by Ilya Gavrichenkov
04/03/2007 | 05:53 PM
We can argue endlessly whether dual-core and quad-core processors are so badly needed these days. Objective analysis suggests that far not all contemporary applications support multi-threading, so the both arguing parties – the parallelism adherents as well as opponents – have indisputable arguments at their disposal. In the end, the need for CPUs with multiple computational cores should be determined in each particular case individually, basing on the actual application field for the computer system in question.
<%BANNER[article]%>However, you should understand that if the clock frequencies of the single-core CPUs were twice as high as the clock frequencies of the dual-core CPUs, then there would be no hard choice to face. It is evident that in this case a single-core processor will be able to deliver at least the same performance level as its dual-core counterpart. However, the entire multi-core approach has actually emerged because it was no longer possible to increase the processor performance by simply raising the clock speed. So, there is no need to argue whether the future lies with multi-core processors: it is an evident fact. However, the CPU developers have been so enthusiastic about introducing their dual-core solutions into the market that they got down to forcing the logical evolutionary transition too hard. For example, Intel is saying that by the end of this year there will be only 5-10% of single-core processors left among the overall CPU shipments. And the saddest thing is that all these will be budget CPUs. So, high-end and mainstream system users can now only choose between dual-core and quad-core processors.
You can see Intel’s attitude to single-core solutions from the way they have been introducing Core micro-architecture. CPUs based on this micro-architecture have been shipping to the market for almost a year already, however there are no single-core processors based on it just yet. The first single-core solutions on new micro-architecture are due to come out only in June. These will be new Celeron 4XX processors. And although Intel was initially planning to also release mainstream single-core Core based processor family aka Pentium E1XXX, they have later transformed this idea into the promising Pentium E2XXX dual-core family.
So, it looks like Celeron 4XX is the last bastion of the single-core processors in Intel’s camp. So our today’s article will be devoted solely to these new processors, and has every chance to become a “single-core memorial” in the end. However, I decided to write this article not because “simplicity in everything” is my motto and I wanted to praise the advantages of single-core architecture. The thing is that I was lucky enough to get my hands on an engineering sample of the promising Celeron 4XX processor. And since Xbit Labs has been known for offering indepth previews of the new processor technologies, we just couldn’t leave this great opportunity unnoticed.
Celeron processors based on Core micro-architecture are scheduled to be officially launched in June. According to the available info, the new Celeron family will first contain only three models marked as 4400, 430 and 420. These processors will be priced very attractively: between $40 and $60. So, Core micro-architecture should get into the budget systems very quickly. Celeron 4XX processors will be based on another modification of Conroe core aka Conroe-L. The main distinguishing features of Conroe-L are single-core design, cut-down 512KB L2 cache and 800MHz Quad Pumped Bus. The first Celeron 4XX processors will run at relatively modest clock speeds ranging from 1.6GHz to 2.0GHz.
The Celeron 4XX engineering sample based on Conroe-L we got at our disposal didn’t have any marking on it and featured an unlocked clock frequency multiplier. We could notice the difference between this CPU and the solutions built on Conroe and Allendale cores with the naked eye. You can easily distinguish this processor from the others by the typical placement of electronic components on the bottom side of it.
We had some issues with the installation of the CPU into our test platform based on ASUS P5B Premium mainboard with BIOS version 402. This BIOS version wasn’t familiar with Conroe-L, so on system boot-up the CPU was recognized as Genuine Intel CPU, and after the POST stage the mainboard reported: «Intel CPU uCode Loading Error. Press F1 to Resume». However, after pressing F1 key the system would boot-up successfully and the system would operate flawlessly after that.
Unlike the BIOS, CPU-Z utility version 1.39 handled the situation better and reported absolutely correct info about our processor sample:
The utility recognized our processor as Celeron with Conroe-L codename. As you can see from the screenshot, our sample allows setting the clock multiplier at up to 12x, which allows us to have this CPU work at up to 2.4GHz frequency taking into account 800MHz bus. Looks like this sample has some hidden reserve for the “future needs”, because the maximum frequency of mass Conroe-L coming out in June will be only 2.0GHz.
The CPU supports 64-bit EM64T extensions present in all contemporary processors. However, some functions typical of the Conroe core will be disabled Celeron family, as usual. Budget CPUs will lose Virtualization Technology and most likely Enhanced Intel SpeedStep support.
So, if we compare the specifications of contemporary Celeron processors on NetBurst micro-architecture with those of the upcoming Celerons on Conroe-L, we will get the following picture:
Celeron D 3XX | Celeron 4XX | |
Processor core | Cedar Mill | Conroe-L |
Micro-architecture | NetBurst | Core |
Number of cores | One | One |
Hyper-Threading | None | None |
Intel EM64T | Yes | Yes |
Intel Virtualization Technology | None | None |
Enhanced Intel Speedstep | None | None |
Execute Disable Bit | Yes | Yes |
L2 cache | 512 KB | 512 KB |
Bus frequency | 533 MHz | 800 MHz |
Clock frequency | Up to 3.6GHz | 1.6-2.0GHz |
Packaging | LGA775 | LGA775 |
Production process | 65 nm | 65 nm |
TDP | 65 W | 35 W |
In other words, Conroe-L processors do not have anything impressive to boast. Of course, new micro-architecture and 800MHz bus are good things, but don’t you think that the clock frequencies Intel set for their budget single-core processors are somewhat too low? However, even during the comparison with “fully-fledged” dual-core processors we concluded that CPUs on NetBurst micro-architecture will have to run at 90-100% higher clock frequency in order to catch up with their counterparts on Core micro-architecture. If this rule also works for the budget Intel processors, then the new Celeron CPUs on Conroe-L core will turn out much faster than their predecessors based on Cedar Mill.
In order to confirm that the new Celeron 4XX will really be faster than the Celeron 3XX family, we decided to compare the performance of the top Celeron D on NetBurst micro-architecture with that of the upcoming Celeron 440 CPU, the fastest single-core representative of the Conroe-L based family. In other words, although our Conroe-L sample could be clocked at up to 2.4GHz, we set its working frequency at lower 2.0GHz value, because this is the top nominal clock frequency of the new processor family. Therefore, we reduced its clock frequency multiplier to 10x.
Note that the today’s fastest Celeron D processor is marked as 365 and works at 3.6GHz frequency. This was the rival we picked for our 2GHz Celeron 440.
Other than that the testbeds were configured as follows:
The obtained results are given in the table below:
The obtained results indicate undoubtedly that the transition of Celeron processors to Core micro-architecture will make them considerably faster. Almost in all tests Celeron 440 outperformed Celeron D 365 and in about half of all cases this advantage is more than 20%. Budget processors on Core micro-architecture are showing even higher results in office applications, games and image editing tasks.
Note that besides the higher performance, the upcoming Celeron processor can also boast much lower heat dissipation and power consumption. The typical heat dissipation for the CPU on Conroe-L core will be 35W, while Celeron D processors on 65nm Cedar Mill core feature 65W TDP.
So, the introduction of Core micro-architecture into the best value Intel processors is definitely great news for the end-users, which will make budget systems with Intel processors faster and more economical.
However, we shouldn’t overestimate the upcoming Celeron CPUs on Conroe-L core, either. If you are hoping to see the times of Celeron 300A will come back, we will have to disappoint you. The further, the bigger gets the performance gap between the fully-fledged and budget CPUs, which is the CPU developers’ policy. And Conroe-L will not break this tendency. Now Celeron can only compete with mainstream and high-end processors only in those applications that do not support multi-threading and their number grows smaller day by day.
However, there is still some hope that a wonder may happen. Maybe Conroe-L will prove extremely overclockable, which will allow us to forget about its single-core design? Let’s find out!
Note that Celeron processor on Conroe-L core looks very appealing for overclockers from theoretical prospective. There are a few indirect indications that this processor should overclock very well. First if all, dual-core processors on Conroe core can reach higher frequencies during overclocking if one of the cores is disabled. Conroe-L has no second core at all. Secondly, this processor boasts very low TDP, which may also be an indication of high frequency potential. Thirdly, rough estimates of the number of transistors in Conroe-L allow us to conclude that this is a relatively simple core. While dual-core Conroe consists of 291mln transistors, Conroe-L should have a little over 100mln transistors. And the simplicity of the semiconductor die is directly connected with its overclocking abilities.
However, we should move from theory to practice now. For our overclocking experiments on the Conroe-L based processor we used the same testbed as for performance testing. We used Zalman CNPS-9500LED cooler for our processor. System stability during overclocking was tested using SP2004/ORTHOS utility that has proven very efficient over the time.
Conroe-L was overclocked with the clock frequency multiplier set to 10x, i.e. we used Celeron 440 parameters as our startup value.
Note that thanks to 200MHz FSB Conroe-L based processors should be relatively simple to overclock, meaning that there will be a lot of mainboards out there that will suit for overclocking this CPU, as no extreme FSB frequencies need to be supported. Of course, we would like to hope that Celeron processors on Core micro-architecture will allow pushing the FSB frequency to 500MHz and higher, but the reality crosses over these dreams.
The thing is that by raising the Vcore to 1.6V we only managed to increase the FSB frequency of our Conroe-L based processor up to 300MHz.
In other words, this promising processor hit its maximum at 3.0GHz. Further investigation revealed that this is a true maximum for this CPU, which has nothing to do with the so-called “FSB wall” as well as with the mainboard features. For example, when we raised the multiplier to 12x, the maximum frequency when the mainboard remained stable dropped down to 250MHz.
So, the practical tests ruined all our rosy hopes for the wonders the new Conroe-L based processor could work. They proved to overclock even worse than Allendale based CPUs, which in their turn overclock worse than the regular Conroe based ones.
However, I have to say in favor of the new processor, that the 1.5 times frequency increase is not bad at all in the end. In fact, overclockers have always liked processors like that until Core micro-architecture came out.
In conclusion to the Conroe-L overclocking investigation, I would like to say a few words about its performance. The main question is: will the owners of the upcoming Celeron processors on Core micro-architecture be able to hit the same performance level as the contemporary dual-core CPUs can offer? To answer this question we compared the performance of the Conroe-L overclocked to 3.0GHz against that of Pentium D 925 and Core 2 Duo E4300.
The numbers speak for themselves. All in all, the overclocked Conroe-L loses to the dual-core Core 2 Duo E4300 showing better results only in those tasks that are not so well-optimized for multi-threaded environments and do not require large L2 cache memory. However, overclocking allows Conroe-L to perform as well as Pentium D 925. In almost half the benchmarks this dual-core processor on the outdated NetBurst micro-architecture runs slower than our single-core hero.
In conclusion I would like to add that the upcoming Celeron processors, if overclocked, will definitely outperform contemporary Celeron D on Cedar Mill core. The thing is that 5.0GHz is the maximum frequency contemporary budget CPUs can hit today, which is only 66% higher than what we can achieve on Conroe-L. And as we have already said before, NetBurst based processors need to run at 90-100% higher clock frequency in order to catch up with Core based processors in performance.
Well, it’s time to sum up what we have discussed today. When new processors codenamed Conroe-L come out this June, Intel will drive Core micro-architecture into the last untouched sphere – low-cost single-core CPUs. This is a pretty remarkable event, because this micro-architecture differs positively from the preceding NetBurst from multiple standpoints. The launch of Celeron processors on Conroe-L core will result into significant performance increase of the budget CPUs and simultaneous reduction of their TDP and power consumption.
It would be very pleasing for the end-users that the new Celeron processors will be selling at very attractive prices. Celeron 420 processor working at 1.6GHz will be offered for about $40, and the top single-core model on Core micro-architecture, Celeron 440 working at 2.0GHz frequency will be priced at around $60.
Unfortunately, despite all the advantages, the new Celeron processors will hardly be able to surprise overclocking fans. The maximum frequency these processors can hit with air cooling should be close to 3GHz, which seems a pretty modest result for CPUs on Core micro-architecture.
However, as you know this article is just a preview of the engineering sample of the upcoming product. In any case once Conroe-L is launched, we will return to this topic and maybe some of the conclusions made today will need to be revised.