples to oranges, but it is pretty clear that K10 will not dethrone Core 2 architecture in terms of performance. Penryn will extend that. On the other hand, with Phenom completely NDAed, we can only derive Phenom's performance from Barcelona, since they share the exact same core, just different package. HT3 will not save Phenom. DDR2 800 will likely slightly improve Phenom's performance, but not to a substantiate degree. AMD's memory divider has been known for its inability to completely accommodate RAMs at different speed. Hence, using DDR2-800 will benefit in some cases, but not all cases.

Image courtesy of George Ou from blogs.zdnets.com/Ou


As a result, you won't see a fixed improvement with DDR2-800. Phenom's only chance is clockspeed, which with AMD's 65nm yield at 30%, I really don't see it happening.

Quote:
also k10 has samples running at 3.7ghz so later revisions of the silicone will render much higher clocks, and hopefully great oc's :-D.

This sentence basically tells me that you've done NO research at all. Where is the demo that showed K10 running at 3.7ghz? What is the system specification? What is the methodology?

The highest clocked K10 was allegedly presented at AMD's Analysis Day, which is 3.0Ghz. However, no one was allowed to run any benchmarks, or view the device manager, or leave the game demo. The "3.0Ghz" was from a screenshot taken by AMD, BEFORE the Analysis Day, that was distributed to the media.

In other words, no one can verify the validity of 3.0Ghz. If we exclude that one, the highest clocked Barcelona is at 2.5Ghz, which was done by Anand.

Quote:
im not a fanboy just want to get the facts straight. if i am wrong in any of this info send me a link to where it says im wrong and i will admit im wrong(we all make mistakes).

How about, put a link up whenever you make a claim to back it up? You just made a bunch of claims that no one can back up, including you. e.g. 3.7Ghz Phenom.

id say if you want upgradeability, go with an AMD setup, if you want the best system now, go with intel. its the end of the line in mid 08 though because of the socket change and intels IMC with quickpath, then things should get interesting :-D. i just realised how much i just typed and im going to shut up now. said:
id say if you want upgradeability, go with an AMD setup, if you want the best system now, go with intel. its the end of the line in mid 08 though because of the socket change and intels IMC with quickpath, then things should get interesting :-D. i just realised how much i just typed and im going to shut up now.

I agree with you, only if you want strictly upgradeability. If you want performance AND upgradeability, Intel is much more suitable for you. True, you will be able to drop AM3 processor into AM2 sockets, but that's will disable some features. In terms of performance, Phenom will likely not be able to compete with Core 2 clock-for-clock. As a result, if you purchase a P35 board, you can drop in Penryn later, which will likely to last you through 2010. As Penryn's speed increase, you'll get much better performance than Phenom.
October 1, 2007 7:17:48 PM

Good grief!

I must take that Evelyn Woods speed reading clourse if the posts are going to be this long!
October 1, 2007 7:21:11 PM

Much disinformation is going on in this thread.

yomamafor1,

I agree with you that the chip is designed differently than the Conroe chips, but I don't agree that the Conroe chip is designed for mobile and the AMD chip for servers. I know Core did stem from mobile processing but they designed the chip to be a good general purpose CPU, as AMD intends to do with it's chips. They don't intend to design a chip that is good for servers and nothing else. It just happens that Core is a better balanced CPU. At least this is my perception of the designs.
October 1, 2007 7:29:01 PM

yomamofor,

AMD historically increases performance of CPU's with each revision. It's safe to say that with new CPU revisions and new board revisions we will see a much faster cpu in the wild. I am not saying it will be Penryn, but I think in the end it will be a worthy competitor. Right now, with the first releases, AMD isn't performing up to where would have hoped, but they will get there. History shows that this is how AMD works (speaking of incremental revisions increasing performance).
October 1, 2007 7:44:45 PM

weskurtz81 said:
Much disinformation is going on in this thread.

yomamafor1,

I agree with you that the chip is designed differently than the Conroe chips, but I don't agree that the Conroe chip is designed for mobile and the AMD chip for servers. I know Core did stem from mobile processing but they designed the chip to be a good general purpose CPU, as AMD intends to do with it's chips. They don't intend to design a chip that is good for servers and nothing else. It just happens that Core is a better balanced CPU. At least this is my perception of the designs.


Conroe is a derivative of the Banias core, a.k.a Pentium M. After Intel abandoned Prescott, they basically took the Banias and beefed it up. As a result, it has a combined advantage of performance as well as efficiency. Thus, I usually refer Conroe as a mobile derived chip.

You can tell by Clovertown's performance in the server field. With 1S or below, Clovertown holds the lead, because there is no scalability. At 2S, you see that K8 will perform similar with Clovertown. At 4S or above, K8 basically reign the area, because its originally designed for server applications, and it has excellent interconnect.

As for K10, the architecture is more suitable for server applications. There are several distinctive features for Barcelona. I'll list two of them: Hypertransport, and native quad core. Let's break it down.

Hypertransport ensures superior scalability, as well as excellent communication with other components in the computer. This is why you see Barcelona will trounce Core 2 in any applications that require constant reading and writing to the RAM.
Native quad core also ensures superior scalability, as well as better efficiency.

The ones I listed generally help in multi-socket, or aka, server applications. However, for desktop and mobile, having all these features don't necessarily translate into performance.

Hypertransport is good for server, and for applications that require constant reading and writing to RAM, but it doesn't help in mobile and desktop applications. Most mobile and desktop applications use very little of the bandwidth provided by HT. As I said above, if HT is essential to computing, why is Core 2 trouncing K8 in mobile and desktop segment?

The impact of having a native multi-core will be significantly reduced due to the above reason. Clovertown is a MCM, yet it still outperforms K8 in these sectors.

So, in conclusion, K10 was designed to ensure AMD have the primary bread and butter they need, (server segment). If they can conquer the other segments, great! But the design is less optimized for desktop and mobile segment than Intel.
October 1, 2007 7:48:57 PM

weskurtz81 said:
yomamofor,

AMD historically increases performance of CPU's with each revision. It's safe to say that with new CPU revisions and new board revisions we will see a much faster cpu in the wild. I am not saying it will be Penryn, but I think in the end it will be a worthy competitor. Right now, with the first releases, AMD isn't performing up to where would have hoped, but they will get there. History shows that this is how AMD works (speaking of incremental revisions increasing performance).


True. We may see a 10% improvement just from B1 revision to BA revision. However, with Penryn in the horizon, it is unlikely that AMD will improve their process enough in time to become competitive in performance.

Another thing I want to bring up is yield. It has been rumored (and slowly confirmed), that AMD's 65nm can only yield Barcelona at 30%. As a result, AMD will likely to improve the yield first, then to the performance.

For performance / price, Barcelona will be a good competitor. But strictly performance speaking, Core 2 still have the reign, and it will likely to extend to 2009 and beyond.
October 1, 2007 8:37:35 PM

yyomamafor1,

As far as yield, I would imagine they have revisions in work while also working on the yield problems. I don't know enough about the process to make an educated guess though. Just trying to use logic.

As far as performance vs Penryn, I think it will depend on one thing, how fast AMD can pump out the revisions. I think they might be able to do some good if they were already closing in on the BA revision or whatever revision is next and only released on B1 out of necessity.

Because of the fact that AMD normally increases performance with each revision I think they will be competitive. They might not be faster than Penryn is when it comes out but I think it is possible that Barcy will be able to compete. It all depends on AMD and how fast they refine the Uarch.
October 1, 2007 8:48:29 PM

yyomamafor1,

Like I said I can understand what you are saying, but these aspects that make the CPU a good performer on servers will not hinder it on the desktop side. I think what hinders it on the desktop side is it's lack of being more balanced like the Core cpu's. I do fully understand what you are saying, but I don't think AMD purposefully left out of the CPU knowing it would lack on the desktop. I am sure they know the HTT and native quad design helps in multi cpu environments, but I don't think it is these things that hinder the CPU in the desktop environment.

As far as Hyper transport not helping on the desktop arena, if FSB's weren't going as high as they are now I think we would definitely see a bottleneck in the desktop system with a 4 core cpu. Also, when the core count goes beyond 4 cores per die I also think they will either need to redesign the FSB or implement HT.

I understand and am aware of all the points above, but I just disagree with the statement that the reason it is slow on the desktop is because it was designed for server applications. The implemented designs to help increase scalability and efficiency shouldn't cause any decrease in performance on desktop apps. If the desktop performance, Int and FP were increased, it would also increase on the server apps.

I think we are splitting hairs here, I just look at it a bit differently than you do.
October 1, 2007 10:42:51 PM

If we can see a thing from different perspectives, the result is a view with more depth.

Phenom will be a hit. Penryn will cost a lot.
October 1, 2007 11:01:37 PM

onestar said:
If we can see a thing from different perspectives, the result is a view with more depth.

Phenom will be a hit. Penryn will cost a lot.
With technological advancements literally taking place overnight i think we're in for one heck of a ride.
I agree that AMD has geared itself towards the server market, Intel is more mainstream, end-user type.
We need AMD to pushing the envelope; without competition Intel would have us over a barrel...
October 1, 2007 11:40:48 PM

weskurtz81 said:
yyomamafor1,

As far as yield, I would imagine they have revisions in work while also working on the yield problems. I don't know enough about the process to make an educated guess though. Just trying to use logic.

...and when you use logic, as well as your comprehension ability, you'll notice that AMD has some manufacturing issue.
1. why are there very limited supplies?
2. why is Barcelona only releasing at 2.0Ghz, instead of 2.3Ghz planned?
3. why is there a tri-core Phenom, that was never on the roadmap until recently?

I already posted in another thread how I obtained a 30% yield. AMD's slide depicted them having 0.5cm^2 defect density, coupled with Barcelona's massive die, you get a yield of 30%.

Now, for AMD to come up with revisions, as well as validate them, they'll need large amount of money. With their current limited quantities, as well as very low margin for AMD, I don't think they have enough money to work on that many revisions. We'll have to wait and see BA's performance.


As far as performance vs Penryn, I think it will depend on one thing, how fast AMD can pump out the revisions. I think they might be able to do some good if they were already closing in on the BA revision or whatever revision is next and only released on B1 out of necessity.

Because of the fact that AMD normally increases performance with each revision I think they will be competitive. They might not be faster than Penryn is when it comes out but I think it is possible that Barcy will be able to compete. It all depends on AMD and [b said:
how fast they refine the Uarch.]
As far as performance vs Penryn, I think it will depend on one thing, how fast AMD can pump out the revisions. I think they might be able to do some good if they were already closing in on the BA revision or whatever revision is next and only released on B1 out of necessity.

Because of the fact that AMD normally increases performance with each revision I think they will be competitive. They might not be faster than Penryn is when it comes out but I think it is possible that Barcy will be able to compete. It all depends on AMD and how fast they refine the Uarch.
[/b]

Correct. Then how fast they can refine the uarch depends on how many engineers as well as resources you can allocate. At the moment, it doesn't look like AMD have the financial means to come up with multiple revisions in a relatively short time.
October 1, 2007 11:46:54 PM

weskurtz81 said:
yyomamafor1,

Like I said I can understand what you are saying, but these aspects that make the CPU a good performer on servers will not hinder it on the desktop side. I think what hinders it on the desktop side is it's lack of being more balanced like the Core cpu's. I do fully understand what you are saying, but I don't think AMD purposefully left out of the CPU knowing it would lack on the desktop. I am sure they know the HTT and native quad design helps in multi cpu environments, but I don't think it is these things that hinder the CPU in the desktop environment.

True, HT and native quad will not hinder its performance in the desktop environment. You're right that K10 might be more specialized in a certain area, while Core 2 is a more general uarch.


As far as Hyper transport not helping on the desktop arena, if FSB's weren't going as high as they are now I think we would definitely see a bottleneck in the desktop system with a 4 core cpu. Also, when the core count goes beyond 4 cores per die I also think they will either need to redesign the FSB or implement HT.

I understand and am aware of all the points above, but I just disagree with the statement that the reason it is slow on the desktop is because it was designed for server applications. The implemented designs to help increase scalability and efficiency shouldn't cause any decrease in performance on desktop apps. If the desktop performance, Int and FP were increased, it would also increase on the server apps.

I think we are splitting hairs here, I just look at it a bit differently than you do. said:

As far as Hyper transport not helping on the desktop arena, if FSB's weren't going as high as they are now I think we would definitely see a bottleneck in the desktop system with a 4 core cpu. Also, when the core count goes beyond 4 cores per die I also think they will either need to redesign the FSB or implement HT.

I understand and am aware of all the points above, but I just disagree with the statement that the reason it is slow on the desktop is because it was designed for server applications. The implemented designs to help increase scalability and efficiency shouldn't cause any decrease in performance on desktop apps. If the desktop performance, Int and FP were increased, it would also increase on the server apps.

I think we are splitting hairs here, I just look at it a bit differently than you do.

I guess you're right in the sense that K10's specialized design in server application doesn't mean it will not perform in the desktop / mobile sector. However, when its compared to Core 2, those specialization in K10 will not improve the performance in a more substantial way.

I guess I'll have to word my arguments a little better :kaola: 
October 1, 2007 11:49:49 PM

onestar said:
If we can see a thing from different perspectives, the result is a view with more depth.

Phenom will be a hit. Penryn will cost a lot.


Sure, let's reconsider the circumstances.

Phenom will be a hit. Sure it will, IF it performs. However, so far we haven't seen any tangible proof on how it performed, except the "test" Anand simulated.

So will it be a hit? No idea.

Let's consider the next one.

Penryn will cost a lot. Compared to Phenom, it might cost a little more. At the moment, AMD can only use price as a mean to attract sales on the desktop Athlon X2s, and I believe they'll have to do the same thing with Phenom too. However, if you compare Penryn with the current quad core, it costs relatively the same.

So does Penryn cost a lot? Not really in the absolute sense.
October 2, 2007 12:10:10 AM

Actually the Penryn will likely cost less than Phenom.

Intel is already listing Penryn prices and they are quite low.

The Penryn will cost Intel LESS to produce than current C2D chips due to die shrinkage.

The Phenom will cost MORE due to the LARGER die size.
October 2, 2007 2:30:51 AM

i was never arguing that it would dethrone intel of the crown. i just said i think clock for clock it will be competitive. and the 3.7, i cant find where i read it, but i know i did at one point or another, i dont mean to argue, this is from what ive read and seen myself. i have done quite a bit of research as i read sites and forums daily for news. im not saying that HTT would make a huge boost in preformance. im just telling you what ive read and that imo it will be a good battle and price wars will wage and we all win :-D
October 2, 2007 3:08:05 AM

yomamafor1,

Yeah, you might want to hire a lawyer in order to make the post so verbose that no one without an exceptional understanding of lawyer lingo could argue with you;) .

No, those specializations today do not really improve the performance in the desktop arena. I suspect though that if you increase the core count above 4, maybe to 8, the FSB will start to show an inability to keep up. If the FSB would not have been boosted up to as high as it is today then I think it would definitely be a limit. And during the time of DDR1, when they were thinking about dual and quads they probably thought they would need to come up with something better than the FSB.

Quick question, if you grab a Q6600 and lower the FSB speed a coupld hundred Mhz, does performance decrease? I think the answer is yes but I cannot remember for sure.

As far as the yields, I don't really doubt they are having yield issues, but I expect this to be taken care of like they have in the past. I really do not know enough about the manufacturing process to make educated guesses. I don't know how introducing a new Uarch to an already well oiled 65nm process is going to affect things. If you have the time and care to explain how introducing a new Uarch to an already tested and working process causes problems.

I just always thought that they shrink the process and bring in the new Uarch when it's running well because it makes things much easier.
a b à CPUs
October 2, 2007 3:35:06 AM

weskurtz81 said:
yomamafor1,

Yeah, you might want to hire a lawyer in order to make the post so verbose that no one without an exceptional understanding of lawyer lingo could argue with you;) .

No, those specializations today do not really improve the performance in the desktop arena. I suspect though that if you increase the core count above 4, maybe to 8, the FSB will start to show an inability to keep up. If the FSB would not have been boosted up to as high as it is today then I think it would definitely be a limit. And during the time of DDR1, when they were thinking about dual and quads they probably thought they would need to come up with something better than the FSB.

Quick question, if you grab a Q6600 and lower the FSB speed a coupld hundred Mhz, does performance decrease? I think the answer is yes but I cannot remember for sure.

As far as the yields, I don't really doubt they are having yield issues, but I expect this to be taken care of like they have in the past. I really do not know enough about the manufacturing process to make educated guesses. I don't know how introducing a new Uarch to an already well oiled 65nm process is going to affect things. If you have the time and care to explain how introducing a new Uarch to an already tested and working process causes problems.

I just always thought that they shrink the process and bring in the new Uarch when it's running well because it makes things much easier.

I work directly with the manufacturing processes that others have talked about. Introducing new Uarch isn't easy. It's like designing something totally different, but still trying to accomplish the same task faster. The die shrinks that you refer to (65nm process) don't have anything to do with different Uarch'ers. When they do die shrinks from 90nm to 65 nm, they are just making the ciricuitry smaller. The wires between point A to point B become smaller, in general. There is alot more to this than what I'm explaining, but this is just a general overview. When you change Uarch, you are doing a total redesign of the circuitry. This takes time to work out.
When a new Uarch is designed, it is first ran through the fabrication process (30-60 days or so) and then tested. There are then changes made to the original design, which then are run through the fabrication process again. This process will take time to work out all of the problems that they come up with. Sometimes they find major issues with the design and have to do major overhauls to it. All of this requires engineering, manufacturing, equipment support and patience to achieve their goals. Sometimes during the redesign of the Uarch they find bottlenecks that require more equipment to meet the demand requested. There is alot of details that I'm sure I left out, but I hope this helps alittle. I've been in this industry since about 1986, so I've seen alot of different changes occur over the years.
October 2, 2007 3:43:02 AM

i believe amd was rushed into hitting this launch date and the silicone was not as revised as they hoped. im sure once later revisions come out yield will become better. (this is all imo no facts....) i just want to see competition. im no fanboy as i own both amd and intel, nvidia and ati :-D whatever is best for the price ill buy. yomamafor1 i didnt mean to try and prove you wrong above, i hope that you didnt think that. best of all, and let the best company win :-D
October 2, 2007 2:41:23 PM

lunyone,

Gotcha, I thought the process was generally at which node they are making CPU's. For example "AMD is having problems with the 65nm process". I would take that as it doesn't matter what they are making, they are having a hard time making it @ 65nm.

Then again, maybe process refers to the overall manufacturing process. So, then it could be any number of different thing throughout the entire process that is causing problems.

The main reason I have a hard time grasping why they would have as many issues is because my thought process is: why would they be able to make X number of CPU's to run as they are supposed to but the other 2X don't turn out right. What changes to make the other 2X not work properly? I picture it as being a either it's going to work or not(with the typical problems).

I just need to do more research on how it works.
October 2, 2007 3:15:27 PM

weskurtz81 said:
yomamafor1,
No, those specializations today do not really improve the performance in the desktop arena. I suspect though that if you increase the core count above 4, maybe to 8, the FSB will start to show an inability to keep up. If the FSB would not have been boosted up to as high as it is today then I think it would definitely be a limit. And during the time of DDR1, when they were thinking about dual and quads they probably thought they would need to come up with something better than the FSB.

It all comes back to, how many times does a core need to access data from the memory. In Intel's case, because they know their MCM and FSB solution cannot keep up with AMD's HT, they basically enlarged L2 cache to accommodate more data to be readily accessed by the core. As a result, even if FSB is slow, and shared by many cores, Core 2 can still pull a victory against K8.

Quote:
Quick question, if you grab a Q6600 and lower the FSB speed a coupld hundred Mhz, does performance decrease? I think the answer is yes but I cannot remember for sure.

Since FSB clock is directly associated with the processor clockspeed, if you decrease the FSB, you also decrease the processor clockspeed.

But I'll play with the idea later by adjusting the multiplier.


As far as the yields, I don't really doubt they are having yield issues, but I expect this to be taken care of like they have in the past. I really do not know enough about the manufacturing process to make educated guesses. I don't know how introducing a new Uarch to an already well oiled 65nm process is going to affect things. If you have the time and care to explain how introducing a new Uarch to an already tested and working process causes problems.

I just always thought that they shrink the process and bring in the new Uarch when it's running well because it makes things much easier.
said:

As far as the yields, I don't really doubt they are having yield issues, but I expect this to be taken care of like they have in the past. I really do not know enough about the manufacturing process to make educated guesses. I don't know how introducing a new Uarch to an already well oiled 65nm process is going to affect things. If you have the time and care to explain how introducing a new Uarch to an already tested and working process causes problems.

I just always thought that they shrink the process and bring in the new Uarch when it's running well because it makes things much easier.

The problem with Barcelona is the die size. AMD's 65nm manufacturing process did not have good yield to begin with. But assuming AMD has 90% yield on their previous X2 lineups, the yield would drop significantly if they move from X2 to Barcelona, as the die size increase by 2x.

I'm sure AMD can adjust and fix their process while pumping out Barcelona, but it is unlikely that they will reach mature yield before they make the jump to 45nm. Providing their yield of Barcelona at 30%, they'll need to work very hard, and implement some ground breaking changes, to improve their yield to a more satisfactory level.
October 2, 2007 3:32:18 PM

lunyone said:
I work directly with the manufacturing processes that others have talked about. Introducing new Uarch isn't easy. It's like designing something totally different, but still trying to accomplish the same task faster. The die shrinks that you refer to (65nm process) don't have anything to do with different Uarch'ers. When they do die shrinks from 90nm to 65 nm, they are just making the ciricuitry smaller. The wires between point A to point B become smaller, in general. There is alot more to this than what I'm explaining, but this is just a general overview. When you change Uarch, you are doing a total redesign of the circuitry. This takes time to work out.
When a new Uarch is designed, it is first ran through the fabrication process (30-60 days or so) and then tested. There are then changes made to the original design, which then are run through the fabrication process again. This process will take time to work out all of the problems that they come up with. Sometimes they find major issues with the design and have to do major overhauls to it. All of this requires engineering, manufacturing, equipment support and patience to achieve their goals. Sometimes during the redesign of the Uarch they find bottlenecks that require more equipment to meet the demand requested. There is alot of details that I'm sure I left out, but I hope this helps alittle. I've been in this industry since about 1986, so I've seen alot of different changes occur over the years.


Exactly. That is why I worry about AMD's future. With the current 30% yield on the Barcelonas, it will take them at least months to come up with the newer design, a new mask, a new wafer....etc. Its simply impossible to compete when Intel's Penryn is over half the die size of Barcelona, and approaching mature yield.
October 2, 2007 5:26:18 PM

humm ... yeah ...
here's what I can tell : if you go LGA775 way I can assure you that the suggested Asus P5K-E/Wifi (somewhere at the beginning of the thread) is a really good board. I have one and I'm very happy with it.
October 2, 2007 6:02:28 PM

Lots of good discussion going on here. I think i'll throw my hat in the ring...

I think one of the best points that's been eluded to in this thread is under what circumstances is HTT/QPI (AKA CSI) needed vs just a simple FSB.

As Intel chips have large caches (with even larger ones coming with Penryn), obviously the RAM - CPU requirements isn't choking the FSB. Really where the FSB is being overloaded is when chips are needing to talk to each other. Very few desktop systems are using multiple sockets - even at the enthusiast level. Servers obviously are the main (and almost sole) location where more sockets are required. At that point, the FSB is probably limiting performance much more than in a system with a single CPU.

So no, HTT isn't going to be a fantastic boon to Phenom over Penryn, since the FSB is probably not gating desktop performance. The constant idea of "FSB is just bad and old" kind of needs to be tempered with its advantages and information regarding how overloaded it actually is in most practical applications.
October 2, 2007 6:52:28 PM

Assurances aside, the original question involved the upgrade path.
Now if the AM2 and AM3 sockets are compatible, I see some advantage for the AMD equation. The other issue was Supreme Commander which will run excellent under an X2 5600 now, and Phenom later, assuming a good video card is acquired.
October 2, 2007 7:42:03 PM

Of course, Supreme Commander will run excellent with a C2D or C2Q right now (better than the 5600) and will run will with a Penryn later (speculation says better than the Phenom, but oh well, we'll just have to see), which is just as easily drop-in able as the AMD route.
October 2, 2007 7:51:31 PM

Wolverine is correct...it will run SC and it might be marginally better, if you consider the numbers only, however in real world terms, the difference is negligible, and it would cost substantially more money without substantial improvement in performance for the applications mentioned.

It occurs to me that some enthusiasts/extremists have no care about the price/performance ratio, however the question asked of the forum seemed to be interested in that aspect.
October 2, 2007 9:30:53 PM

onestar said:
Wolverine is correct...it will run SC and it might be marginally better, if you consider the numbers only, however in real world terms, the difference is negligible, and it would cost substantially more money without substantial improvement in performance for the applications mentioned.

It occurs to me that some enthusiasts/extremists have no care about the price/performance ratio, however the question asked of the forum seemed to be interested in that aspect.


Substantially more?

From Newegg:
E6750: 194.00
6000+: 169.00

This is the performance chart of Supreme Commander from Tom's CPU charts.
http://www23.tomshardware.com/cpu_2007.html?modelx=33&m...

So, for 30USD more, you get ~10 FPS more. Aside from that, E6750 dissipates a lot less heat as well.

For the upgrade path, from the data we have now, Penryn is going to dominate the gaming market on price / performance and absolute performance. It will likely to run cooler due to 45nm and HK/MG, and consume less energy than Phenom. (although I have no data at the moment).
October 3, 2007 4:40:22 PM

errr thats a 6000...not 5600. to me its about even price to pay for the fps. in mid 08 though, 775 will no longer be used is my only issue, am2, am2+ and am3 will be lasting atleast another few years i believe(they realised the 939-AM2 pissed the 939 users off alot....like me) i like to lean towards the underdog also, so if i were in your position id go amd.
October 3, 2007 5:29:17 PM

1749900,3,61808

The AMD's put off alot more heat than the AMD chips.

(Note: Current Mobos will accept both Phenom and Pennryn)[/quotemsg said:


Huh? and what?
October 3, 2007 6:07:34 PM

bwdsmart said:
errr thats a 6000...not 5600. to me its about even price to pay for the fps. in mid 08 though, 775 will no longer be used is my only issue, am2, am2+ and am3 will be lasting atleast another few years i believe(they realised the 939-AM2 pissed the 939 users off alot....like me) i like to lean towards the underdog also, so if i were in your position id go amd.


There aren't any new processors coming for Socket 775, doesn't mean you'll have to toss out your system the first day Nehalem launches. Plus, the performance of Penryn will be sufficient to last two~three years, unless you use your computer for something else far more intensive than gaming, and going on webs.

However, I guess everyone has their preferences.
October 3, 2007 6:38:16 PM

By the way, did I mention you have to buy a new motherboard for that E6750? Add in a few dollars for that, oh yes, and please do not forget a CPU cooler. Did someone say 30$ ?

let's see now, it appears the bottom line is a bit more than that.


Fanboys are like blonds, they mean well but you can't teach them anything.
October 3, 2007 6:55:40 PM

onestar said:
By the way, did I mention you have to buy a new motherboard for that E6750? Add in a few dollars for that, oh yes, and please do not forget a CPU cooler. Did someone say 30$ ?

let's see now, it appears the bottom line is a bit more than that.


Fanboys are like blonds, they mean well but you can't teach them anything.

Are you really trying to prove me as a fanboy huh?

Take a look at his previous system again.
Quote:
My current (dead) setup is:
AMD X2 4200+ Manchester
nForce4 Ultra 939
nVidia 7800GT
OCZ 2GB DDR PC3200
Zalman 7500-CU


Now go ahead and tell me he doesn't need a new AMD board.
CPU cooler? I'm sure 6000+ needs more cooling than E6750.

It costs him relatively the same to build either way, and I was just merely suggesting him that with Intel, he'll get more performance, and a better upgrade.
!