I had been thinking of switching out my P965 mobo for a P35 but reviews all pretty much state that there isn't much difference in performance. Then I find articles on X38 which say its not much better than P35. What's going on??? I don't plan to run more than 1 vid card so the X38 doesn't interest me but this lack of significant improvement in mobos bugs me. CPUs and vid cards get measurably better all the time but mobos seem to be stuck in place. Anybody got any ideas why?
Why so little improvement in mobos?
- Thread starter perdomot
- Start date
The P35 chipset works very well with all Intel CPUs. Easy 500MHz FSB with a quality MB. The only advantage with X38 is higher FSB speed (if you need more than 500MHz) and a small speed improvement with high performance DDR3 RAMs.
Hmm yeah I agree that in the MAIN areas motherboards aren't getting a LOT better.
But the reason for that architecturally is really Intel. They make the CPU and
they make (or at least DiCTATE) the motherboard chipset that is compatible
with the CPU.
The motherboard is basically nothing more than a glorified SOCKET for the CPU
that, oh by the way, gives you a place to plug in cards, USB stuff, drives,
and other such things.
The peripheral architecure -- PS/2 ports, USB 2.0, Firewire, IDE, SATA, AC97 sound,
etc. basically hasn't changed much in years. Those standards take like 5-10 years
to actually change meaningfully and even once new versions are agreed on they
take years to be adopted in motherboard chipset designs.
So the Intel Socket 775 CPU design hasn't changed much in years, the DDR2 RAM
design hasn't changed in like a decade, all the USB / Firewire / PCI / PCI-Express stuff
hasn't changed.
So really motherboards don't change MUCH other than getting SLIGHT incremental
improvements in speed or numbers of peripherals included just because there's
no other way FOR them to change and still be "PC compatible" and compete in the
same cost and form-factor market.
What REALLY needs to happen are some bold companies that are willing to
COMPLETELY redesign the PC form factor and platform / peripherals, but not many
try and not many of those succeed in the market. VIA's doing some cool stuff
with Mini-ITX / Nano-ITX PCs, companies like IBM or SUN or NVIDIA take
PC like technology and turn it into big clusters of super-computers or blade servers
for special applications like 3D special effects rendering farms, internet servers, etc.
I'd even settle for a BIOS that doesn't SUCK with tons of bugs on a motherboard but
it seems like even that is too much to ask.
As long as the focus is to save the most $0.00001 (NOT kidding) on each motherboard,
do NOT expect better QUALITY or FEATURES or DESIGN from the "next generation"
motherboard. It'll just be more of the same with slight tweaks Intel adds to make
sure there's a reason for people to buy a new CPU *and* motherboard if they
want to upgrade. Now we get to throw away all our old DDR2 memory and
buy DDR3 memory for our next generation motherboards. Why? Is it faster?
No, not really, and it won't be for several more years; DDR2 is just as good.
Is it cheaper? HELL no, it's like twice the price. Why do it? Intel decided to
force it on the market so it'll be compatible with the high end CPUs they won't
even start making for another 3-4 years.....
Just keep what you've got as long as it works.
But the reason for that architecturally is really Intel. They make the CPU and
they make (or at least DiCTATE) the motherboard chipset that is compatible
with the CPU.
The motherboard is basically nothing more than a glorified SOCKET for the CPU
that, oh by the way, gives you a place to plug in cards, USB stuff, drives,
and other such things.
The peripheral architecure -- PS/2 ports, USB 2.0, Firewire, IDE, SATA, AC97 sound,
etc. basically hasn't changed much in years. Those standards take like 5-10 years
to actually change meaningfully and even once new versions are agreed on they
take years to be adopted in motherboard chipset designs.
So the Intel Socket 775 CPU design hasn't changed much in years, the DDR2 RAM
design hasn't changed in like a decade, all the USB / Firewire / PCI / PCI-Express stuff
hasn't changed.
So really motherboards don't change MUCH other than getting SLIGHT incremental
improvements in speed or numbers of peripherals included just because there's
no other way FOR them to change and still be "PC compatible" and compete in the
same cost and form-factor market.
What REALLY needs to happen are some bold companies that are willing to
COMPLETELY redesign the PC form factor and platform / peripherals, but not many
try and not many of those succeed in the market. VIA's doing some cool stuff
with Mini-ITX / Nano-ITX PCs, companies like IBM or SUN or NVIDIA take
PC like technology and turn it into big clusters of super-computers or blade servers
for special applications like 3D special effects rendering farms, internet servers, etc.
I'd even settle for a BIOS that doesn't SUCK with tons of bugs on a motherboard but
it seems like even that is too much to ask.
As long as the focus is to save the most $0.00001 (NOT kidding) on each motherboard,
do NOT expect better QUALITY or FEATURES or DESIGN from the "next generation"
motherboard. It'll just be more of the same with slight tweaks Intel adds to make
sure there's a reason for people to buy a new CPU *and* motherboard if they
want to upgrade. Now we get to throw away all our old DDR2 memory and
buy DDR3 memory for our next generation motherboards. Why? Is it faster?
No, not really, and it won't be for several more years; DDR2 is just as good.
Is it cheaper? HELL no, it's like twice the price. Why do it? Intel decided to
force it on the market so it'll be compatible with the high end CPUs they won't
even start making for another 3-4 years.....
Just keep what you've got as long as it works.
Motherboards can't add performance capability that isn't there. Chipsets since Intel 975X have been good enough to expose 90% ~ 95% of the maximum performance inherent in contemporary technologies such as DDR2, PCI, PCI-X, PCI Express 1.x, SATA 3Gbps, USB2.0, 10/100/1000 ethernet, and so on. Motherboards and chipsets can't do anything about performance limits of current hard drive, RAM, GPU, and processor technologies.
So what you are in effect complaining about is that Intel has consistently developed well-balanced and highly efficient chipsets for the past couple years. That's a bad thing?
Its not as though Intel is somehow failing to deliver on increased processor performance, as AMD can't even keep up.
tcsenter the point that is being argured here isn't so much that intel has been providing bad products, but rather, they have failed to seek to improve - they are satisfied with mediocre, and we the consumers of computer hardware are the losers. Intel shouldn't allow themselves to be satisfied with being satisfied.
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,787
- 136
I sort of disagree. I think may be some sort of saturation of chipset performance here. New chipsets don't even bring 2% performance increase, half the difference of what the same generation 875P brought over 865PE. I wondered why they couldn't make better utilization of bandwidth. Dual channel DDR matched with 1066MHz FSB has theoretical bandwidth limit of 8.4GB/s. However, you can see in synthetic memory tests that it reaches only ~5GB/s. Perhaps using DDR2-800 rather than DDR2-667 raises it by another 100-200MB/s or so. Why is it so poor?
I thought it might be just Intel's inability to make better chipsets, until I think of chipsets from Nvidia/ATI. They don't do much better. Either they are all slacking off or there's a limitation of performance to be extracted.
Because of that, I am now viewing the worst R&D spending is trying to develop chipsets with higher memory performance. Useless, as some sites even show the "mighty" X38 chipset slower than P35 in all cases. One might think they might benefit more(at least Intel) by spending those extra R&D costs going nowhere(aka memory controller performance) into areas that are pitifully funded, like their integrated graphics.
I think what is lacking maybe a nice OCable board that has a strong on-board video. AS for features, the only thing I can think of may be to add something like a L4-cache or something. Or maybe extra cache for HD controller. Since HDs are so slow relatively, if the m.b. has an additional cache system could be helpful. Maybe on-board wifi but that's about it. It is kind of true that since P965 there has not been much speed improvements with subsequestn P35 X38 X48 etc. So I guess that's sort of what OP is complaining about.
A powerful GPU will cost more $ and generate more heat. A lot of users are perfectly happy with the slowest PCI-E or PCI graphic cards.
Integrated video will always have a demand due to value-conscious buyers.
Integrated video will always have a demand due to value-conscious buyers.
I thought we dealt with the question of whether Intel was somehow failing to release processors with respectable improvements? If that were so, why has Intel completely out-classed AMD at every turn for the past 18 months? There is another capable and innovative player here who should be able to capitalize from an Intel resting on its laurels. Mediocre is by definition relative to some higher standard or benchmark. What is mediocre about the best performing processors and chipsets on the market?
NVIDIA has grown an engineering team and culture that is very good at pushing the envelope (sometimes beyond what will reliably work). Is NVIDIA getting better DDR2 performance? USB2.0? SATA 3Gbps? PCI Express? Nope.
Intel can't make DDR2 work differently than it is designed to. It can support higher operating frequencies and lower latencies to the maximum effective limits of the technology (which have already been reached), but it can't make DDR2 transfer 'extra' data every clock than is defined by the technology.
Again, the complaint here in effect is that Intel is guilty of consistently developing well-balanced and highly efficient chipsets for the past couple years. For shame!
Intel 865P was Intel's first generation dual channel memory controller. 875P was second generation, or at least an improved/tweaked 865P, and still only managed to squeeze out an extra 5%. I will note that this yielded very good synthetic memory bandwidth scores, but in real-world performance benchmarks, VIA's single channel PT800 consistently performed within 5% ~ 7% of the best performing dual channel PAT-enabled 865P and 875P motherboards.
The additional bandwidth may have been there, but P4 Netburst couldn't utilize much of it.
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,787
- 136
So is the X38 chipset, a improved/tweaked variant of the P35 chipset. Rumors indicated radical improvements with a seperate die that's significantly larger than P35 chipset with all new designs required for mobos. Reviews have shown that such claims were bull****. I have seen sites where X38 chipset based mobos are slower than the P35 chipset.
Considering how performance leadership changes with 3-5%, it would have been good to see even that much of an improvement, since we all know that's pretty big for chipset improvements. However, its not faster. It's equal/slower than the P35 chipset. Ridiculous I say.
Regarding to PT800, that 5-7% difference in performance is significant difference. That's almost how much Northwood brought in average from Willamette. That's quite significant. Back when CPUs were so close together, its enough to make certain Intel processors beat AMD's.
If some chipset manufacturer manages to somehow make dual channel utilization go from current 50-55% to 80-85%, performance will improve noticeably.
X38 doesn't do that. It's not faster at all. Not even 1%. Average reviews of all X38 motherboards from all sites and it'll probably be ending up not advantageous at all compared to P35. To add insult to that injury, it doesn't even overclock better.
Well that's because most of the performance come from your devices running on the mobo, not from the mobo itself. If X38 improved performance by 50% given you have same cpu/mem/video..etc, it means your old mobo have been limiting your devices' performance by 50%, which is not a good thing.
When buying a mobo, it's best to see if the new feature offer something to you. For example, if you do SLI/Xfire, X38 give u more bandwidth. The new chipset also offers better compatibility to the latest cpu, with compatible bus speed. If you don't need any of that, stick to your old mobo.
I personally look for features, quality, layout, and maybe stability when OC, when buying mobo. Performance is really not a big factor since all most mobo with same chipset, or even different chipset like P965 vs P35, have very slim performance differences.
There have been some big improvements in motherboard cooling, such as all of the fancy heat pipes on many motherboards now. I don't know if those are gimmicks, or actually help. I picked my motherboard specifically beucase it had no northbride fan.
Still need cold air to hit the fins to remove heat. A much more elegant and cost effective solution is to use a good down-draft CPU cooler and an extra 92mm fan positioned above the NB heat sink.
There are always those who try to fill voids with ungrounded predictions, speculations, or wishful thinking, particularly among enthusiasts. Most of the time this stuff does not emanate from the company but from the land of zealous fanboy-dom. You throw those types a little crumb and they'll spin it into a smorgasbord of speculation and wishful thinking.
I was among those who, way back in August, took the position that X38 was not going to be the performance 'crown jewel' that others were anticipating:
I knew full-well there was not much more to be wrung out of 65nm Conroe/Kentsfield C2D and DDR2, with DDR3 bringing only modest gains largely in benchmarks heavily influenced by memory bandwidth (at two and three times the cost of DDR2). As we've discussed, a chipset can't 'extract' something that isn't there. Regarding X38 memory performance...
Intel has been moving toward more emphasis on DDR3 than DDR2. Although P35 was first to support DDR3, it was not slated to become Intel's flagship enthusiast/performance chipset, and lacked support for DDR3-1333 and DDR3-1600. When I read that Intel reportedly put considerable effort into optimizing/tweaking the memory controller on X38, I wondered to myself whether that effort was entirely focused on DDR3 performance, and if so, whether DDR2 performance might not be improved at all.
Let me rephrase for added clarity:
VIA's single channel PT800, found on motherboards priced between $50 ~ $70, performed within 5% ~ 7% of the fastest dual channel 865P and 875P motherboards priced between $175 ~ $200, in spite of a full 50% ~ 60% advantage in memory bandwidth for 865P/875P.
That's not as significant as you suggest, given it required spending at least 100% more to get that 5% ~ 7% increase. Numerous review sites concluded it was barely significant or not at all:
VIA PT800 vs. Intel 875P @ HardOCP.com
VIA PT800 vs. Intel 875P @ Legion Hardware
VIA PT800 vs. Intel 875P @ Tweak Town
VIA PT800 vs. Intel 875P @ The Tech Report
VIA PT800 vs. Intel 875P @ Hexus.net
From HardOCP's review:
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,787
- 136
You must not fully understand what I am talking about.
Core 2 Duo has 1333MHz bus in the latest generation. The real max throughput of such system is equal to the FSB, or 10.8GB/s(as long as the memory sticks are equal/greater than FSB's throughput).
Go see synthetic memory benchmarks. They are nowhere near 10.8GB/s. I see them at ~6500MB/s. Considering how before Dual channel memory, it had much higher utilization rate( 85%+) this seems there is still more to be extracted. 400MHz FSB with PC3200 got ~3000MB/s in memory bandwidth tests. That's very close to the theoretical max of 3.2GB/s. With similar utilization rate, we should see ~10000MB/s, not 6500MB/s. DDR3 isn't faster at all. Lower latency DDR2 modules have equal/better bandwidth. And that's on the X38.
There are 2 insignificant advantages X38 gives over P35:
1. It allows DDR3 to catch up with DDR2(P35 couldn't do that with DDR3)
2. It allows Crossfire performance to catch up with an old chipset, the 975X
You see, the X38 with DDR3 will bring no advantages with P35 and DDR2.
You surely are not understanding me here. Of course 5-7% isn't significant. People are spoiled nowadays. It used to be back in AthlonXP/Willamette, that was significant. It was enough to change lead back and forth between Intel and AMD. And from my calculations, 7% was what the Northwood brought over Willamette in overall performance increase. If someone used a Via chipset instead, they'll lose the performance gain that Northwood got over Willamette.
And for chipsets, which aren't known to make big gains, 5-7% is BIG.
And lets see that memory utilization again. I am not talking about the ability for the CPU to "utilize" that memory bandwidth. I am not stupid. I am talking about memory technology utilization, where the real life bandwidth measurements are in relation to the theoretical memory bandwidth.
Via PT800 with max bandwidth of 3.2GB/s: 3147MB/s
875P with max bandwidth of 6.4GB/: 4703MB/s
The theoretical max difference between two chipsets are 2x. The real bandwidth difference is only 49%.
The 3-5% per component in your system adds up to be 20-30% overall. Sure there is a price increase that doesn't seem to justify it. But lets look at it this way: X38 is an enthusiast chipset. For the average user these are the rich money wasters. They spend hundreds of dollars to get few % faster. This is why the ultra-high end video cards sell. If nobody thought like them, nobody would get the high end video cards and opt for 8600/HD 2600. Think of the other situations. Think of the world's fastest runners. They compete based on fractions of a second. Think of how much faster they are in terms of percentages, and its not too much considering all the effort went in to make them perhaps 60% faster than the average person.
Here's the exception though. The X38/X48 chipset. I was hoping for 3-5% performance increase. No, its not faster. Not even 1%. In other words, its worth it to no one.
No, we shouldn't. DDR2 has twice the latency of DDR at the same clock frequency. As bandwidth increases, utilization becomes more sensitive to latency. This relationship can't be changed. It can only be masked or offset by improvements elsewhere. Futhermore, your utilization rate assumptions are incorrect. Single channel DDR400 throughput on P4 Netburst achieved ~3100MB/s (95%), but when we moved to dual channel DDR400 on the same architecture, utilization rate dropped to 80% (~5000MB/s).
This hardly changed until DDR was paired with a new CPU architecture with latency advantages - AMD64 - which of course took the chipset out of the equation. With S939 AMD64, dual channel DDR400 throughput reached the same 95% utilization. When AMD64 moved from DDR to DDR2, utilization rate plummeted as it did when Intel moved to DDR2. Moving from DDR to DDR2 on AMD64 provided a significant increase in bandwidth but little performance gain.
In fact, DDR2 throughput on C2D is actually LOWER than last-gen P4 Netburst on the same chipset, as shown in this Anandtech article. And yet, C2D is vastly superior to P4 Netburst and AMD64 in putting that bandwidth to better use. Clearly, Intel made design choices in C2D that rather successfully traded utilization for efficiency (leveraged bandwidth). It can't just increase utilization without also impacting the trade-off in some way, until a major redesign in CPU architecture.
The pattern that develops from all this is clear: The reason we are seeing low utilization rate with DDR2 is due to current C2D architecture (and higher latency of DDR2), not the chipset.
This is true with 65nm C2D, but as I've shown, major changes in CPU architecture can change that. X38 was developed specifically for 45nm Penryn and DDR3. We will have to wait to see how X38 + DDR3 compares to P35 + DDR3 when both are running Penryn.
However, it isn't quite correct to say that P35 didn't enable DDR3 to 'catch up' with DDR2. Indeed, you imply that you would be satisfied if new chipsets were bringing 3% ~ 5% improvements in memory bandwidth. P35 alone delivered three times that much over P965:
So you could think of it as Intel skipped a chipset but more than made up for it in the next.
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,787
- 136
875P got 4600-4700MB/s. The later optimized motherboards brought more, but that's not really comparable.
Actually, Prescott had higher bandwidth results than Northwood. Prescott got more Write Combining buffers, which contributed to higher burst bandwidth, but that's really beside the point.
Show me in previous cases where it actually came true-new CPU brought more benefits. I haven't seen that.
I am not talking about 3-5% improvement in memory bandwidth, I am talking about average performance. And, X38 doesn't deliver anything over the P35.
I think you will see more improvemnts with the 45nm processors which will have an addittional 54 instructions for improvement to Video processing using SSE4 instruction set. Even if there is no remarkable improvement, the processor running cooler would seem to have a benefit for overclocking and overall quiet and cooler performance. For instance in line with this is the X35 or G35 chipsets with Integrated HDMI Video with Dual monitor support.
An example is the Asus P5E-VM HDMI motherboard with the GMA X3500 video which should be available soon. Having an HDMI port may be considered a benefit, but probably will not do much for gaming.
An example is the Asus P5E-VM HDMI motherboard with the GMA X3500 video which should be available soon. Having an HDMI port may be considered a benefit, but probably will not do much for gaming.
You also need to consider that the motherboard market is different in structure than the others. Where video and CPU tech are duopolies, the motherboard market is a polyopoly. While some manufacturers seem to stand out - like ASUS - it is really an even market with many evenly-matched competitors. What happens in this case is that manufacturers become risk-adverse, and each competitor is really just trying to meet established demand. Because the competition is so tight among many different manufacturers, any manufacturer who tries something revolutionary (like a new small form-factor gaming motherboard) risks complete bankruptcy. That's why components on motherboards are meticulously chosen to save that $0.00001, because every $0.00001 counts. That's also why we haven't seen the end of those blasted ugly serial/parallel/game ports.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter