« Computex | Main | Gossip & Tips »
Intel Nehalem to be called Core i7
Intel has just announced that the upcoming 45nm 'Nehalem' processors will carry the Core i7 moniker. The logos above show the variants for mainstream models (blue) and the Extreme Editions (black). Nehalem is due to launch by the end of 2008, and includes novelties such as an integrated memory controller and a graphics chip built into the CPU package. It also introduces Intel's replacement for the traditional front-side bus, known as QuickPath Interconnect (QPI). Hyperthreading also returns, allowing two threads per core.
Remote voice recording kit
A new wireless device allows you to record meetings and make audio notes onto your PC frpm up to 30 metres away, says vendor Lindy.
The £69.99 (inc Vat) Wireless Voice Recorder package consists of a wireless handset, a USB dongle and software.
Recordings can be played back either on your PC or on the handset's built-in speaker. The integrated lithium battery lasts for up to twelve hours, Lindy says.
Phenom X4 9850 overclock record: 3.1GHz
We’ve had the Phenom X4 9850 for a day now and we’ve clocked all four cores stably to 3.1GHz using a regular Akasa AK-876 air cooler. That’s a modest 24 per cent boost over its native 2.5GHz clock speed.
We used a Gigabyte GA-MA790FX-DQ6 motherboard, 2GB Corsair TwinX 1066MHz Ram (downclocked to 800MHz but linked to the bus so it increased to 888MHz), a 0.225V core boost, a 14x CPU multiplier and a 222MHz bus speed to achieve this magic number.
That voltage boost is rather big, but our sample was also happy to sit at 2.8GHz with a simple 0.05V increase. It even booted at 3.2GHz, but couldn’t load Windows properly.
When overclocked, our preliminary results show it is 13 per cent faster than a 2.66GHz Core 2 Extreme QX6700 in PCmark05 and two per cent faster in Cinebench 9.5. It’s still slower than a 3GHz QX6850 and a long way off high end Penryn quad cores.
At its default speed (2.5GHz) the Phenom X4 9850 can’t keep up with Intel’s slowest Core 2 Quad Q6600 processor in PCmark05 CPU test, Cinebench 9.5 or X. It does beat the Q6600 by 160 points in 3Dmark06’s CPU test though.
The Phenom X4 9850’s big hope lies in a decent retail price. Pre-order pricing lies at £170 which, in the UK, is £20 more than a Core 2 Quad Q6600.
A full review will follow, but here are some more scores from our 3.1GHz overclocked sample:
3Dmark06 CPU: 4526
Cinebench 9.5: 478 (single core), 1473 (quad core)
Cinebench X: 2707 (single core), 9959 (quad core)
PCmark05: 9172 (just 7396 at 2.5GHz)
A full review will come soon, but for now these results can be directly compared to PCW's other CPU tests at www.reportlabs.com
Why Moore's Law needs AMD
Intel appears to have walked all over AMD in the past few days, releasing new quad-cores to spoil the much-heralded unveiling of its rival's Barcelona processors and then flourishing 32nm and 45nm wafers to point up the fact that AMD is stuck at 65nm.
There was little new on the processor front at the Intel Developer Forum, but the mood was upbeat. Intel founder Gordon Moore said that his famous law predicting that transistor density (and thus processing power) would double every 18-24 months looked good for up to 15 more years.
Enter the second, er... the third, er... the fourth AMD generation
Nobody from PCW Towers could make it to today's launch of AMD's Barcelona chips in Barcelona (where else?) and we could find no-one at the company to elucidate a somewhat confusion announcement. The nine new quad-cores all have names of the form 83xx or 23xx. The first digit, according to AMD, indicates the scalability: eight means that number can be used on a single motherboard.
The last two digits indicate relative performance, and the second digit the generation. So, following this logic, all today's processors are third generation.
But, according to an AMD presentation, all the quad-cores including 45nm versions launching next year, are second generation and the third generation will come in 2009 with the launch of 45nm octal-cores. And yet another AMD paper refers to them as fourth-generation, though this could relate to the manufacturing process used. We trust that the processors themselves won't get their data so mixed up.
Barcelona will rule data centres
Barcelona will double the number of cores in rack servers without changing the power budget if server managers choose it rather than Intel's quad core chips, AMD announced at its latest press confidence in London yesterday.
The claims join AMD's change in emphasis from performance advantages, to the power savings Barcelona offers compared to Intel's quad cores.
"We were guilty of talking too much about native quad core designs versus multi-chip. Now we've changed the language a little bit to speak about the benefits of native design" said EMEA channel market development manager Felipe Payet.
His colleague Steve Demski, Opteron Product Manager, then went on to hype up power-savings offered by native design. Firstly he said comparing AMD's and Intel's Thermal Design Power (TDP) wasn't an apples to apples comparison, since AMD's TDP takes into account the memory controller, whereas Intel's memory controller resides in the Northbridge.
Demski also made some interesting observations regarding AMD's use of DDR2 memory, since it consumes far less power than Intel's FBDimms.
The difference in power is around 60 Watts for eight Dimms and is AMD's trump card in the server space.
The change in emphasis has also led to AMD only comparing Barcelona performance to Rev F (the latest dual-core) Opterons explicitly. AMD will no longer state that Barcelona will be faster than Intel, instead just stating it will be "competitive".
While this doesn't mean much in the server space (where performance per watt is king), it could be a sign that upcoming consumer quad-core chips won't be significantly faster.
The fastest Barcelona quad-cores will leave AMD at 2GHz, going to manufacturers in August and hitting retail in September. Higher-clocked quad cores will arrive in late 2007. Look out for a review soon.
Intel to stick the boot in as AMD launches slow-clocking Barcelona
News that AMD is to launch its much-anticipated quad-core Opterons, codenamed Barcelona, in August seem to confound rumours of a major delay in the launch. But the picture is more complicated than that. AMD has been claiming that Barcelona would beat Intel's quad-core Xeons on both performance and power efficiency.
The first Barcelona chips will clock a modest 1.9GHz and 2.0GHz, with faster clock rates coming only later this year. Of course, as AMD has itself demonstrated over the past few years, clock rates are no longer a good measure of performance. AMD's design has an edge over Intel's in that it is true quad core, with the processors able to share cache, rather than being two dual-cores in one module. Still AMD is likely to be relying on faster clock rates to justify its early Barcelona hype.
Meanwhile Intel is not standing still, and has been expected to release a "spoiler" Xeon quad-core when Barcelona comes out. According to The Register, this will be a four-core Clovertown Xeon clocking 2.0GHz, and using at most 50 watts. It will also use a fastest-yet 1.333GHz front side bus.
Kingston announce DDR3 pricing
Kingston have revealed details about their DDR3 memory modules in both the HyperX and ValueRAM product lines. Available in 512MB and 1GB modules as well as 1 and 2GB kits, the first modules are clocked at DDR3-1375 (1,375MHz) for the HyperX line and DDR3-1066 (1,066MHz) for the ValueRAM. Kingston will also launch 1,333MHz modules in the ValueRAM line for the upcoming X38 chipset from Intel.
The latency’s (CL7-7-7-20 CAS7) of these first DDR3 modules are higher than DDR2 but it’s expected they will drop quickly once DDR3 gets established, and while the latency may be higher than DDR2 the voltages are lower 1.8v and 1.5v for the 1375 and 1066 speeds respectively
Pricing details are listed below
HyperX
KHX11000D3LLK2/2G 2GB 1375MHz DDR3 Kit £265(+ VAT)
KHX11000D3LLK2/1G 1GB 1375MHz DDR3 Kit £144(+ VAT)
KHX11000D3LL/1G 1GB 1375MHz DDR3 DIMM £133 (+ VAT)
KHX11000D3LL/512 512MB 1375MHz DDR3 DIMM £72 (+ VAT)
ValueRAM
KVR1066D3N7K2/2G 2GB 1066MHz DDR3 Kit £226(+ VAT)
KVR1066D3N7K2/1G 1GB 1066MHz DDR3 Kit £123(+ VAT)
KVR1066D3N7/1G 1GB 1066MHz DDR3 DIMM £113 (+ VAT)
KVR1066D3N7/512 512MB 1066MHz DDR3 DIMM £61 (+ VAT)
First AMD Fusion chips like Geodes?
AMD's first Fusion chip will only contain one CPU and one GPU, if a slip by Giuseppe Amato, technical director of sales and marketing at AMD is true.
The combined CPU/GPU chip should launch in 2009 and Amato said it would only have one CPU and one GPU on a single piece of silicon. When I asked him to confirm this he said a very definitive yes, but later backtracked saying he would rather not talk about the issue and that actually the first Fusion chips might have multiple CPU cores.
From this, I'd suggest a basic birth for the new processor. Getting one CPU and one GPU talking to each other on one piece of silicon is no small feat and this has to be the first step before you start adding multiple CPUs, crossbars and caches.
Additionally, talk of multiple GPUs on one die (internal-crossfire, say) should never become an talking point, since multiple GPUs on the same die are just an amalgamation of pipelines into a bigger stream processor, unlike multicore CPUs.
During a different part of his presentation, Amato spoke about the $100 one laptop per child initiative. Fascinatingly, if you delve into its specs it quickly becomes clear that AMD is already using its Fusion ideas in developing the $100 laptop CPU, the Geode GX-500.
It has an integrated graphics controller and unified memory architecture, so AMD already has some experience in doing CPUGPUs. A small number of these laptops are already knocking about, although they won't officially ship until 2008.
Finally, and we all know it's coming, Amato dropped some hints about ATI chips being made in AMD fabs. ATI, now know internally as AMD north since they're Canadians, will transition its stuff to AMD foundries "where it makes sense to do it". Originally AMD kept schtum on the issue, perhaps because it didn't want to piss off TMSC and UMC, which currently manufacture the graphics chips.
How Intel shrank processors to 45nm without taking a leak
Pictured left is a die shot of one of Intel's new 45nm Penryn processors, which the company claims represent the biggest breakthough since the sixties. Its development forced Intel to address one of the biggest problems of miniaturisation: leakage current.
A standard transistor of the type used in processors consists of source and drain electrodes sitting in a silicon substrate with a tiny gap between them. Above this gap is a thin layer of insulator, or dielectric; and sitting on that is the gate electrode 
(click image at right for pop-up diagram). Toggling the voltage at the gate toggles the charge distribution across the gap, and thus its ability to pass current.
This solid-state switch is never quite perfect because there are tiny current flows even in the off state. Most important is the "leakage" across the insulating layer under the gate. This layer is made of silicon dioxide (SiO2) in current designs and when it becomes only a few atoms thick, as it does as processor transistors get smaller, leakage becomes prohibitively high.
So why not have thicker insulation? The problem is that the thinner the layer, the higher is the capacitance of the structure - the amount of charge it can hold. The higher the capacitance, the better the current flow in the on state, and the faster the switching. In other words if you thicken the insulation to reduce leakage, you slow the transistor down.
What Intel has done is to replace the SiO2 with a 'high K dielectric', based on the element Hafnium, which allows a thicker (and thus less leaky) layer of insulation without reducing the capacitance.
Intel has also replaced the usual silicon gate with what it vaguely refers to as a mix of metals. Kaizad Mistry, product manager for Intel's 45nm logic technology development, said Intel was keeping this secret as the precise proportions of these and hafnium are critical.
The overall effect is to boost current flow in the on state, providing fast switching, and cutting leakage in the off state.
Intel claims that relative to 65nm technology the Penryn chips will pack twice as many transistors in a given area, with a 30 percent reduction in switching power, 20 percent faster switching, and a tenfold reduction in leakage across the gate dielectric.
It also claims a fivefold reduction in current leaking between the source and drain.
