Sign-in / Sign-up
Your question

FSB limits performance on Intel?

Tags:
  • CPUs
  • Performance
  • Memory
Last response: in CPUs
July 11, 2008 12:51:19 AM

Check this

http://forums.guru3d.com/showthread.php?p=2740928

I think I have found the culprit: It's the FSB!

I did the following:
- changed my CPU clock from 2400 (9*266) to 1600 (6*266) - no change, 27 FPS
- changed my memory clock from 400 to 266 - no change, 27 FPS
- changed my FSB from 266 to 333 - success, 34 FPS

The FPS actually changed exactly according to my FSB change - a 25% increase in both cases.
It might still be memory bandwidth though, since with Intel CPUs the FSB is the limiting factor.
I have never actually seen any game that is limited by FSB or memory bandwidth nor any mention of such behaviour. I think we have found a new type of bottleneck here.

EDIT: I just ran another test: I decreased the memory clock to 266 while keeping the FSB at 333: Still 34 FPS. So it actually is the FSB.
Time for Nehalem, I'd say.

More about : fsb limits performance intel

a b à CPUs
July 11, 2008 1:05:43 AM

Intel is still faster then AMD
July 11, 2008 1:09:03 AM

I can play your silly game too:

Quote:
Hi

I recently got an AMD Phenom a decision that I am regretting now.

I've got a
* ASUS m2n-sli
* AMD Phenom 9500
*Thermaltake Soprano with a 430W (XP5500NP)
* 4Gb = (2 Kits of (2*1 GB Ram A Data)) 800
* Geforce 7900GS
*19 Screen (1280x1024)

I currently have another computer that's a p5n-e sli Core 2 Duo E6600 4 GB of ram and a 8800gts(320mb), same case but with a 22" monitor. This computer literally smashes the performance of the AMD.

I even tried putting the 8800gts into the Phenom and it only improved graphic performance mildly. The Core 2 Duo beats the Phenom even running at 1680x1024 vs 1280*1024.

The games I have tested it with are TF2 and Supreme commander.

Supreme commander is interesting as after 1 minute the processor performce deteriorates even just playing a 1 on 1 AI. This thing runs it worse than my old Athlon 3500 single core, even a 3800+ X2 I borrowed of a friend works better.

Have I configured something funny in my system or is it something else?

I know the Phenom have a bug that a Bios updates has to fix
The board I'm using is a AM2 (not AM2+) which has phenom compatibility
I heard AMD sent benchmarkers a core which didn't have these faults and higher clock speeds just to win. Is the phenom just a huge lemon?

Should I ?


http://forums.whirlpool.net.au/forum-replies-archive.cfm/944295.html
Related resources
July 11, 2008 1:20:48 AM

Since forum posts now equal reliable sources to you - here is another, running the same game, and much more recent:
Quote:
i have a phenom 9500 2.2ghz quad core, 3gb ram, 8600gts. In any town i barely get 30 fps. out of town i get a wonderful 50 fps. Now what i dont understand is how i only get 30 in town. So what i did is went to a place where i only got 27 fps and i saved it and turned the res down to 800x600 and i got the same 27 fps. what is wrong with my setup


http://answers.yahoo.com/question/index?qid=20080708190039AAuguOO

So, go wander through more forums looking for the magical saturated FSB.
a c 213 à CPUs
a b } Memory
July 11, 2008 1:24:28 AM

The TF2 is a great FPS to use for true CPU performance since Source is very CPU limited.

Of course I almost feel like this was made up. But I can't go to the sit ehe linked so meh.

Also I like that the guy never posted his multiplier for the 333FSB.
July 11, 2008 1:24:35 AM

NMDante said:
So, go wander through more forums looking for the magical saturated FSB.


Read the text! He did some informative OC to check where he was getting more performance and found out that for that game it was the FSB that was the bottleneck. This could be the case why Intel users notice a performance gain when the OC.
July 11, 2008 1:27:29 AM

Read the text - Phenom having same issues. Guess it's the Phenom's FSB too?

a c 213 à CPUs
a b } Memory
July 11, 2008 1:31:35 AM

NMDante said:
Read the text - Phenom having same issues. Guess it's the Phenom's FSB too?


LMAO

+1

Its more to do with the game itself than anything. Its poorly optimized. And at the res he was playing a poorly optimized game engine will kill your FPS no matter what hardware you have.
July 11, 2008 1:34:25 AM

jimmysmitty said:
You do understand it was a joke right? I mean.... come on. Hes not an idiot.
Man you must not have comedy in your area.


Aha, I do comedy in my own language ;) 
July 11, 2008 1:37:21 AM

kassler said:
Aha, I do comedy in my own language ;) 


Yeah, and the same FSB joke is getting kinda old... :p 
a c 213 à CPUs
a b } Memory
July 11, 2008 1:38:17 AM

I just realized something..... Your own post here is in a way proving you wrong. You talk about how the FSB limits performance since it cannot access memory faster. Well if this was true then when the guy lowered his memory clock speed he also lowered his memory bandwidth.

But still its more of the games fault than the CPU here.
July 11, 2008 1:42:34 AM

kassler said:
Phenom doesn’t have a FSB. Read about the hypertransport here: http://en.wikipedia.org/wiki/HyperTransport

fsb: http://en.wikipedia.org/wiki/Front_side_bus

Nehalem have removed the FSB


No way. Really?

So, explain why the Phenom system has almost the exact same FPS as the Intel "FSB problem" system.
Both are claiming to have low FPS, no matter what the resolution is set at. We are waiting for your explanation, since it can't possibly be a GPU or game coding issue. It must be an interconnect problem, right?

I'll be waiting for your technical explanation.
July 11, 2008 2:00:39 AM

but a rockin AMD = Intel with FSB problem
hmmm
o.0?
July 11, 2008 2:31:27 AM

I am not familiar with the Rampage Formula but on any of the later Intel chipset boards I have seen (p35and later) it doesn't allow setting the mem speed lower than 1:1 and don't see how he ran at 333 fsb with a 266 mem speed. I wonder if he just changed the strap rather than slowing the memory down to 266. If so how credible is anything else he said?

Also just look at the article that Toms just did on the p45 and see how much gain was shown going from 1333 to 1600 on the fsb (should be quite noticeable if the fsb is a bottleneck).

http://www.tomshardware.com/reviews/intel-p45-chipset,1961-15.html

From that article you can show that 1333 is faster than 1600 or that 1600 is faster than 1333 depending on the test result you pick.
Quote:
We ran all benchmarks at least three times, and used the average result of all benchmark runs. As you will see, there is no clear performance winner, as the differences are too small. This includes the enthusiast X38 and X48 chipsets, as well as the mainstream P35 and P45 chipsets, and 1333 MHz versus 1600 MHz—the effective difference is close to zero.
a c 118 à CPUs
July 11, 2008 3:05:48 AM

kassler said:
Check this

http://forums.guru3d.com/showthread.php?p=2740928

I think I have found the culprit: It's the FSB!

I did the following:
- changed my CPU clock from 2400 (9*266) to 1600 (6*266) - no change, 27 FPS
- changed my memory clock from 400 to 266 - no change, 27 FPS
- changed my FSB from 266 to 333 - success, 34 FPS

The FPS actually changed exactly according to my FSB change - a 25% increase in both cases.
It might still be memory bandwidth though, since with Intel CPUs the FSB is the limiting factor.
I have never actually seen any game that is limited by FSB or memory bandwidth nor any mention of such behaviour. I think we have found a new type of bottleneck here.

EDIT: I just ran another test: I decreased the memory clock to 266 while keeping the FSB at 333: Still 34 FPS. So it actually is the FSB.
Time for Nehalem, I'd say.


I can think of three reasons that he may have had better performance <i>initially</i>:

1. Northbridge I/O bottlenecking being relieved by higher northbridge clock speed. Northbridge clock speed is the same as the FSB clock speed, so if there was an internal northbridge I/O bottleneck this would most certainly help. He has two GTX280s in SLi and the PCIe x16 slots come from the northbridge, not the southbridge, so it is certainly a possible culprit and this is the kind of result you would see if you were GPU-limited and your GPUs were I/O-starved.

2. His memory straps aren't what he thinks they are and the strap stayed the same when he bumped the FSB clock from 266 to 333, overclocking his memory. That can give you better performance in a memory-intensive application.

3. The FSB actually was the bottleneck.

After reading his edit, I am almost certain it is #1 if he is accurate with his "I set my memory to 266 MHz when the FSB was at 333 MHz" bit. It makes #2 incorrect and #3 unlikely. An FSB clocked at 333 MHz carries up to 12.7 GB/sec while dual-channel DDR2 at 266 MHz (DDR2-533) only outputs 8.5 GB/sec. The FSB in a Core 2 Quad *does* carry information both between dies as well as to and from the northbridge, but inter-die FSB congestion is unlikely. Why do I say that? He was playing Oblivion and that is not very multithreaded: http://www.tomshardware.com/forum/240359-28-core-quad-performance-oblivion That means that there isn't going to be a whole lot of cache coherency traffic between the two dies and thus the FSB is basically going to be used for CPU <--> RAM data transfer. We'd need to get a hold of his system and try a few things such as a single-die C2D and play more with FSB, northbridge, and RAM speeds as well as using a single GPU to accurately tell what's going on, but I *highly* suspect it has to do with the northbridge being swamped with PCIe traffic from the two GPUs.
July 11, 2008 3:20:25 AM

It is an Intel chipset which means no sli so its a single card. It wasn't the original poster that is quoted in the post here it was Reaping_Ant on post #6.
a c 213 à CPUs
a b } Memory
July 11, 2008 4:49:03 AM

I have a P35 mobo and it doesn't have a ratio setting but lets you select the memory frequency. I haven't had to mess with it since I got it to a stable 3GHz and the memory to a 1:1 @ 1333MHz so I can't remember if you can go lower than 1:1.

MU I will only disagree with you that its not only Intel rigs that have problems with this game, it is also AMD rigs. This game has a pretty badly optimized engine and I need to try it on my new system but on my old one I had to cut AA and AF @ 1024x768 and move up to 1280x1024 (to try to get the same AA/AF effect). And I had a pretty dam kick ass system for the time, Pentium 4 EE 3.4GHz, X850XT and 2GB DDR800. It ran pretty smooth then but I had to have HDR off or it would kill my system.
July 11, 2008 5:38:04 AM

First he states that "changed my CPU clock from 2400 (9*266) to 1600 (6*266) - no change, 27 FPS"... Erm that means that the CPU is 30% slower at the same FSB, 30% less data transfer but still the "FSB is saturated"... Not likely...

Second item "changed my FSB from 266 to 333 - success, 34 FPS", he doesn't mention if he lowered the multiplier to compenstate. If he didn't well then it does prove that he is an idiot because then his clock speed and FSB both went up by 25% and performance increased by 25% but lower the clock speed by 25% and leaving the FSB where it is at did not produce a performance change. All of this indicates poor coding / config instead of a saturated FSB. Most likely a config issue as Obilivion + GTX280 SLI... I'm suspecting a driver issue here.

If it was an FSB related problem why does the AMD Phenom and its infamous HT also suffer of poor performance?

Thirdly at that resolution on a 680i motherboard... Well now theres a problem. My Q6600 @ 3.4ghz + 8800GT SLI general speaking produced a worse experience than my X38 + Q6600 @ 3.2ghz + a single 8800GT. This is more of a motherboard than FSB issue in my opinion.

Poeple must either post more complete details or not post anything at all. I play with a Q6600 @ 400x8, 4gb RAM and a 8800GT @ 1280x1024(my monitors max), max details, max viewing distance (this is a real killler, not max detail), AAx4 and AFx16. I get 40-50fps most of the time unless I'm viewing a area with 100+ objects such as a mountain side with hundreds of grass and tree objects. Surely if a 1066mhz FSB can bearly handle being in town then my 1600mhz FSB should die on most scenes with heavy grass/trees and many objects?
July 11, 2008 7:15:05 AM

MU_Engineer said:
3. The FSB actually was the bottleneck.


Oh ****! you don't say

How hard can it be to understand that the FSB is a problem on Intel motherboards ;) 
a b à CPUs
a b } Memory
July 11, 2008 7:26:06 AM

How like you Kassler. You take what he said, and completely misinterpret it.
July 11, 2008 8:35:14 AM

Kassler, you would love to meet this guy, he's called thunderman, and he continually goes on about how great AMD CPUs are and how the FSB is a bottleneck despite the fact that intel CPUs do better in most benchmarks... or wait, are you him?
July 11, 2008 8:49:18 AM

spuddyt said:
Kassler, you would love to meet this guy, he's called thunderman, and he continually goes on about how great AMD CPUs are and how the FSB is a bottleneck despite the fact that intel CPUs do better in most benchmarks... or wait, are you him?

hahaha, where is Self-criticism ;) 

The REAL problem is that there are some people that have problems to understand that Intel isn't best on everything. I find it rather amusing to see how long they can go to explain their feelings about Intel and how hard it is for them to understand technology.
July 11, 2008 9:20:56 AM

Kassler, most of us here will admit that the FSB is a limiting factor for Intel in certain aspect but we are talking about if the FSB is responsible for the problem the original poster claims it is. That is however doubtful otherwise AMD chips with HT wouldn't be as badly effect.

The original poster is claiming its the FSB when it clearly is not.
July 11, 2008 9:54:11 AM

MU_Engineer said:
I can think of three reasons that he may have had better performance <i>initially</i>:

1. Northbridge I/O bottlenecking being relieved by higher northbridge clock speed. Northbridge clock speed is the same as the FSB clock speed, so if there was an internal northbridge I/O bottleneck this would most certainly help. He has two GTX280s in SLi and the PCIe x16 slots come from the northbridge, not the southbridge, so it is certainly a possible culprit and this is the kind of result you would see if you were GPU-limited and your GPUs were I/O-starved.

After reading his edit, I am almost certain it is #1 if he is accurate with his "I set my memory to 266 MHz when the FSB was at 333 MHz" bit. It makes #2 incorrect and #3 unlikely. An FSB clocked at 333 MHz carries up to 12.7 GB/sec while dual-channel DDR2 at 266 MHz (DDR2-533) only outputs 8.5 GB/sec. The FSB in a Core 2 Quad *does* carry information both between dies as well as to and from the northbridge, but inter-die FSB congestion is unlikely. Why do I say that? He was playing Oblivion and that is not very multithreaded: http://www.tomshardware.com/forum/240359-28-core-quad-performance-oblivion That means that there isn't going to be a whole lot of cache coherency traffic between the two dies and thus the FSB is basically going to be used for CPU <--> RAM data transfer. We'd need to get a hold of his system and try a few things such as a single-die C2D and play more with FSB, northbridge, and RAM speeds as well as using a single GPU to accurately tell what's going on, but I *highly* suspect it has to do with the northbridge being swamped with PCIe traffic from the two GPUs.


That's what i thought too. The memory bandwidth is more than enough but i think the Northbridge can't switch fast enough between who has control over what area of the memory. Basically the cards are fighting over access to the memory which results in the northbridge ownership switching between CPU/GPU/GPU. This is quite sad since both GPUs actually request almost the same set of data which makes a lot of the switching redundant. Nvidia has already noticed the problem as they introduced "PWShort" and "Broadcast" (source) for their latest intel chipsets.

I think an AMD system would do better because it can keep the data flow up. This is not intels shortcoming though as they don't offer any multi-card solutions which actually create the problem.
July 11, 2008 10:13:47 AM

oh my god.....

Seriously, this "omfgwtfbbq FSB is the limiting factor on Intel CPU" thing needs to stop. Yes, FSB is the limiting factor on Intel CPU in server segments, but in desktop? Not so much.
July 11, 2008 10:33:00 AM

yomamafor1 said:
oh my god.....

Seriously, this "omfgwtfbbq FSB is the limiting factor on Intel CPU" thing needs to stop. Yes, FSB is the limiting factor on Intel CPU in server segments, but in desktop? Not so much.


WOW!! You are starting to understand ;) 
July 11, 2008 10:36:23 AM

Looking at this presentation
Yes FSB is a limiting factor
a b à CPUs
July 11, 2008 10:41:36 AM

If the FSB is saturated on a single socket Intel system then how come it still blitzes the phenom for every game at all ofthe decent resolutions for useful gaming?

Superior cache architecture, superior prefetch logic, superior transistor design for the cache - smaller, cooler, faster. lower latency for L1 and L2.

Plus it is wider issue and has a slightly higher IPC.

Why didn't you just say that MU ?

When your talking 3 or more sockets then the barcelona design which allows faster and more efficient inter-socket communication really pays off.

Why are we having this conversation ,,, Phenom is a good chip but it has limited headroom and when compared to an equivalent core2 duo or quad does not have the same IPC at the same frequency ... it is slower consistently by 9 to 13 % ... with the exception of a couple of areas of no concern to me as a gamer.

Kassler I have stuck up for you but clearly the link you posted is just crap ... interesting but crap.

cjl, MU and JDOCS are trying to be patient.

There is no point starting another flame war.

Can we ditch the rating stuff now ... it is getting annoying.

Here is something funny :

http://www.theinquirer.net/gb/inquirer/news/2008/07/09/...

I recon he had a Hdd full of decent pron ... what other reason would you go into a burning house unless it ws for your family??
July 11, 2008 10:51:03 AM

Slobogob said:
That's what i thought too. The memory bandwidth is more than enough but i think the Northbridge can't switch fast enough between who has control over what area of the memory.


In fact, bandwidth isn’t that important but this is how it is used to be described. The key to performance is less latency. And that’s what happens if you increase the speed of the FSB on Intel.
Memory bandwidth on AMD Phenom isn’t that big, but what they have done is to decrease the latency on the Phenom because there is where you improve speed. However this is a much more complicated discussion and it is difficult to sell products if it’s complicated to describe. AMD Phenom is more than twice as fast (less latency) as intel working with memory if memory is fast, that is also why AMD users may feel that their computers responds faster. Exactly the same impression that Intel users notice when they clock.
Intel has a very large L2 cache, and why is that good? L2 cache is very low on latency.
Processor cores are so fast today that they spend most of the time waiting for data.
Latency is very hard to measure though. It varies depending on what’s running. When the FSB is starting to be full of data then the latency goes up. Increase the FSB and the whole computer speed is increased.

July 11, 2008 10:56:25 AM

Why is C2D better than C2Q for gaming?
C2Q has a bit more latency when data travels through the FSB and that hurt performance. But the thing is that users seldom understand this and they think it is the same package with two more cores.
a b à CPUs
July 11, 2008 11:14:52 AM

You should be able to answer that question yourself ...

But you can't.

Why?
July 11, 2008 11:35:28 AM

Reynod said:
Superior cache architecture, superior prefetch logic,

But that also makes the FSB less effective because it need to handle more data than the processor wants. Also running a lot of applications or many threads this decrease performance because then cache is trashed. On reviews this isn’t noticed because then often run one single application.
Intel has tried to increase the cache as much as they can, using this they win two things. Less traffic in the FSB and L2 cache is fast. Perfect for single threaded applications and good for applications using two threads. Increase thread count and it isn’t that good anymore...
I think that there is a lot of Intel users that is going to be really disappointed on their computers when more applications will arrive that really takes advantage of threading.

July 11, 2008 11:41:52 AM

FSB is not saturated at all, there's plenty of room left. If you think FSB is a problem you're clearly not with it and need to do some more research.
a b à CPUs
July 11, 2008 11:48:05 AM

kassler said:
Why is C2D better than C2Q for gaming?

Because it's cheaper and it OCs better on motherboards under $600 :D 
a c 213 à CPUs
a b } Memory
July 11, 2008 1:44:09 PM

So even if a Phenom has the same problem it cannot be that the game is poorly coded or the game engine is just badly. I mean that would make sense really cuz it does happen even from the best game companies.

Here is the thing, every system has the same problems with this game. It is not the FSB or that crap you keep saying.
July 11, 2008 1:50:57 PM

Quote:
FSB limits performance on Intel?


Yes, in workstation and server applications as well as a small number of professional 'desktop' applications, the FSB based architecture will be a limiter - even for single socket systems this can be the case.


For the home user, it is largely irrelevant.

For the 2P workstation/server user, it may be critical, for the 4+P workstation/server, it almost certainly will be critical. (i.e. better off with AMD)
July 11, 2008 2:13:20 PM

wake me up when these fsb saturation test pipe dream is done on a ut3/modernwarfare engine game.*yaaaawwwwnssssss
a b à CPUs
a b } Memory
July 11, 2008 3:06:05 PM

kassler said:
Why is C2D better than C2Q for gaming?

Higher clocks and single threaded apps.
July 11, 2008 4:48:15 PM

kassler said:
WOW!! You are starting to understand ;) 


No, its that you're still delusional.

EDIT: You're still unable to show me a series of test where FSB is clearly limiting performance.

http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx...

As Anand's tests showed, very few applications take advantage of wider FSB bus. The ones that take advantage of it yield about 10% improvement. So, in short, 25% more bandwidth yields about 10% improvement. How does that prove FSB is limiting the performance.

You should really try harder next time :non: 
July 11, 2008 9:22:13 PM

more proof

http://news.firingsquad.com/hardware/ati_r...mance/page9.asp
We’ve excluded Radeon 4870 CrossFire results from this graph, as we couldn’t get the cards to completely scale in CoH. It’s not that the cards weren’t scaling under CrossFire, they certainly were, but their scaling performance was barely an improvement over the 4850 CrossFire setup. We’re confident we weren’t CPU-bound either, as we saw this occur under both 4xAA and 8xAA with max game settings. No matter what we tried the cards were only a couple of percentage points faster than 4850 CrossFire. We’re going to look into the issue further, we have a feeling we probably ran into a snag with the cards under Vista. CoH was the only game that exhibited this behavior with the 4870’s running CrossFire so we’re confident it was an isolated incident.

They used this hardware
Intel Core 2 Extreme QX9770

EVGA nForce 790i Ultra SLI motherboard (for GeForce cards)
ASUS P5E3 Premium WiFi AP Edition (for Radeon cards)
4GB OCZ DDR3 @ 1333MHz

compare to this because that was one game that intels had most problems with
http://www.overclockersclub.com/reviews/intel_q9450/14.htm

The cards can't perform (scale) because the FSB
a b à CPUs
a b } Memory
July 11, 2008 10:02:33 PM

No, they probably can't scale because of drivers.
July 11, 2008 10:13:21 PM

I would say its the drivers. Someone at the other forum also observed similar phenomenon, and his FSB was at 1600Mhz. At 4xAA, CFed 4870 actually performed lower than single 4870.

By the way, CoH seems to be optimized specifically for AMD.
a b à CPUs
a b } Memory
July 11, 2008 10:27:07 PM

Interesting note about CoH. I wonder how it will perform on Nehalem.

I run it on my laptop though, and with everything but DX10 cranked, it's perfectly smooth (DX10 kills it though from what I hear - I don't know, as my laptop is DX9 only)
July 11, 2008 11:11:00 PM

cjl said:
No, they probably can't scale because of drivers.

is it a driver problem on the other games as well then?
The cards differ more when used alone compared to using them in crossfire. Why is that?

July 11, 2008 11:34:04 PM

kassler said:
is it a driver problem on the other games as well then?
The cards differ more when used alone compared to using them in crossfire. Why is that?


By the same token, are you also claiming that Xfire performance increase should be the same across the board?
a c 213 à CPUs
a b } Memory
July 12, 2008 4:16:57 AM

kassler said:
is it a driver problem on the other games as well then?
The cards differ more when used alone compared to using them in crossfire. Why is that?


Yes. Early drivers have a lot of bugs and normally in Vista they are lacking.

Also did you not read the part where he specifically stated that they were not CPU bound?

Heck there are games that even with SLI/CF enabled there is no improvement.

Driver optimizations do not only concern the GPU they also concern how well the GPU handles that game in single and CF configs.

Just stop pls...... its no longer relevant to anything.
a b à CPUs
July 12, 2008 4:45:34 AM

Just stop wasting your time with this stupid troll
July 12, 2008 8:31:59 AM

jimmysmitty said:
Also did you not read the part where he specifically stated that they were not CPU bound?

No of course it isn't CPU bound, it's FSB limited so other hardware is going to sit and wait for data to arrive
      • 1 / 2
      • 2
      • Newest