Thursday, August 21st 2014
Intel Core i7-5820K Features Fewer PCI-Express Lanes After All
It turns out that our older report suggesting that the most affordable of Intel's new Core i7 "Haswell-E" HEDT processors will feature a slimmer PCI-Express root complex, even if it gives you 6 cores at a [hopefully] sub-$400 price-point, holds true, after all. Intel's wacky approach to its latest HEDT processor lineup was confirmed by leaked manuals of Gigabyte's socket LGA2011-3 motherboards, based on the Intel X99 Express chipset. The manual confirms that while Intel's $500-$750 Core i7-5930K and >$1,000 Core i7-4960X offer bigger 40-lane PCI-Express Gen 3.0 root complexes; the Core i7-5820K features a narrower 28-lane one. This means that multi-GPU configurations on systems running the chip won't be too different from those on LGA1150 "Haswell" platforms.
On motherboards with, say, three PCI-Express 3.0 x16 slots, the i7-5930K and i7-5960X will let you run two slots at full x16 bandwidth, and a third slot at x8. On systems with the i7-5820K, the second slot won't go beyond x8, and the third one will cap out at x4. On boards with four slots, one of them will run out of bandwidth. The trade-off for this narrower PCI-Express interface is the fact that you're getting six "Haswell" cores, twelve logical CPUs enabled with HyperThreading, about 12 MB of L3 cache, and a quad-channel DDR4 memory interface, at a price-point not too far off from the Core i7-4790K. So for enthusiasts with no more than two high-end graphics cards, the i7-5820K could provide an attractive gateway option to Intel's new HEDT platform. You can find the leaked manuals in this thread.
On motherboards with, say, three PCI-Express 3.0 x16 slots, the i7-5930K and i7-5960X will let you run two slots at full x16 bandwidth, and a third slot at x8. On systems with the i7-5820K, the second slot won't go beyond x8, and the third one will cap out at x4. On boards with four slots, one of them will run out of bandwidth. The trade-off for this narrower PCI-Express interface is the fact that you're getting six "Haswell" cores, twelve logical CPUs enabled with HyperThreading, about 12 MB of L3 cache, and a quad-channel DDR4 memory interface, at a price-point not too far off from the Core i7-4790K. So for enthusiasts with no more than two high-end graphics cards, the i7-5820K could provide an attractive gateway option to Intel's new HEDT platform. You can find the leaked manuals in this thread.
53 Comments on Intel Core i7-5820K Features Fewer PCI-Express Lanes After All
The fact of the matter is, if you don't actually need 6 cores, there has been little reason at all to use 2011. What did the inexpensive 2011 options offer before? Well, not even extra cores, because anything sub-$400 was a quad-core. The extra PCI-E lanes had almost no benefit to most HEDT users. Sure, you can run dual-SLI/CF with x16/x16, but that has no performance advantage over x8/x8. You can run triple-SLI/CF in x16/x16/x8, but again that has no real performance advantage over x8/x8/x8. The real advantage of 2011(and 1366) is the extra cores. If you aren't buying them for the extra processing power of the extra cores, then you are wasting your money.
Oh, and pay attention slick, you aren't paying more for less lanes. The 5820K still has significantly more lanes than the Performance platform. In fact it is offering a 75% increase in the number of PCI-E lanes over standard Haswell. So stop trying to be over dramatic.
So its a compromise with respect to 40 lanes but the entry level HEDT has always been a compromise in some way.
So its a compromise with respect to 40 lanes but the entry level HEDT has always been a compromise in some way.
all bitching over which way intel is really fucking us all :lol:
it matters not, we still get fucked!
In the end personally I would prefer the Quad variant with all the lanes if I was choosing this chip. But I can definitely see where they are placing this chip and it makes alot of sense in all honesty.
I will still be on the middle ground one in the end (The 5930K, but thats because I want it all).
28-16=12
12/16=.75
That's a 75% increase.
x58 - 36 pci-e lanes
x79 - 40 pci-e lanes
x99 - 28 pci-e lanes or 40 pci-e lanes
Getting charged more for less lanes [on the entry level] and a presumably low clocked six-core sucks plain and simple. Can't wait to see if Skylake brings a six-core to the mainstream, it won't arrive for desktop Broadwell.
I think most people who seriously take issue with the decisions that Intel has made here with the HEDT line never actually bought into the Sandy Bridge-E / Ivy Bridge-E LGA2011 platform and aren't actually candidates for such a platform to begin with.
I'll also point out though that there is room to recant. Intel can refresh the platform later with a full 40 PCIe lane part with 4 cores 8 threads at the same price.
Some people would probably still complain about that one too though.
As I said before the entry level HEDT option is and always has been a compromise. The mid range is a different kind of compromise as well because the previouse mid level options all had the same number of cores / threads as the extreme option. The extreme option is more or less a no compromise option ( other then compromising on paying top dollar).
Also no one is forcing you to buy a quad channel DDR4 $$$ kit with Haswell-E. If I buy into the platform I'll probably only buy a single 4GB or 8GB 2133 DDR4 DIMM to start off with while I phase in the system. I'll buy more RAM later as needed or due to declining prices
I probably paid something like ~$40 or ~$50 for my first DDR3 DIMM which I had before the release of the first Core i7 platform. I used that RAM initially with a Core 2 Quad Q6600 (not expecting any performance gains over DDR2) and then I used the same DIMM later with a Core i7 920 system (when released). Since the RAM is still in use I still consider it money well spent.
Imo Skylake-e 8core for mainstream 5820K price or bust - what with X179? chipset :D
Z170 is for LGA1151
en.wikipedia.org/wiki/Skylake_%28microarchitecture%29
1) Most games use significantly less than 4 cores effectively. That's if they even use multiple cores well when they are doing threaded loads.
2) Most games aren't CPU bound once you get to higher resolutions. Gamers aren't pushing for huge leaps in computational accuracy, they're pushing for huge leaps in graphics hardware to sate 4K monitors.
3) Haswell-e is functionally supposed to be an enthusiast level platform. Enthusiasts aren't all gamers, and the converse is true.
Based upon this logic, SB-e and IB-E are successes. They took their mainstream offerings, and added more of everything. If I needed a RAID card, I could pop one in. If I needed to pump out audio I had plenty of extra PCI-e lanes to do it. What I wanted from my computer was everything, and I could do it. It sucked that that functionality wasn't built-in, but when I wanted something I didn't have to compromise anything.
SB, IB, and Haswell mainstream boards fit their niche as well too. I might not have all the flexibility an enthusiast demands (and pays for), but I can do one task very well.
In steps Haswell-e's base offering. This is a bastardization. You've got less flexibility than the other Haswell-e options. What you receive in trade is more cores. I don't get it. More cores don't help the majority of games, if they did socket 2011 would have been used by more gamers. More cores do help compiling and video editing, but after you get two GPUs, a decent NIC, and a couple of other expansion cards (audio and USB3 or firewire), you're out of room. You can't fit a second NIC (think rendering farm), a RAID controller, and this is assuming that your GPU doesn't cover one or more of the PCI-e slots (unlikely at best with a double wide GPU).
You've given up a truly flexible rig for something with useless cores. That is why I think the 5820 is a poorly aimed CPU. You can argue semantics about how games are going to be threaded better soon, but Haswell is already having its threaded instruction sets pruned down. I guess show stopping known glitches will do that. Call me cynical and judgmental, but Intel seems to be polishing a turd here. Whatever bad mouthing there is for SB-e and IB-e, they were targeted at their niches well.
The 5820k is aiming at being a better overall platform than the 4790k (more cores, more PCI-e, and a bunch of new features). Intel won't cannibalize their 4790k sales, so the 5820k isn't going to be coming in at the $300 price point. Intel isn't that kind of stupid. I'm waiting for Intel to say the price point here is about $400-500. That means you've got the 4790 at $300, the 5820 and $400-500, and the 5930 at $600. Intel is going to differentiate the mainstream offerings from the enthusiast platform, and justify it with more cores. It's not putting forward the effort to make the differentiation between products better, it's making more shades of grey out of a relatively poorly defined black and white spectra which continues to shrink every generation.
However, I kind of consider the 4790k a little like the 2700k. A little tardy to the part and not very relevant with respect to its predecessor the 2600k.
Pricing and positioning will be known soon enough but history is the best indication of the future IMO which suggests ~$300. However, you very well could be right and ~$400+ could be more accurate.
I don't see it as a deal breaker either way but ~$400+ is close to the ~$500 I spent on my Core i7 3930K. Going up to ~$400+ on the entry level suggest going up on the other two offerings in the lineup as we'll.
It's a crowded market and intel is only competing with themselves.
There was a reason the 3820 and 4820k were priced at ~$300 USD and I don't see what has changed in with respect to the 5820k.
SB-e came to the party very late, and the 3820 was a locked chip. You had an unlocked chip competing with a locked one. The feature set distinction was clear. At price point $300 you got either a locked or unlocked quad core. If you settled with the locked chip you got a boat load of PCI-e. The markets were very clear, and the similar price point wasn't a deal breaker.
IB versus SB-e only increased the divide. That $300 still bought flexibility, or more speed.
IB versus IB-e is a repeat of SB versus BS-e, as Haswell versus IB-e is basically the same as SB-e versus IB.
Now welcome to Haswell. The differentiation is an unproven DDR, less PCI-e lanes than other Haswell-e offerings, but a couple of extra cores. Exactly how does that makes sense to put in place versus a similarly priced Haswell chip?
I'd put money on the assertion that if Broadwell released on time there'd be no discussion on this front. A 10% increase in IPC would have been enough improvement for Intel to have a 4 core Broadwell versus a 4 core Haswell-e. Instead, Broadwell development has stagnated and produced a true loggerhead. The enthusiast platform has caught up to the mainstream, so it isn't a previous architecture competing with a new one. Direct competition and Intel have been out of sync for nearly three years, so Intel is differentiating products by compromising features. They design two high end chips, disable chunks to make a range of products, and strategically try not to let the MSRPs overlap. Great for business, chip yield, and binning. Poor for consumers.
I'll put it a simpler way. You've got $500 for a system. Would you pay for six cores and new features, or four of the same cores and less features. This is why Intel would be peeing on their own feet if a 5820k was priced the same as a 4790k. It doesn't take a genius to see that, and Intel isn't a company of idiots. They'll artificially segment the market before competing with themselves.
Subb'd. ROFL.
The entry level quad core Sandy Bridge-E Core i7 3820 was locked (or partially locked as some have referred to it) and its successor the entry level quad core Ivy Bridge-E Core i7 4820K was unlocked and deserving of the "k" designation. Finally the 6 core Haswell-E 5820K will presumably also be deserving of the "k" nomenclature.
Each entry level part had a slightly different approach and could just as easily been the result of experimentation and uncertainty.
The Core i7 3820 was delayed and thus not released with the 3930K / 3960X. It also had a slightly different core design but my initial impression was that Intel was trying to decide how to best gimp the ~$300 product.
As for competing with ones self, if there is no competition from another company in this segment of the market then Intel is indeed competing with themselves. Beyond that point, the market is IMO a bit crowded and it's crowded with Intel parts. Perhaps Intel will discontinue some additional processors or sort it out some other way. Or perhaps they will simply allow some overlap in pricing, we will see but it only really matters to those willing to pony up for it,....to all else it's purely academic.
Also one small point, even if there were some overlap with the same pricing for two or more chips, the X99 / LGA2011-3 platform would likely still be more expensive based on X99 motherboard pricing and DDR4 RAM pricing. Given this likely eventuality perhaps price overlap on two or more different processors is a bit more forgivable from Intel's perspective.
Again we will see soon enough,...
As for clock speeds, it doesn't really matter, it is a K processor. If you don't like the clock speeds, change them.
Going by that assumption, you've still got to explain the 4820. The 3820 coming later to the party is a reasonable point to bring up. Intel deciding how best to gimp an already gimped chip (having two cores lasered off is gimped in my book) is nothing new for this industry.
I concede that this would be the logical point for a paradigm shift in what entry level enthusiast chips offer. Intel is switching up a bunch of what they are doing, so one more switch shouldn't matter too much. My problem then becomes Broadwell.
Let's say it's 6 months down the line. Broadwell on the mainstream platform launches. If the 5820 is still around and unlocked there are only two possibilities. The 5770 (or whatever they call the high-end mainstream offering) starts to compete with the 5820. Yes there is a generational gap, but most people will overlook a 10% increase in IPC for 50% more cores and more PCI-e.
So tell me, Intel is competing with a superior platform against their own less superior platform. The only way that could fly is if the 5770 were also 6 core, and even then it'd be a question of same price, but better features on the enthusiast platform.
Do you see how an unlocked entry level enthusiast platform price competing with the mainstream high end offering screws everything up that Intel has built over the past four years? Why would anyone at Intel believe this is a rational decision?
This is why I'm reasonably certain that entry level 2011-3 options aren't going to hit the $300 option. Spin it however you must, but they aren't going to scalp sales of their high end mainstream options in order to sell entry level enthusiast platforms. They've proven that they'll take the hit on delivered performance, in order to make sure they don't have to compete with themselves. Why would they sacrifice even more sales in the future?
I can see one situation in which I am wrong. They could pursue this pricing if 80% of their chips were binned with a slightly defective PCI-e root complex. Sell off all the defective units at competing prices, to prevent chip loss. They could do this while refining the process for better yields, and slowly phase out the 5820 as production no longer had problems with producing the large PCI-e root complex.
Of course, this assumption is based upon Haswell and Broadwell being a particularly bad turning point for the process of CPU development. The point at which we'll see an even more dramatic slow-down of performance gains per generation than we already have. Moore's law is great, but we're a bit behind on it. I guess Moore didn't really factor in the eventual issues of quantum physics...
On that note, I'm done. This debate is fun and all, but we'll see exactly what Intel has decided upon soon enough. To that end, you've got the last words in the discussion.