taltamir
Lifer
- Mar 21, 2004
- 13,576
- 6
- 76
We don't need any more at the moment.
to put things in perspective... remember the rediculous NF200 chip nvidia makes? that gives more PCIe SLI lanes? that is ALL the X58 chip actually does... it is an NF200 chip from intel. everything is else is on the southbridge or the CPU... the i5 architecture gets rid of that chip... P55 is just a renamed southbridge and the CPU is now handling the pcie.
And honestly, who the hell needs two pcie v2 16x lanes? i am fine with single slot cards, or with multi gpu of cheap cards... only people doing quadfire setup really benefit from the X58 setup. and they get the "we screw you" tax from intel.
I don't know ANYONE who has 4 AMD gpus in one computer, so I don't know anyone who could even USE more than 16 lanes.
16 lanes is the perfect amount for users with single-card setups, and will be fine for years as long as you only use one card.
But why so much hyperbole in your post? Far less than quadfire will saturate an 16x PCI-e connection. That means each GPU only gets 4 lanes! There is a very good reason people want to run their PCI-express slots at full speed.
This review tests two low-end 4850's in crossfire on a P45 and X48 motherboard, and you can see that in high resolution gaming there are considerable compromises when you start halving bandwidth, particularly in crysis.
http://www.tweaktown.com/artic...erformance/index4.html
If you never plan on running multiple GPUs, then you'll be fine with a P-series chipset. However, taking a configuration that works for you and widely applying it to other people's systems is just wrong. This is what idontcare has called forced market segmentation. If you want full-speed multi-GPU (and with dx11 cards this may be even more important), you will need a socket 1366 system.
But why so much hyperbole in your post? Far less than quadfire will saturate an 16x PCI-e connection. That means each GPU only gets 4 lanes! There is a very good reason people want to run their PCI-express slots at full speed.
This review tests two low-end 4850's in crossfire on a P45 and X48 motherboard, and you can see that in high resolution gaming there are considerable compromises when you start halving bandwidth, particularly in crysis.
http://www.tweaktown.com/artic...erformance/index4.html
If you never plan on running multiple GPUs, then you'll be fine with a P-series chipset. However, taking a configuration that works for you and widely applying it to other people's systems is just wrong. This is what idontcare has called forced market segmentation. If you want full-speed multi-GPU (and with dx11 cards this may be even more important), you will need a socket 1366 system.
There was a review that showed a GTX 280 gaining FPS going from a PCI-E 1.0 to 2.0. One G300 will come close to maxxing out a 2.0 slot if the rumors are true. There should be 3.0 spec motherboards next year. Plus people don't generally upgrade CPU's every year so that bandwidth needs to last 2-3 years or more. Bottom line is you can never have too much bandwidth or speed or power...only for the moment.
Zensal
Senior member
- Jan 18, 2005
- 740
- 0
- 0
No one will want to run Quad SLI/CF on a P55 mobo. The same issues exist on P45 but most people don't care because they only run 1 card.
But 2 low-end cards like the 4850's in your link are still perfectly good cards in CF at or under 1920x1200. A vast majority of people on these forums and elsewhere use monitors at or below 1920x1200.
But 2 low-end cards like the 4850's in your link are still perfectly good cards in CF at or under 1920x1200. A vast majority of people on these forums and elsewhere use monitors at or below 1920x1200.
VirtualLarry
No Lifer
- Aug 25, 2001
- 56,587
- 10,227
- 126
I use four 9600GSOs in one PC, in a Folding@home box. Granted, I'm not running SLI, but the 4x8 lanes of PCI-E on my K9A2 platinum mobo (790FX, I think) help a lot.
VirtualLarry
No Lifer
- Aug 25, 2001
- 56,587
- 10,227
- 126
OILFIELDTRASH
Lifer
- May 13, 2009
- 12,333
- 612
- 126
:disgust:You know the more I think about i5 the more I don't like the whole thing. First off why not just have one platform like 775. You could get yourself a decent mobo and cheap cpu. Later when you could afford it or needed the extra performance you could get a bad ass cpu then instantly you have a sweet system. So now your telling me if I want the best performance I'm gonna have to buy $1000 cpu's. WTF. I now am being priced out of top tier computing. Not because I can't afford it but because to me $1000 cpu prices are ridiculous for something that does not contribute to my livelyhood. It looked like i7 was heading in the right direction. Mobo's were coming down in price. The cpu's were reasonable. Now if I buy an i7 platform(even though its affordable now) later on if I want a better cpu I'm gonna shell out close to a thousand dollars. Mobo prices will also be incredibly high as mentioned in articles since the new i7 buyers will be high dollar spenders.
Intel is just getting greedy. Hell it seemed they were doing pretty well. They just got fined for billions of dollars and it still hasn't affected day to day operations.
I'm seriously considering another hobby. I will not be intel's bitch.
Intel is just getting greedy. Hell it seemed they were doing pretty well. They just got fined for billions of dollars and it still hasn't affected day to day operations.
I'm seriously considering another hobby. I will not be intel's bitch.
- Sep 28, 2005
- 21,128
- 3,659
- 126
taltamir b4 u continue with your debate, have you seen an i5 overclocked?
Can they even overclock? i can care less about turbo on if i cant overclock the cpu.
also rumor has it they have a massively high QPI which makes QPI overclocking near impossible.
Can they even overclock? i can care less about turbo on if i cant overclock the cpu.
also rumor has it they have a massively high QPI which makes QPI overclocking near impossible.
You know if AMD had anything competitive to offer, this kind of gouging wouldn't be happening. Will AMD have an istanbul type chip for the desktop? It feels like we are back to the pentium II days vs k6-2 or k6-3.
775 isn't enough pins for all the components they added on the die of the i5/i7 (integrated memory controller etc.). You don't have to buy the $1000 6 core CPU. The 4 core i7 should be plenty until Sandy Bridge comes out next year and its mainstream variants in 2011. There should be a $300 version then also. You also forgot about overclocking so you can buy a $250 CPU and make it faster than a $1000 CPU.
Zensal
Senior member
- Jan 18, 2005
- 740
- 0
- 0
What OILFIELDTRASH is trying to say is that instead of having 2 different sockets and 2 different platforms you should just have LGA-1366. Just like 775 was. You still had the P45 vs X48 but the processors were drop in upgrades. There was no technical reason that it couldn't have been done the same way as 775.
I guess it could have been like that. X58 offers much more than X48 ever did and it is cheaper to go from P55 to X58 than P45 to X48. I had a P35 and couldn't justify upgrading to P45 or X48 because there weren't any real features to justify it. With X58 you have QPI, triple channel RAM and SLI for only a $40-$50 premium. Plus I'm sure there will be many more OC'ing options and i7 will more than likely OC more than i5. X48 offered none of these things for what it cost except for more BIOS options and 2.0 PCI-E slots. There are better options this time so Intel decided to make it a separate platform. I am just happy they made the 920 at all.
Even if Intel made the i5 (lynnfield) to be socket-compatible with the i7 (bloomfield), the motherboard itself would not be compatible with both CPUs as the supporting chipsets for the CPUs are different.
Imagine the customer confusion when they have to figure out whether their LGA1366 motherboard has the right chipset to support Bloomfield (communicates via QPI with the X58 chipset and PCIe lanes from there) or whether the motherboard contains all the electronics and path traces needed to support Lynnfield (on-die PCIe, communicates externally via DMI link to P55 chipset)
Even in the day of LGA775 the server world had LGA771. Xeons were made for both sockets. All Intel did with bloomfield is reduce the supported socket count from two to just the one. Now both desktop and server sockets for bloomfield uses the same socket LGA1366 (not talking about Beckton, which is not bloomfield, just talking bloomfield here).
So now in world of nehalem we have two sockets still, LGA1156 and LGA1366. I don't get what the "beef" is with this. The two chips are electrically incompatible with each other as they communicate with the motherboard differently for a very good reason...bloomfield is meant to be multi-socket capable and that requires QPI. Lynnfield is meant to be single-socket and aimed at people who care about things connected to the PCIe (i.e. gamers and their GPU's) so its tailored for that purpose.
Zensal
Senior member
- Jan 18, 2005
- 740
- 0
- 0
I'm just saying why not use QPI on Lynnfield also? Intel could have limited the bandwidth between the CPU and the IOH to make the P55 chipset less attractive (and thus cheaper) then the X58. Lynnfield could also still use dual channel memory instead of triple channel. And the IOH with QPI is not expensive to manufacture. Intel prices the X58 chipset artificially high.
taltamir
Lifer
- Mar 21, 2004
- 13,576
- 6
- 76
the problem with your argument is that X58 does not give any more bandwidth for one slots... only if you have TWO slots than it gives the same bandwidth as one slot, while the P55 gives half bandwidth to each slot.
So, 2 years down the road, you buy a GTX580 that needs pcie v3... and compare its performance on a single slot P55 mobo to a single one of those cards in an X58, it will be identical...
but if you try to run two GTX580s in SLI (Assuming you even COULD) you would have slightly less atrocious performance loss due to bandwidth in the X58 mobo... however anyone interesting in running two top end cards in SLI 2 years down the line is gonna be upgrading his mobo and CPU to something that does PCIe V2...
How many people today are bemoaning their v1 PCIe 8x SLI mobos, saying they should have gone with 2x16x PCIeV1 so that they could SLI two GTX275 cards without a performance hit? You are assuming someone who pays ridiculous amounts of money for top the line multi card solutions would use an outdated mobo and CPU and as such would be better off paying hundreds of dollars more for it today for "future compatibility".
I see what you are saying now...its not a "why release LGA1156 CPUs instead of making them all be LGA1366" debate, its a "why create an entirely new electrical interface thus negating the possibility of socket compatibility for the two product lineups?" discussion.
I agree. It seems entirely needless to create DMI. Why not just keep using QPI to the chipset and then create artificially castrated chipsets to further enforce your market segmentation?
The value-angle of DMI and LGA1156 versus QPI and LGA1366 topologies has not been communicated very well to the consumer.
Not going to dispute that as I haven't seen any BOM cost breakdowns or analyses.
But I would like to see one, so I'm compelled to ask you "how do you know?" in hopes of a link or something coming out of it so I can educate myself.
That statement makes little sense to me, so I guess I am missing some technical documentations on the DMI specs that would make it self-evident to me as to why DMI is incompatible with QPI.
Why (or rather, how) is DMI incompatible with QPI?
QPI is needed for 2S and more. I'm quite confident in my understanding of the how's and why's of QPI...it's the DMI angle I am not so well versed in.
It's been made quite clear that Lynnfield uses DMI, and thus LGA1156 cannot be made to be electrically compatible with LGA1366 mobos and hence no reason/need to make it physically compatible either...but what hasn't been well communicated is why Lynnfield requires DMI.
I don't see a need for two sockets, I see it as a convenient way to enforce market segmentation. (which I'm not against, but lets call a spade a spade and not insult our intelligence about it in that case)
taltamir
Lifer
- Mar 21, 2004
- 13,576
- 6
- 76
Anand's lynfield article said so.
When I say non compatible I mean you can't just replace the DMI for QPI and have it on the same exactly electrical connections, you would be putting a DMI communicating CPU into a socket that transmits it to the X58 northbridge which expects QPI and QPI only. there is just no way that the X58 can just accept that and work.
If it is abundantly clear then why ask? this is what I was referring to.
Anand said cost.
QPI can do 1S to 8S specifically. QPI enables up to 8S systems.
Anand's article spoke about platform/chipset pricing, nothing about interface topology costs. I just read it again to confirm my initial impressions weren't misguided. He's quite clear about it with repetitious use of the wording.
Further he said Intel prices the x58 at a premium, not that it cost a premium to implement/produce/manufacture. He further stipulated that Intel's decision was to not lower the price on x58 but rather to introduce a lower-price DMI-based product (again, emphasis is on price to the consumer, i.e. market segmentation, not cost or gross margins impact on Intel's side of it).
I was giving you the benefit of the doubt by assuming you were talking about there being some technical reason (unbeknownst to me) why a topology can't exist with both DMI and QPI implemented in parallel...it is quite clear they are not electrically compatible hence my confusion on why anyone, yourself included, would bother stating something that simply need not be stated? I was thinking there was meat to your statement, my error.
To be pragmatic about it, x58 not accepting DMI inputs is simply a matter of engineering/management decisions during the design phase, i.e. the decision was made to have the x58 not support DMI.
It still does not speak to the "necessity" of DMI. P55 could have communicated with lynnfield over QPI just as well. I'm not seeing innovation here, I am seeing marketing winning again and it makes me shudder because this has subtle characteristics of the old Intel resurfacing in regards to how the Pentium4 came about based on marketing dictating the design directions rather than the engineers.
QPI is flat-out simply required for 2S and higher. The memory topology of Nehalem requires it.
It is not a case of being "really more for 8 socket systems" as you originally stated.
It is required 2S and higher. QED.
It is also not intrinsically limited to 8S (i.e. it is not simply 1S to 8S specific as you state), it varies by implementation. The EP's QPI supports up to 4S while Beckton's QPI implementation supports up to 8S. QPI can also support much larger systems as well...Tukwila uses QPI for example and current Itanium systems scale to much more than 8S systems so I expect tukwila to scale equally in sockets and that will require even higher QPI counts.
From what I can tell, Lynnfield could have just as easily used QPI to talk with P55 and the QPI bandwidth could have been crippled to some paltry 2-4GB/s bandwidth and still have met the objective.
So why DMI? Still no technical answers, just marketing ones IMO.
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: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter