Kaveri and Counting Cores

With the move towards highly integrated SoCs we've seen a variety of approaches to core counts. Apple, Intel and Qualcomm still count CPU cores when advertising an SoC. For Apple and Qualcomm that's partially because neither company is particularly fond of disclosing the configuration of their GPUs. More recently, NVIDIA took the somewhat insane stance of counting GPU CUDA cores on its Tegra K1 SoC. Motorola on the other hand opted for the bizarre choice of aggregating CPU, GPU and off-die companion processors with the X8 platform in its Moto X smartphone. Eventually we will have to find a way to characterize these highly integrated SoCs, particularly when the majority of applications actually depend on/leverage both CPU and GPU cores.

AMD finds itself in a unique position with Kaveri where it has a truly unified CPU/GPU architecture and needs to establish a new nomenclature for use in the future. With 47% of the Kaveri die dedicated for GPU use, and an architecture that treats both CPU and GPU as equals, I can understand AMD's desire to talk about the number of total cores on the APU.

AMD settled on the term "Compute Core", which can refer to either an x86 (or maybe eventually ARM) CPU core or a GCN compute unit. The breakdown is as follows:

  • Each thread on a CPU is a Compute Core
  • Each Compute Unit on the IGP is a Compute Core
  • Total Compute Cores = CPU Compute Cores + IGP Compute Cores

This means that the high end SKU, the A10-7850K will have a total of 12 compute cores: four from the CPU (two Steamroller modules supporting four threads) and eight from the IGP (due to eight compute units from the R7 graphics).

There are some qualifications to be made on this front. Technically, AMD is correct – each compute unit in the IGP and each thread on the CPU can run separate code. The Hawaii GCN architecture can spawn as many kernels as compute units, whereas a couple of generations ago we were restricted to one compute kernel on the GPU at once (merely with blocks of work being split across the CUs). However, clearly these 12 compute units are not equivalent: a programmer will still have to write code for the CPU and GPU specifically in order to use all the processing power available.

Whenever AMD (or partners) are to promote the new APUs, AMD tells us clearly that two sets of numbers should be quoted in reference to the Compute Cores – the total, and the breakdown of CPU/GPU on the APU. Thus this would mean that the A10-7850K APU would be marketed at a “12 Compute Core” device, with “(4 CPU + 8 GPU)” following immediately after. I applaud AMD's decision to not obfuscate the internal configuration of its APUs. This approach seems to be the most sensible if it wants to tout the total processing power of the APU as well as tell those users who understand a bit more what the actual configuration of the SoC is. The biggest issue is how to address the users who automatically assume that more cores == better. The root of this problem is very similar to the old PR-rating debates of the Athlon XP. Explaining to end users the intracacies of CPU/GPU programming is really no different than explaining why IPC * frequency matters more than absolute frequency.

When a programmer obtains an APU, the OpenCL profiler should locate the eight compute units of the GPU and display that to the user for offloading compute purposes; at the same time it is up to the programmer to leverage the threads as appropriately as possible, even with AMD’s 3rd generation Bulldozer modules implementing a dual-INT + single-FP solution.

At launch, AMD will offer the following configurations:

  • A10-7850K: 12 Compute Cores (4 CPU + 8 GPU)
  • A10-7700K: 10 Compute Cores (4 CPU + 6 GPU)
  • A8-7600: 10 Compute Cores (4 CPU + 6 GPU)

The problem with distilling the processing power of the APU into a number of compute cores is primarily on the CPU side. AMD will be keeping the GPU at approximately the same clock (720 MHz for these three) but the CPU frequency will differ greatly, especially with the A8-7600 which has configurable TDP and when in 45W mode will operate 300-400 MHz less.

Steamroller Architecture Redux

We've already documented much of what Steamroller brings to the table based on AMD's Hot Chips presentation last year, but I'll offer a quick recap here.

Steamroller brings forward AMD's Bulldozer architecture, largely unchanged. We're still talking about a dual-core module featuring two independent integer execution cores with a single shared floating point execution core capable of executing instructions from two threads in parallel. A single module still appears as two cores/threads to the OS.

In Bulldozer and Piledriver, each integer core had its own independent scheduler but the two cores shared a single fetch and decode unit. Instructions would come in and decodeded operations would be fed to each integer pipe on alternating clock cycles. In Steamroller the decode hardware is duplicated in each module, so now each integer core gets its own decode unit. The two decode units are shared by the one FP unit.

L1 instruction caches increase in size from 64KB to 96KB per module, which AMD claims reduces misses by up to 30%. There's an updated branch predictor which is responsible for reducing mispredicted branches by up to 20%. Both integer and FP register files grow in size as well as an increase in the size of the scheduling window, a combination of which increases the number of dispatches per thread by up to 25%.

There are huge improvements on the store side. Steamroller can now issue up to 2 stores at the same time compared to 1 in Bulldozer/Piledriver. The load/store queue sizes go up by around 20% as well. It's remarkable just how much low hanging fruit there was in the Bulldozer design.

GCN in an APU

The integrated graphics solution on Trinity/Richland launched with a Cayman-derived VLIW4 architecture, which unfortunately came just shortly after desktop side of the equation finished moving from VLIW5/VLIW4 to GCN. Having a product stack with largely different GPU architectures doesn't help anyone, particularly on the developer front. Looking forward, sticking with GCN was the appropriate thing to do, as now Kaveri is using GCN, the same architecture found in AMD’s high end R9-290X GPU, based on the Hawaii platform.

This enabled AMD to add in all the features they currently have on Hawaii with little to no effort – the TrueAudio DSP, the upgraded Video Coding Engine and Unified Video Decoder are such examples. Whether or not AMD decides to develop an APU with more than 8 GCN CUs is another matter. This is a point we've brought up with AMD internally and one that I'm curious about - do any AnandTech readers have an interest in an even higher end APU with substantially more graphics horsepower? Memory bandwidth obviously becomes an issue, but the real question is how valuable an Xbox One/PS4-like APU would be to the community.

Kaveri: Aiming for 1080p30 and Compute Llano, Trinity and Kaveri Die: Compared
Comments Locked

380 Comments

View All Comments

  • jaydee - Tuesday, January 14, 2014 - link

    I just don't "get" adding three $300 Intel CPU's to a review of 3 sub $200 AMD CPU's. We all know, or can find out how i7 SB, IB, Haswell compare to each other. I can see adding one of these CPUs to show a baseline of how AMDs top-of-the-line compares against Intels (albiet at very different price points), but having all three of them gives the impression that you want to make sure everyone knows who's boss...

    Is there going to be an update on power draw? I'm really curious to see what the 45W Kaveri draws (idle and full power) considering it is so competitive with the 95W Kaveri.
  • jaydee - Tuesday, January 14, 2014 - link

    I stand corrected, four $300+ Intel i7 CPUs in this comparison, not three. And two Intel CPU's that are actually in the price range of the AMDs. It feels as if to do this review, AT just cobbled together whatever was laying around instead of being intentional about putting together the best test bench possible to compare the review product against its real competitors.

    It's really a shame, because the commentary, the in-depth look at the architecture and the conclusions are outstanding; no website out there has tech writers as good and do as thorough of a job as Anandtech. The fact that the test benches are just after-thoughts in some of these reviews are really disappointing.
  • UtilityMax - Tuesday, January 14, 2014 - link

    I think this was mentioned somewhere in the beginning of the review. Intel likes to have the i7 reviewed instead of other parts, so the send the i7 CPUs to everyone. This is kind of like what the car makers do. Most people just buy a reasonably priced, mid-spec car model. However, the journos always get to review these ridiculously over-optioned cars, that hardly sell, like the $35 Ford Fusions, even though the base car sells for 10 grand less.
  • jaydee - Tuesday, January 14, 2014 - link

    I am well aware that when having the choice, Intel would rather you review a $340 Intel CPU against a $179 AMD CPU. But is there not any way, given the ad revenue of Anandtech, to obtain a $190-200 mid-range Intel i5 CPU (such as the i5-4440) in order to have a relevant test bench for an eagerly anticipated AMD mid-range CPU launch?
  • srkelley - Tuesday, January 14, 2014 - link

    "...do any AnandTech readers have an interest in an even higher end APU with substantially more graphics horsepower?"

    Yes, oh yes! I'm letting the APU be the core of my system build in a few months and plan to upgrade as needed. I'd like a simple solution like an APU instead of having to go with a discrete card right away. If it lets me spend more on ram and other things, keep the psu and power draw low I'm happy. The most demanding game that I play right now is the Witcher 2. Eventually I will have to go discrete for Star Citizen and the Witcher 3 but if I can get solid enough results with a high powered apu then I'll simply jump to that instead.
  • Conduit - Tuesday, January 14, 2014 - link

    This has been a long time coming thanks to Always Major Delays (AMD). Even know they can't get their sh!t together.
  • nissangtr786 - Tuesday, January 14, 2014 - link

    http://techreport.com/review/25908/amd-a8-7600-kav...
    The fpu still not improved miles still behind intel.
    http://browser.primatelabs.com/geekbench3/326781
    http://browser.primatelabs.com/geekbench3/321256
  • A5 - Tuesday, January 14, 2014 - link

    "do any AnandTech readers have an interest in an even higher end APU with substantially more graphics horsepower? "

    Maybe in the context of a Steam Machine? But for my main gaming PC, no way. Maybe something they can try out after the next die shrink if SteamOS really takes off.
  • Xajel - Tuesday, January 14, 2014 - link

    I believe the main reason for AMD to not have a new FX is SOI process.. it was okay in it's glory days, but it can't keep it up with bulk silicon...

    They want to increase the clock at least but they're trapped with 32nm SOI, and why not moving toward 28nm or lower SOI because I think that AMD is already working to convert it's entire CPU's to Bulk Silicon, so it's not logical to make a new design for still not ready yet 28nm or lower SOI while they are already designing the new core for bulk silicon.. knowing that both SOI and bulk require a complete redesign of the silicon just to convert from one to another... so it's not even logical to convert Pilediver or maybe even Steamroller to bulk silicon for only one year while they're working on the next architecture which will be Bulk Silicon...
  • jimjamjamie - Thursday, January 16, 2014 - link

    That makes sense, I was very confused as to why AMD were not going to refresh the FX line - even if HSA is the future for AMD, I presumed new FX sales throughout this year would have helped things along.

    Perhaps it wasn't worth the cost, but at the same time AMD could really do with keeping the fanboys on side.

Log in

Don't have an account? Sign up now