Date: Friday , August 11, 2017
As is often the case with Intel, a new CPU usually comes with a new chipset launch. Intel's launch of its new Skylake-X and Kaby Lake-X processors naturally equate to the launch of a new chipset. Intel's X299 Express Chipset has been launched along-side these new CPUs. Was a new chipset necessary? On some technical level, I am sure that it was. If nothing else X299 brings Intel and its motherboard partners into parity with Z270 and adds HSIO architecture to the HEDT platform which is arguably the space that needs it the most.
Feature-wise, the chipset doesn’t provide that much of an upgrade over the venerable X99 Express Chipset. The reality is that we get very little with the new X299 chipset that we didn’t already have with the X99. The features we had with X99 mostly carry over at the expense of two SATA ports.
While I hope to provide some technical details here for inquiring minds this is more of an editorial on the stagnant nature of the HEDT platform and provide some possible answers to anyone who is asking if they should sell off their X99 setups in favor of X299. When there is a significant improvement in hardware, that question is more easily answered. A lot of our readers still rock chipsets that support their ancient Sandy Bridge CPUs (which still perform very well). With a modern graphics card, we’ve shown that these CPUs are still relevant (The Definitive AMD Ryzen 7 Real-World Gaming Guide). Even the improvements in subsequent chipsets never provided the clearly beneficial upgrade path that enthusiasts are looking for.
Intel’s doing some odd things with its LGA 2066 CPU lineup. Kaby Lake-X and Skylake-X launching at the same time for the HEDT platform is an unusual move. Intel’s also officially moved the top of the product stack into the stratosphere with a $2,000 Extreme Edition. Intel’s fastest CPU right now is the Core i9 7900X which runs right around $1,059 on Newegg at the time of this writing. That’s perfectly in-line with the Extreme Edition CPUs we’ve seen for years. However, it’s announced Core i9 7980XE will have an astounding 18 cores and 36 total threads for a whopping $2,000 price tag. I’m not shy about spending money on a CPU, but that’s getting ridiculous. I don’t begrudge anyone who can afford a two million-dollar car for buying one, but this isn’t a product of superior craftsmanship or something with real collector’s value. This is a CPU. Functionally, in games it won’t provide anything over the 10 core, 20 thread i9 7900X today or anytime soon. Intel is getting as lethargic with its marketing and product branding as it is with its IPC gains.
As a guy who's spent almost 20 years in the IT industry, I'm always going to go towards an option that provides more CPU cores. Yet, I skipped Intel's 6950X and I've stuck with my 5960X since that product launched. Why? Broadwell-E offered little to nothing over Haswell-E's IPC. Sure, you picked up two more cores but those won't help you eek out anymore FPS over the older 5960X. Those two extra cores and almost non-existent IPC improvements weren’t worth the $1,700 price of admission. Broadwell-E also doesn't overclock as well and after degrading 6950X's max clock speeds on my test bench I wasn't about to take that plunge. I'm not frugal, but I don't have more money than sense either. I quickly discerned that Broadwell-E had nothing to offer me so I skipped it. Skylake-X is potentially compelling, but anything beyond 10 or 12 cores at this point would be wasted cash.
What about Skylake-X? On the surface, there is something to a potential upgrade to Skylake-X. The $1150 Core i9 7900X for example has it all. It offers more cores, better clock speeds, and more PCIe lanes. It's a bit power hungry of course and it does run hot. I'm no stranger to those issues and frankly, I do not care about such things. Where Skylake-X really shines for gaming comes down to potential for higher performance due to the architectures scaling with faster memory speeds. It’s around the price I normally spend for CPU’s and while it’s not the fastest on the block, I think we’ve reached a point in the HEDT space where going for maximum core count isn’t the way forward. We are going to want a balanced approach between cores and clock speeds. The TDPs of CPUs with 12 or 18 cores will be insane. We have already seen Kyle having plenty of issues dialing in an enthusiast overclock on the $1000 Threadripper 1950X due to heat. Kaby Lake-X in its current form is stupid, so I won’t talk too much about it. 5.0GHz+ on such a relatively cheap processor and having Intel’s strongest IPC may be compelling for some. I wouldn’t argue that outside of the fact that the chip is crippled. Most of the reasons to buy into HEDT are locked away when pairing a motherboard with 8x DIMM slots and up to 68 PCIe lanes with a Kaby-Lake X CPU.
So, let’s talk about the actual X299 Express chipset and how it compares to the outgoing X99 Express chipset. X99 launched in Q3 2014. It lithography is 32nm and it has a TDP of 6.5watts. It's has a 25mm x 25mm package size. It supports only 8x PCIe Gen 2.0 lanes. It supports a total number of 14 USB ports natively. Only 6 of which can be USB 3.0 without added hubs or controllers. It does support upwards of 14 USB 2.0 ports. Beyond that X99 has the usual integrated gigabit Ethernet and is constrained by DMI 2.0. This, and the limited 8x PCIe 2.0 lanes were the real problems with X99. When the platform launched it was only at relative parity with Z97 and it quickly fell behind Z170.
X299 on the other hand features full PCI-Express 3.0 support and offers 24x HSIO lanes and up to 24 PCIe lanes. For those that don’t know, HSIO adds flexibility on the back end by allowing motherboard manufacturers more flexibility with lane allocation and integrated devices. They can allocate lanes in blocks or groups of 4x PCIe lanes to be used for things like up to 3x M.2 slots and other things. X99 didn’t do that and relied on expensive PCIe switches. The most flexible solutions used PEX PLX8747 chips to offer the greatest amount of flexibility but these chips aren’t cheap and have actually become more expensive as of late.
The X299 chipset is built on a 22nm lithography and as a result has reduced its TDP to a mere 6 watts which doesn't sound like much, but is a reasonable reduction in power considering how little either uses. The package size has been reduced from 25mm x 25mm to 24mm x 24mm. Again, this doesn't sound like much, but it can be a big deal as real-estate on some of these motherboard PCBs is at a significant premium. USB support has also improved as X299 can now handle allocating up to 10 of its 14 total ports as USB 3.0. We still don't see USB 3.1 capability which means a native Intel approach still requires Alpine Ridge to be paired with X299. This is not an inexpensive solution. One area where X299 backslides a little is in its SATA support. X299 only supports 8x SATA ports instead of 10x like X99 does. I'm not sure of the reasoning behind this but I doubt most people will care. I'm using 8 or 9 ports in my own machine, so this is a knock against X299 as far as I’m concerned. Albeit a minor one.
Two more improvements in X299 over X99 are Optane support and DMI 3.0. DMI 3.0 gives you double the interconnect bandwidth between the CPU and the chipset which is a good thing. It’s effectively a PCI-Express x4 link that’s a dedicated path between the CPU and the chipset. I've argued before that this interconnect was already anemic at launch, but that's beyond the scope of this article. I'll summarize things in order to keep things brief: Theoretically, you can easily max out DMI 3.0 doing something as benign as benchmarking a RAID 0 M.2 NVME SSD volume. X299 doesn't change that. Add in some file transfer in the background or a few other activities and you can reach saturation of this bus easily. Now, I don't think that's realistic as you won't be hitting these limits often, if ever. DMI 2.0 on the other hand, is asthmatic compared to DMI 3.0 so score another point for X299.
I don’t truly have a solid conclusion on the chipset itself. If you want to get technical, X299 is an upgrade in almost every way over the outgoing X99 Express chipset and for many people that’s good enough. Still, we are talking about something that is clearly evolutional and not revolutionary. DMI 3.0 and more PCIe lanes are the largest benefits I see to X299’s design over its predecessor, X99. At least from the consumer’s perspective. HSIO is a major benefit on the back end, but that’s a benefit to implementation and design. The added PCI-Express lanes are nice, but that opens another can of worms. With talk of multi-GPUs demise around various tech sites, including our forums it’s hard to see the justification for so many PCIe lanes on the consumer side of things. You get upwards of 68 PCIe lanes on X299 with the right CPU which sounds like more than enough for even two or three-way GPU based systems. That's a significant jump ahead of X99 which had a maximum 48 PCIe lanes to offer. Score another point for X299.
Even in the HEDT market, it’s getting harder to justify multi-GPU systems but again there are things like electronic currency mining and compute type tasks which can leverage multiple GPUs even if SLI and Crossfire are off the table. Let’s also not forget that DX12 and Vulkan support multiple GPUs, so support for these might simply be evolving beyond the hands of its creator's and into something tackled by the industry in a different way apart from the GPU manufacturers. That's something I'd like to see but then again, we have little need for multiple GPUs outside of VR applications. Even at 4K, I can drive my system off a single GTX 1080 Ti and max out most games' IQ levels. It's even possible we could see a return of systems that have more than two GPUs inside. This is all wishful thinking on my part so I can't say for sure whether or not the industry will actually embrace these technologies.
X399 and Threadripper are worth considering as well, but I haven't any hands-on experience with that platform as the time of this writing. I'll probably follow up on this once I do. In the interim, I think Intel's built a good platform once again but as usual, it's an incremental step just as we've seen for the last five or six years. Back in the day when games would evolve in significant steps each time, I felt compelled to upgrade just to get good performance in games. Often, I'd be reaping benefits in a single game for one or two upgrade cycles as technology matured. These days games are multi-platform friendly and the need to do that has diminished to the point where an ancient Core i7 2600K can keep up so long as it is paired with a modern GPU. I'm not sure I'll be parting with any cash this time around. That's how little I'm impressed over all.
There are a lot of comments about Intel having knee jerk reactions to AMD's Threadripper and I think there is some truth to that. That said, I talked with Intel about X299 well over a year ago and so far, in my hands X299 has felt as mature as anything Intel's launched in the last few years. The only real concern I have for it revolves around specific motherboards using single 8-pin CPU power connectors for power hungry chips and questionable VRM cooling. There are a lot of people in the industry expressing such concerns and I am in the same camp.
I can say that X299 is a definite improvement over X99, but where the rubber meets the road it's not a reason to upgrade on it's own. If this platform has any real value as an upgrade, it will be on the merits of the processors that will inhabit motherboards using this chipset. I'm somewhat skeptical given how little Skylake and Kaby Lake brought to the table. The X versions are cool, but I don't think they are necessarily worth the price of admission over existing X99 systems. We will ultimately find out how things shake out over the next couple of months or so. I’d still most likely be all for an upgrade consisting of X299 with the right CPU, but I think I’ll wait on Threadripper before I can ultimately decide on where my hard-earned money will be headed next. I think you should probably wait it out as well.