Intel’s Roadmap: Ivy Bridge, Panther Point, and SSDs
by Kristian Vättö on May 6, 2011 1:05 AM EST- Posted in
- CPUs
- SSDs
- Intel
- roadmap
- Ivy Bridge
- Panther Point
Last week, we told you about Sandy Bridge-E and X79 chipset. Today, we have a lot of interesting news about other Intel products, including a look at the Ivy Bridge platform and upcoming SSDs. Intel still isn’t taking the wraps off of their Ivy Bridge architecture, but yesterday’s 3D Tri-Gate announcement certainly changes the expectations.
Ivy Bridge
Ivy Bridge will be a die shrink of Sandy Bridge and represents a “tick” in Intel’s product line. That means the micro-architecture is mostly the same as Sandy Bridge, but it will be manufactured using 22nm process instead of 32nm. That will bring lower power consumption and thus less heat while keeping the same performance level. Unfortunately, we don’t have any information on core counts, clock speeds, model numbers, cache sizes, or the IGP. At the same core counts as SNB, we might see some fairly high clock speeds (>4GHz, anyone?) since current quad-core Sandy Bridge CPUs already offer Turbo Boost up to 3.8GHz, and adding additional cache to the design is almost a given. Intel is putting additional resources into their IGP as well, so we expect to see some healthy performance and capability improvements.
| Sandy Bridge versus Ivy Bridge | ||
| Sandy Bridge | Ivy Bridge | |
| Manufacturing process | 32nm | 22nm |
| Transistor technology | 2D (Planar) | 3D (Tri-Gate) |
| PCI-Express (version) | 16 lanes (2.0) | 16 lanes (3.0) |
| Turbo Boost version | 2.0 | 2.0 |
| Memory support | Up to DDR3-1333 | Up to DDR3-1600 |
| Quick Sync | 1.0 | 2.0 (?) |
| DirectX | 10.1 | 11 |
| IGP shader count | Up to 12 EUs | Up to 16 EUs (?) |
| OpenGL | 3.0 | 3.1 (?) |
| OpenCL | N/A | 1.1 (?) |
The table above summarizes most of the currently known differences. Ivy Bridge will have enhanced AVX support, the on-die PCI-Express graphics links become version 3.0 instead of SNB’s 2.0, and official memory speed support from the IMC gets bumped to DDR3-1600. While we won’t know about the power consumption until we actually get to test an IB CPU, the roadmap lists TDPs that are the same as SNB (95W, 65W, 45W, 35W). The shrink to 22nm and 3D transistors (FinFET) almost represents a two-node process technology jump, so we expect performance at various power levels to increase quite a bit. A final interesting point for many users is that Ivy Bridge is pin compatible with Sandy Bridge, and it will work on current LGA1155 motherboards with the appropriate chipset and a firmware and BIOS update (H61, H67, P67, and Z68 are capable of support IB). Intel will also launch new 7-series chipsets, which we’ll get into below.
Wrapping up the discussion of improvements, let’s focus on the IGP a bit more. As with Sandy Bridge, we expect Intel will have several IGP variants with Ivy Bridge’s graphics. We don’t know what they will be, but we do know that Intel is calling it their “next Gen Intel HD Graphics” and the core GPU will be DX11 capable. It also looks like Intel will add OpenCL 1.1 support and increase the maximum number of EUs from 12 to 16, though either of those elements may change. Intel lists “Next Gen Quick Sync” as another feature, and with the increase in EU count and additional functionality Ivy Bridge might be double the speed of SNB when it comes to transcoding video.
Facebook
Twitter
RSS
74 Comments
View All Comments
DanNeely - Friday, May 6, 2011 - link
looks like it's not going quite as fast as was expected; prior rumors had it stripped from the x79 chip too.OTOH if filling in all the expansion slots in an ATX board would've been tricky with x79, it's going to be even worse for enthusiast z7x boards unless the number of lanes on the southbridge is increased, especially if DMI isn't given a boost to allow the SB lanes to run at 3.0 speeds as well. 2x gb nics, and 3rd party audio will take 3 lanes. Additional usb3 and sata 6gb controllers will take 1 each. That leaves at most 3 SB lanes to make 1x slots. The situation is worse if the board makers insist on adding legacy firewire or pata ports. Thunderbolt ports will IIRC eat a PCIe lane each as well. I suspect 4 to 8 lane pcie bridge chips (or possibly PCIe to PCI bridge chips) will end up being fairly common on these boards.
JarredWalton - Friday, May 6, 2011 - link
Do the current 6-series chipsets connect to the CPU with a 20Gbps DMI link, which isn't even enough for all the potential peripherals if they're all maxed out, but in practice that almost never happens.8x PCIe 2.0 lanes = 40Gbps
2x SATA 6Gbps = 12Gbps
4x SATA 3Gbps = 12Gbps
14x USB 2.0 = 6.72Gbps
Gigabit Ethernet = 1Gbps
--------------------------------
Total Potential = 71.72Gbps!
I believe the 7-series is still using DMI 2.0, so it remains at 20Gbps, but now the total potential bandwidth for all ports is:
8x PCIe 2.0 lanes = 40Gbps
4x USB 3.0 = 19.2Gbps
2x SATA 6Gbps = 12Gbps
4x SATA 3Gbps = 12Gbps
10x USB 2.0 = 4.8Gbps
Gigabit Ethernet = 1Gbps
--------------------------------
Total Potential = 89Gbps
Realistically, the SATA ports are pretty much never going to all be 100% utilized at the same time, and the same goes for the PCIe ports and USB ports. It's possible Intel will tweak the DMI interface to boost bandwidth as well, which could easily accommodate the additional bandwidth heavy devices.
DanNeely - Friday, May 6, 2011 - link
I was looking at the number of devices to connect, not total bandwidth (pcie to pci or pcie-pcie (more lanes) bridge chips don't help with that.My observation is that after you connect all the devices to the PCIe bus that need to be connected, you don't have enough lanes left to fill out all 7 slots on an enthusiast level full atx board. Depending on the chipset you have anywhere between 1 and 3 slots off the CPU, and at most 3 1x slots off the southbridge. With thunderbolt potentially eating a 6th, and a few of the 6x series boards having enough USB3 controllers to have 8 or 10 USB3 ports, it's entirely possible to use up all of the southbridge's 8 lanes just on onboard controllers. The main advantage I see in potentially running the SB lanes at PCIe3 speeds isn't total bandwidth which isn't likely to be saturated; but allowing board makers to use 4xUSB,3 4x SATA6GB, or 2x GB nic controllers to get their port inflation numbers up while only consuming a single PCIe lane instead of 2.
DanNeely - Friday, May 6, 2011 - link
That or they should bump the number of lanes on the southbridge from 8 to 12ish to allow connecting as many low average bandwidth devices as they could when the PCI controller was still available. IIRC the controller on Intel's current desktop boards supports 5 PCI devices.Ammaross - Friday, May 6, 2011 - link
If you remember, current P67 motherboards share PCIe lanes. If you drop a card in the bottom PCIe gfx 4x slot, it disables some on-board things (firewire, some USB headers, etc). In essence, the lanes are shared amongst the devices already.DanNeely - Friday, May 6, 2011 - link
I wasn't aware of anything like that, I assume the disabled headers are USB3 since there's no reason to disable any of the USB2 ports. The firewire port surprises me though, I'd always assumed it would have been stuck on the legacy PCI bus since there's plenty of capacity there and no need for the higher speeds of PCIe.DanNeely - Friday, May 6, 2011 - link
While you're probably right about it not seeing much traction in the enthusiast market, assuming it trickles down to the mobile chipset as well, support for a third monitor would be really nice for my work laptop.iwod - Friday, May 6, 2011 - link
This basically are the rumors we have heard from other places summed up here. Of coz being Anandtech this means the rumors are realistic.The EU are properly a lot different to current SandyBridge, We expect to have double the Gfx Performance from Ivy Bridge. If any of the previous Intel slide are to be trusted. 22nm would also allow much higher Clock Speed for GPU part.
I hope Anandtech could find out whether Ivy Bridge have FMA.
$100 for a 20GB SLC SSD is really expensive, compare to the previous rumors of 40 - 50 USD.
Kristian Vättö - Friday, May 6, 2011 - link
Ivy Bridge IGP will only have 16 EUs which is only 33% more than what SB has. Like Anand mentioned in the SB review, it unfortunately looks unlikely that the IB IGP would be twice as fast, which is a shame.In the original version, I said the "Larsen Creek" SSD will be priced at around 50$ like the rumors have suggested but apparently Jarred edited that. Of course it is safer to say 100$ and then be happily surprised when it turns out to be 50$ instead of getting those angry comments when it turned out to be 100$ instead of 50$.
iwod - Friday, May 6, 2011 - link
My thought is that these "Next Gen EU " would be completely different to current EU, that is why it is not fair to compare EU numbers to performance.And if this new EU runs 33% faster per clock, and runs at 33% higher clock while having 33% more EU then previous EU, should land us at 100% performance increase.
Of coz, Hardware dont matter much on GPU, not at all. It is software, drivers that makes ALL the difference. As we have seen with S3. Nvidia has double the Software Engineering compare to hardware working on it. And if Intel continue the way it is, their GPU hardware will never get the respect they deserve.
I am looking forward to the Larsen Creek SSD.