Monday, August 11th 2008
Evaluation of the 45nm AMD Deneb Reveals an Efficient Processor in the Making
Chinese website Hardspell conducted a comprehensive pre-release evaluation of the upcoming Deneb 45nm Quad-core processor by AMD. The Deneb core incorporates thrice the amount of L3 Cache (that's 6 MB), and uses the same SIMD sets as its 65nm counterparts.
Here's a shocker: While the Phenom X4 9650 (65nm, 2.30 GHz, B3) consumes 104.1 W at load (peak), the 45nm Deneb (45nm, 2.30 GHz) peaks at an astonishing 57.3 W according to Hardspell's findings, go to see, the Deneb has an added load of transistors due to a 300% increase in the L3 Cache size. Let's bring in some numbers and figures.
CPU-Z Identification
Version 1.46.2 and above of CPU-Z detects the processor except the CPUID string which the engineering samples don't usually bring along. The 2.30 GHz Deneb Part comes with a 1.80 GHz HT link with 3600 MT/s of system bandwidth over the HyperTransport 3.0 bus. The L3 Cache uses 48-way set-associative paths.Power Consumption
Specifications of the test-bed are provided. The 65nm Agena part was compared to the 45nm Deneb at the same clock-speed, idle and load consumptions were measured (first chart: idle, second: load):Benchmark Scores
Fritz Chess (higher is better):W-Prime Multithreaded Benchmark (time, lower is better):POV-Ray 3.7 beta23 SSE2 (higher is better):H.264 Encoding (time, lower is better):3DMark Vantage (CPU score, higher is better):The test bed was configured as follows:
Source:
Hardspell
Here's a shocker: While the Phenom X4 9650 (65nm, 2.30 GHz, B3) consumes 104.1 W at load (peak), the 45nm Deneb (45nm, 2.30 GHz) peaks at an astonishing 57.3 W according to Hardspell's findings, go to see, the Deneb has an added load of transistors due to a 300% increase in the L3 Cache size. Let's bring in some numbers and figures.
CPU-Z Identification
Version 1.46.2 and above of CPU-Z detects the processor except the CPUID string which the engineering samples don't usually bring along. The 2.30 GHz Deneb Part comes with a 1.80 GHz HT link with 3600 MT/s of system bandwidth over the HyperTransport 3.0 bus. The L3 Cache uses 48-way set-associative paths.Power Consumption
Specifications of the test-bed are provided. The 65nm Agena part was compared to the 45nm Deneb at the same clock-speed, idle and load consumptions were measured (first chart: idle, second: load):Benchmark Scores
Fritz Chess (higher is better):W-Prime Multithreaded Benchmark (time, lower is better):POV-Ray 3.7 beta23 SSE2 (higher is better):H.264 Encoding (time, lower is better):3DMark Vantage (CPU score, higher is better):The test bed was configured as follows:
67 Comments on Evaluation of the 45nm AMD Deneb Reveals an Efficient Processor in the Making
Next, I remember reading something regarding their no architecture change here at TPU. If I find that article I will post a link.
Next, AMD never boasted about a new architecture in the process, it always says that it has fine tuned.
If you find something to prove then please do post here.
The lower power consumption would be excelent news in numerous ways, but what comes to mind first is the server front. In server business the nr.1 enemy is heat and power draw, with the new 45nm process AMD could tackle those both and make great increase on their server market share.
Deneb is really only a placeholder - but a good placeholder!
Am I right assuming it's the hype-name for their next architecture?
en.wikipedia.org/wiki/Bulldozer_(processor)
the current form, in wiki form to read
2. If you don't trust the Chinese, prepare not to trust CPU-Z, CPUID SDK and its creators as well, the primary specs are put up in those CPU-Z windows.
I feel safe saying that by my tests that AMD really isn't really all that far behind Intel (with the exception of OCability of course, but I don't think that matters all that much considering only ~10% of PC users overclock their hardware.)
Like you said, I am absolutely happy with my Quad after disabling the TLB patch. I just wanted AMD to release a killer CPU and get back in the race.
most stock users look up reviews, hear a chip was a 'great clocker' have no idea what it means and buy it anyway.
Edit: also every review I've ever seen (other than TPU) has given crappy little overclocks that really don't amount to anything. Pity really.