Intel’s new LGA 1156-based processors, namely the Core i5-700- and Core i7-800-series, arrived with a bang. Our launch article by Chris Angelini provides all the key information. But, in short, the new processor delivers increased performance compared to Core 2 Quad. No surprise there. But Intel also claims that power consumption went down significantly, so we decided to tackle this prospect with a closer look.
The Core i5 Innovation
The new P55 platform and Intel’s Lynnfield processors make a are impressive enough from a performance angle, but I’d like to look at the facts from a different perspective. The platform’s foundation still centers on the Nehalem architecture first seen as Core i7 for LGA 1366 late last year, hence the level of innovation in Lynnfield is less revolutionary and more of an evolution.
Intel is great at tweaking, though. From a technology standpoint, Core i5 incorporates a lot more integration, stepping closer to the inevitable SoC (system-on-chip) future by adding PCI Express onto the die. A more aggressive implementation of Turbo Boost adds extra performance for non-threaded applications. Considering that Intel plans to integrate graphics with Clarkdale in coming months, this step could also be looked at as a necessity. Improved Turbo Boost is another building block for turning dynamic overclocking into a real feature. In reality, though, the large motherboard makers have been offering similar features for a few years already.
I see two main benefits of the new LGA 1156 processors for buyers. First, you have decreased power consumption thanks to the integrated memory controller and PCI Express 2.0 interface. The second benefit is performance. Turbo Boost helps to elevate LGA 1156-based parts to Core i7-900-class performance levels in certain applications. Both benefits became obvious in our article, Intel’s Mainstream Magnum Opus.
Performance Increase Where It's Needed
The current generation of Turbo Boost accelerates a single core by four 133 MHz increments (or bins) in the case of Core i5-700-series and by as many as five increments for the Core i7-800-series. This equals clock speed bumps of 533 and 667 MHz. Two cores can accelerate by four clock speed increments, while three and four cores still may receive a simultaneous 133 MHz boost (up to 266 MHz for the top model Core i7-860). Obviously, the new chips operate with higher power consumption every time Turbo Boost kicks in, though they're kept under a 95W ceiling by Intel's on-die Power Control Unit.
Power Down?
We want to answer some basic questions: What is the real reduction in system power consumption compared to Core 2 Quad platforms? What is the efficiency (performance per watt) for the new LGA 1156 generation? What’s the difference in efficiency for a LGA 1156 system with and without Turbo Boost enabled?
Power Play On TDP
Once upon a time, an Intel platform consisted of three components: a processor, a chipset northbridge (with a graphics interface and memory controller, also called the Memory Controller Hub or MCH), and a southbridge (the I/O Controller Hub, or ICH). The potential power consumption of these three components has to be added together. The processors are rated from 65W up to 130W, a P45 MCH requires up to 22W, and the ICH10R is rated at 4.5W.
The new LGA 1156 platform, including the Core i5-700- and i7-800-series CPUs, introduces Intel’s first mainstream two-chip design. The memory controller slipped into the processor with Bloomfield (Core i7 on LGA 1366) and the PCI Express interface now follows suit. As a result, the need for a separate northbridge is gone, leaving mainly I/O and management functionality behind. Given this slimmed-down arrangement, the P55 chipset is officially called the Platform Controller Hub, or PCH. Since large chunks of the PCH correspond to equivalent areas of the old ICH, net complexity and power consumption are similar.
Effectively, the new processors stay at the same power level as as some of the preceding Core 2 Quads: 95W. Compared to Core i7 on LGA 1366, the new processors are already rated 35W lower. Moreover, the fact that there's no longer a discrete MCH means that 20-some watt piece of logic no longer contributes to overall power consumption. Here is a summary:
The Math on Power Consumption
| Core 2 Duo And P45 | Core 2 Quad And P45 | Core i7 And X58 | Core i5/i7 And P55 | |
|---|---|---|---|---|
| Processor TDP | 65W | 65-95W | 130W | 95W |
| Northbridge TDP | 22W | 22W | 24.1W | - |
| Southbridge TDP | 4.5W | 4.5W | 4.5W | 4.7W |
| Total Platform TDP | 91.5W | 91.5 - 121.5W | 158.6W | 99.7W |
If you look at the highest possible specified power levels, Core i7 and X58 top out at nearly 160W, while a Core 2 Quad platform requires up to 122W. The new processor’s sub-100W platform ceiling already represents a significant advance in potential efficiency without us having run a single power or performance benchmark. However, keep in mind that these numbers reflect maximum power consumption with regards to thermal limits (thermal design power). Reality might very well turn out to be a bit different.
The math also does not include power required by the motherboard and on-board components. Voltage regulators may be more or less efficient, and additional audio, network, or RAID chips all consume power, as well. We also have to add memory and a hard drive or SSD. Then there’s the graphics card. Entry-level models require around 20W idle power while high-end components easily eat up 60W and more without doing anything but displaying the Windows desktop. Hence we use a mainstream GeForce GTX 260² from Zotac, which is an efficient but still modern-enough model. Installing a faster graphics solution would just reduce the effective power consumption difference between various configurations based on Core 2 or Core i5/i7.
Test Setup
| System Hardware | |
|---|---|
| Hardware | Details |
| Performance Benchmarks | |
| Motherboard (LGA 1156) | Intel DP55KG (Rev. 1.0) Chipset: Intel P55 BIOS: KGIBX10J.86A.3456 (08/22/2009) |
| Motherboard (LGA 775) | MSI P45D3 Neo (Rev. 1.0) Chipset: Intel P45, ICH10R BIOS: 4.2 (02/18/2009) |
| CPU Intel I | Intel Core i5-750 (45nm, 2.66 GHz, 4 x 256KB L2 and 8MB L3 Cache, TDP 95W, Rev. B1) |
| CPU Intel II | Intel Core i7-870 (45nm, 2.93 GHz, 4 x 256KB L2 and 8MB L3 Cache, TDP 95W, Rev. B1) |
| CPU Intel III | Intel Core 2 Quad Q8200S (45nm, 2.33 GHz, 4MB L2 Cache, TDP 65W, Rev. R0) |
| CPU Intel IV | Intel Core 2 Quad Q9550S (45nm, 2.83 GHz, 12MB L2 Cache, TDP 65W, Rev. E0) |
| CPU Intel V | Intel Core 2 Extreme Q9650 (45nm, 3.00 GHz, 12MB L2 Cache, TDP 130W, Rev. C0) |
| RAM DDR3 (Dual) | 2 x 2GB DDR3-1600 (Corsair CM3X2G1600C9DHX) 2 x 1GB DDR3-2000 (OCZ OCZ3P2000EB1G) |
| Graphics | Zotac Geforce GTX 260² GPU: Geforce GTX 260 (576 MHz), Graphics RAM: 896MB DDR3 (1998 MHz), Stream Processors: 216, Shader Clock: 1242 MHz |
| Hard Drive | Western Digital VelociRaptor, 300GB (WD3000HLFS) 10,000 RPM, SATA/300, 16MB Cache |
| Blu-Ray Drive | LG GGW-H20L, SATA/150 |
| Power Supply | PC Power & Cooling, Silencer 750EPS12V 750W |
| System Software and Drivers | |
| Operating System | Windows Vista Enterprise Version 6.0 x64 Service Pack 2 (Build 6000) |
| Drivers and Settings | |
| AMD Chipset Drivers | Catalyst Control Center 9.4 |
| Intel Chipset Drivers | Chipset Installation Utility Ver. 9.1.1.1015 |
| Intel Storage Drivers | Matrix Storage Drivers Ver. 8.8.0.1009 |
Power Consumption
Clearly, the new Core i5/i7 processor generation on LGA 1156 requires considerably less idle power than Core 2 Quad. In this example, we even used the S-models (low-power) of Intel’s Core 2 Quad family, which are more efficient at idle and particularly at peak power. However, for comparison at the other end of the spectrum, we also tested the Core 2 Extreme QX9650, the most power-hungry model in its family. On our test system, using a GeForce GTX260², the LGA 1156 platform requires between 6W and 16W less idle power than the Core 2 family.
At peak power using Prime95, the two Core 2 Quad S-models remain the most efficient products. However, keep in mind that they were designed for efficiency (Ed.: and priced accordingly; that Q9550S is still a $219 part). They’re rated at a 65W TDP, while the new LGA 1156 processors are 95W models. The Core 2 Extreme QX9650 processor is a juice-guzzling 130W CPU.
Peak power of these processors directly corresponds to maximum clock speed. If we enable Turbo Boost on the Core i7-870, the processor can reach up to 3.6 GHz. As a result, peak power consumption will also be very close to the thermal envelope’s top edge in an effort to maximize performance. Disabling Turbo Boost in the Core i7-870 results in an obvious decrease of peak power from 182W to 178W. The Core i5-750 keeps its maximum clock speed much lower: 3.2 GHz in Turbo Boost mode or 2.8 GHz with four active cores.
Next, we tracked the total power used (in watt-hours) by the test systems to complete one full PCMark Vantage run. In our Core i5/i7 launch article, you could see that the new CPUs outperform Core 2 Quad models. The power-efficient S-models require less power for this workload, but they take more time to complete the benchmark. Intel’s old Core 2 Extreme QX9650 is the most voracious processor here. The LGA 1156 chips aren’t particularly low on power consumption, but keep in mind that Turbo Boost provides an extra performance kick. Let’s turn to performance per watt now.
Efficiency Comparison
Efficiency Comparison
First, we need the average power requirement during a PCMark Vantage run. We obtained this number by tracking power consumption throughout a full PCMark Vantage run and calculating the average. Once again,you can see that the new processors are roughly at the level of old Core 2 Quad systems, but you get much increased performance at the same average power consumption. Now let’s relate the PCMark Vantage score to the average power required to get a performance per watt result.
This is the real deal. Since the new Core i5/i7 generation on socket LGA 1156 delivers much better performance at similar peak power consumption and decreased idle power, the performance per watt delivered by the new platforms is greatly improved. However, there’s a surprise waiting: Turbo Boost does not help increase power efficiency. In fact, it decreases it, at least in the case of PCMark Vantage. Clearly, the Core i7-870 requires considerably more power with Turbo Boost enabled and hence has a rather negative impact on power efficiency. A Core i7-870 at 2.93 GHz may deliver less performance, but the power savings without Turbo Boost are significant enough to have a very noticeable efficiency benefit.
Our efficiency diagram shows power consumption for each of the system configurations at every point of the PCMark Vantage run.
Conclusion
There’s no denying that Intel did a great job of optimizing its Nehalem architecture for the mainstream. System idle power went down considerably, and even though system peak power didn’t decrease much, you get a lot more responsiveness. Intel integrated the PCI Express interface into the processor and adjusted clock speed management for each of the four available cores via its second-generation Turbo Boost technology to maximize performance when needed.
As a result, the Core i5/i7’s efficiency (performance per watt) increased nicely. The new platforms can do more work at slightly reduced peak power consumption levels. You’d suspect that Turbo Boost deserves credits for this development, but this isn’t quite true, at least for the Core i7-870. Turbo Boost's massive clock speed bumps (up to 667 MHz) utilize the available thermal envelope more aggressively, which, in the end, actually decreases the power efficiency of the Core i7-870 system when compared to the same machine not running Turbo Boost. We don’t want to generalize these results, but they’re certainly the case when benchmarking PCMark Vantage.
In everyday life, Turbo Boost will probably kick in for smaller periods of time and we’d assume it typically does so for one or two cores. Scenarios in which three or all cores can be accelerated requires heavily threaded applications, and in such cases you probably want all the performance you can get. Efficiency obviously isn’t a concern at these times. Having one core running at 3.6 GHz will certainly not increase system power consumption enough to significantly impact efficiency.
Source: http://www.tomshardware.com
No comments:
Post a Comment