Addressing the Memory Bandwidth Problem

Integrated graphics solutions always bumped into a glass ceiling because they lacked the high-speed memory interfaces of their discrete counterparts. As Haswell is predominantly a mobile focused architecture, designed to span the gamut from 10W to 84W TDPs, relying on a power-hungry high-speed external memory interface wasn’t going to cut it. Intel’s solution to the problem, like most of Intel’s solutions, involves custom silicon. As a owner of several bleeding edge foundries, would you expect anything less?

As we’ve been talking about for a while now, the highest end Haswell graphics configuration includes 128MB of eDRAM on-package. The eDRAM itself is a custom design by Intel and it’s built on a variant of Intel’s P1271 22nm SoC process (not P1270, the CPU process). Intel needed a set of low leakage 22nm transistors rather than the ability to drive very high frequencies which is why it’s using the mobile SoC 22nm process variant here.

Despite its name, the eDRAM silicon is actually separate from the main microprocessor die - it’s simply housed on the same package. Intel’s reasoning here is obvious. By making Crystalwell (the codename for the eDRAM silicon) a discrete die, it’s easier to respond to changes in demand. If Crystalwell demand is lower than expected, Intel still has a lot of quad-core GT3 Haswell die that it can sell and vice versa.

Crystalwell Architecture

Unlike previous eDRAM implementations in game consoles, Crystalwell is true 4th level cache in the memory hierarchy. It acts as a victim buffer to the L3 cache, meaning anything evicted from L3 cache immediately goes into the L4 cache. Both CPU and GPU requests are cached. The cache can dynamically allocate its partitioning between CPU and GPU use. If you don’t use the GPU at all (e.g. discrete GPU installed), Crystalwell will still work on caching CPU requests. That’s right, Haswell CPUs equipped with Crystalwell effectively have a 128MB L4 cache.

Intel isn’t providing much detail on the connection to Crystalwell other than to say that it’s a narrow, double-pumped serial interface capable of delivering 50GB/s bi-directional bandwidth (100GB/s aggregate). Access latency after a miss in the L3 cache is 30 - 32ns, nicely in between an L3 and main memory access.

The eDRAM clock tops out at 1.6GHz.

There’s only a single size of eDRAM offered this generation: 128MB. Since it’s a cache and not a buffer (and a giant one at that), Intel found that hit rate rarely dropped below 95%. It turns out that for current workloads, Intel didn’t see much benefit beyond a 32MB eDRAM however it wanted the design to be future proof. Intel doubled the size to deal with any increases in game complexity, and doubled it again just to be sure. I believe the exact wording Intel’s Tom Piazza used during his explanation of why 128MB was “go big or go home”. It’s very rare that we see Intel be so liberal with die area, which makes me think this 128MB design is going to stick around for a while.

The 32MB number is particularly interesting because it’s the same number Microsoft arrived at for the embedded SRAM on the Xbox One silicon. If you felt that I was hinting heavily at the Xbox One being ok if its eSRAM was indeed a cache, this is why. I’d also like to point out the difference in future proofing between the two designs.

The Crystalwell enabled graphics driver can choose to keep certain things out of the eDRAM. The frame buffer isn’t stored in eDRAM for example.

Peak Theoretical Memory Bandwidth
  Memory Interface Memory Frequency Peak Theoretical Bandwidth
Intel Iris Pro 5200 128-bit DDR3 + eDRAM 1600MHz + 1600MHz eDRAM 25.6GB/s + 50GB/s eDRAM (bidirectional)
NVIDIA GeForce GT 650M 128-bit GDDR5 5016MHz 80.3 GB/s
Intel HD 5100/4600/4000 128-bit DDR3 1600MHz 25.6GB/s
Apple A6X 128-bit LPDDR2 1066MHz 17.1 GB/s

Intel claims that it would take a 100 - 130GB/s GDDR memory interface to deliver similar effective performance to Crystalwell since the latter is a cache. Accessing the same data (e.g. texture reads) over and over again is greatly benefitted by having a large L4 cache on package.

I get the impression that the plan might be to keep the eDRAM on a n-1 process going forward. When Intel moves to 14nm with Broadwell, it’s entirely possible that Crystalwell will remain at 22nm. Doing so would help Intel put older fabs to use, especially if there’s no need for a near term increase in eDRAM size. I asked about the potential to integrate eDRAM on-die, but was told that it’s far too early for that discussion. Given the size of the 128MB eDRAM on 22nm (~84mm^2), I can understand why. Intel did float an interesting idea by me though. In the future it could integrate 16 - 32MB of eDRAM on-die for specific use cases (e.g. storing the frame buffer).

Intel settled on eDRAM because of its high bandwidth and low power characteristics. According to Intel, Crystalwell’s bandwidth curve is very flat - far more workload independent than GDDR5. The power consumption also sounds very good. At idle, simply refreshing whatever data is stored within, the Crystalwell die will consume between 0.5W and 1W. Under load, operating at full bandwidth, the power usage is 3.5 - 4.5W. The idle figures might sound a bit high, but do keep in mind that since Crystalwell caches both CPU and GPU memory it’s entirely possible to shut off the main memory controller and operate completely on-package depending on the workload. At the same time, I suspect there’s room for future power improvements especially as Crystalwell (or a lower power derivative) heads towards ultra mobile silicon.

Crystalwell is tracked by Haswell’s PCU (Power Control Unit) just like the CPU cores, GPU, L3 cache, etc... Paying attention to thermals, workload and even eDRAM hit rate, the PCU can shift power budget between the CPU, GPU and eDRAM.

Crystalwell is only offered alongside quad-core GT3 Haswell. Unlike previous generations of Intel graphics, high-end socketed desktop parts do not get Crystalwell. Only mobile H-SKUs and desktop (BGA-only) R-SKUs have Crystalwell at this point. Given the potential use as a very large CPU cache, it’s a bit insane that Intel won’t even offer a single K-series SKU with Crystalwell on-board.

As for why lower end parts don’t get it, they simply don’t have high enough memory bandwidth demands - particularly in GT1/GT2 graphics configurations. According to Intel, once you get to about 18W then GT3e starts to make sense but you run into die size constraints there. An Ultrabook SKU with Crystalwell would make a ton of sense, but given where Ultrabooks are headed (price-wise) I’m not sure Intel could get any takers.

Haswell GPU Architecture & Iris Pro The Core i7-4950HQ Mobile CRB
POST A COMMENT

177 Comments

View All Comments

  • HisDivineOrder - Saturday, June 01, 2013 - link

    I see Razer making an Edge tablet with an Iris-based chip. In fact, it seems built for that idea more than anything else. That or a NUC HTPC run at 720p with no AA ever. You've got superior performance to any console out there right now and it's in a size smaller than an AppleTV.

    So yeah, the next Razer Edge should include this as an optional way to lower the cost of the whole system. I also think the next Surface Pro should use this. So high end x86-based laptops with Windows 8 Pro.

    And NUC/BRIX systems that are so small they don't have room for discrete GPU's.

    I imagine some thinner than makes sense ultrathins could also use this to great effect.

    All that said, most systems people will be able to afford and use on a regular basis won't be using this chip. I think that's sad, but it's the way it will be until Intel stops trying to use Iris as a bonus for the high end users instead of trying to put discrete GPU's out of business by putting these on every chip they make so people start seeing it CAN do a decent job on its own within its specific limitations.

    Right now, no one's going to see that, except those few fringe cases. Strictly speaking, while it might not have matched the 650m (or its successor), it did a decent job with the 640m and that's a lot better than any other IGP by Intel.
    Reply
  • Spunjji - Tuesday, June 04, 2013 - link

    You confused me here on these points:

    1) The NUC uses a 17W TDP chip and overheats. We're not going to have Iris in that form factor yet.
    2) It would increase the cost of the Edge, not lower it. Same TDP problem too.

    Otherwise I agree, this really needs to roll down lower in the food chain to have a serious impact. Hopefully they'll do that with Broadwell used by the GPU when the die area effectively becomes free thanks to the process switch.
    Reply
  • whyso - Saturday, June 01, 2013 - link

    So intel was right. Iris Pro pretty much matches a 650m at playable settings (30 fps +). Note that anandtech is being full of BullS**t here and comparing it to an OVERCLOCKED 650m from apple. Lets see, when intel made that 'equal to a 650m' claim it was talking about a standard 650m not an overclocked 650m running at 900/2500 (GDDR5) vs the normal 835/1000 (GDDR5 + boost at full, no boost = 735 mhz core). If you look at a standard clocked GDDR3 variant iris pro 5200 and the 650m are pretty much very similar (depending on the games) within around 10%. New Intel drivers should further shorten the gap (given that intel is quite good in compute). Reply
  • JarredWalton - Sunday, June 02, 2013 - link

    http://www.anandtech.com/bench/Product/814

    For the games I tested, the rMBP15 isnt' that much faster in many titles. Iris isn't quite able to match GT 650M, but it's pretty close all things considered.
    Reply
  • Spunjji - Tuesday, June 04, 2013 - link

    I will believe this about new Intel drivers when I see them. I seriously, genuinely hope they surprise me, though. Reply
  • dbcoopernz - Saturday, June 01, 2013 - link

    Are you going to test this system with madVR? Reply
  • Ryan Smith - Sunday, June 02, 2013 - link

    We have Ganesh working to answer that question right now. Reply
  • dbcoopernz - Sunday, June 02, 2013 - link

    Cool. :) Reply
  • JDG1980 - Saturday, June 01, 2013 - link

    I would have liked to see some madVR tests. It seems to me that the particular architecture of this chip - lots of computing power, somewhat less memory bandwidth - would be very well suited to madVR's better processing options. It's been established that difficult features like Jinc scaling (the best quality) are limited by shader performance, not bandwidth.
    The price is far steeper than I would have expected, but once it inevitably drops a bit, I could see mini-ITX boards with this become a viable solution for high-end, passively-cooled HTPCs.
    By the way, did they ever fix the 23.976 fps error that has been there since Clarkdale?
    Reply
  • dbcoopernz - Saturday, June 01, 2013 - link

    Missing Remote reports that 23.976 timing is much better.

    http://www.missingremote.com/review/intel-core-i7-...
    Reply

Log in

Don't have an account? Sign up now