Today the Net offers a lot of reviews on hardware with performance tests and short conclusions. Unfortunately, not all hardware sites are able to follow prices on-the-fly. And a user is to decide himself what ratio is better "price/performance" or "price/quality".
In my opinion such approach suits only printers or scanners, i.e. those peripherals that can be considered as a separate device. The CPU is a heart of the system, its performance greatly influences the performance of the whole computer. Therefore, the "price/performance" is a complicated issue. Besides, you have to use several configuration systems for an objective estimation.
Moreover, price for components changes much. That's why expediency of inclusion one or another component into a system depends as well.
Today I'd like to tell you about how to choose hardware nowadays (within several months forward). What will be in future, I don't know - so I don't want forecast for that time.
Note that I will mention only WHOLESALE PRICES. They are changing more efficiently than the retail prices. A reader can calculate himself an ideal retail price with a simple multiplication by a typical mark-on coefficient.
A potential buyer has two alternatives - to buy a new processor for already known motherboard, or to buy a new CPU and a new motherboard (the second case can be also to get a new case and a new memory as well). As a result, many users follow the second variant. Happy owners of Socket370 or Slot1 motherboards make an exception. These boards allow usage of FCPGA or SECC2 CPUs Intel correspondingly with frequency from 500-600 MHz and higher.
Intel or AMD?
What do we have to start with? There are two platforms - AMD and Intel with different sockets - Slot A and Socket A (Socket 462) on the one hand, and Slot 1 and Socket 370 - on the other hand. The both slots are already dying. Slot A died quickly and unremarkably. Lately, the prices for Socket A processors have fallen down and have become lower than that of the Slot A counterparts. Intel doesn't hurry, and today you can find nearly all Pentium III in Slot version with prices being much higher than that of Socket analogs. Anyway, variant with Socket is preferable, since such processor can be changed without touching a motherboard at least during the nearest 1.5 year. The chance is small, but the lower CPU voltage that can be given by a board, the higher is the chance. That's why it makes sense to buy the most modern models of motherboards.
The question - Intel or AMD? - is answered simply. If you are not a lover of Intel, then you must realize that AMD solution is preferable today. Not to be unfounded, I'll give you some references to the reviews like "AMD vs. Intel".
The Athlon on "old" and "slow" chipsets (KT133, SiS730, ALi MAGIK1) in games is a bit slower or equal to the Pentium III; on new ones (KT133A, AMD760) it gets faster. In programs intensively using FPU, such as 3D Studio MAX, Athlon is considerably faster than PIII (sometimes, 30% ahead).
Duron is nearly always much faster than Celeron: 15 - 30%.
Athlon and Pentium4 in different applications show completely different scores, but in general AMD 1200/133(266)/DDR is quicker than Intel 1500/100(400)/RDRAM. As optimized for P4 applications are appearing, this ratio can change in favor of Intel, but not in radical way.
Duron is slower than Athlon in various applications in 0-25%, in average 5-12%, and as the speed grows the gap increases this way: from 7% at 650MHz to 11% at 950 MHz.
EOL - End of Life.
More data on the market state:
1) The market doesn't have cheap Socket A boards - minimal price is $79 for a motherboard of AÒX form-factor, and $74 for MicroATX mainboard. At the same time, all these boards were released not so long ago, and in technical aspect are up-to-date. Socket370 (not necessary FCPGA!!!) boards start from $50. It's quite difficult to choose out a fresh, modern board that will be supporting all Intel CPUs (often VIA Cyrix III as well). Unfortunately, there are no integrated solutions for Socket A platforms, but by April-May the situation will change.
2) AMD processors, especially old Athlon's, consumes and dissipates more power than their Intel counterparts. There come out 2 peculiar features of assembling a computer based on the Socket A:
It raises the price of such case by $15-20 if Athlon is used. Besides, it's not recommended to use new models of Pentium with a cheap cooler and a power supply unit, but it's not admissible for Athlon (since AMD processors do not feature an integrated thermo sensor which allows in case of overheating to switch off the system).
3) "Cool" boards for Socket A platforms (e.g. allowing to change a multiplier from BIOS, change a FSB frequency with 1 MHz steps, and core voltage - with 0.025 V steps) in average is not more expensive than it's analogs with FCPGA-connector.
4) A range of Socket A mainboards is much more narrow than that of Socket 370 motherboards, but their quality is higher in average, and possibilities are more adequate to modern conditions.
5) The Intel Pentium 4 can be used together with memory of only Direct RDRAM type which costs 4 times more than normal SDRAM. Besides, the prices for P4 boards hover around $250-300, and as a result, the cost of one case is too high.
Taking into consideration all aforementioned facts, I can state that practically in all cases when buying a new machine or upgrading an old one by a replacing a motherboard it's beneficial to use AMD CPUs.
In this case we are limited in choice of a chipset: the today's market offers only VIA KT133, though the KT133A mainboards are apearing literally today. In the nearest months the market will get inexpensive integrated SiS730 and VIA KM133 chipsets. And the main thing is that we will see mainboards based on chipsets with a new memory type support - DDR SDRAM.
SDR or DDR?
This question started arising since release of motherboards with DDR support and the modules themselves. It happened not so long ago, and as usual the prices for them turned to be too high: the boards cost 1$70-190, the memory modules are three times more than SDRAM of the same capacity.
However, it's useless to assemble a computer based on DDR right now. But what, then, we are to do in March-April, when the boards and the memory will get cheaper? An answer to this question can be given after comparing the performance of DDR SDRAM and SDR SDRAM
As you can see, DDR gives from 1 to 8% performance gain, mainly in games and programs oriented for graphics, in average 3-4%.
Moreover, usage of PC1600DDR SDRAM makes no sense (that's why further we will talk only about PC2100DDR). Obviously, usage of DDR-memory is irrational until the prices get just a little higher than those for normal memory like SDR SDRAM.
Let's estimate. Assume, that in spring of 2001 DDR will be 4% faster than SDR, and in autumn - 6% faster (for example, there will appear applications exacting to the memory bandwidth). A good system case costs $800. If you are ready to pay 5% more in order to get 5% increase in performance it will make $40. It means that we are to pay in addition exactly such sum in order to get DDR SDRAM instead of a normal SDR and boards for this new memory. If a memory size is 128 MBytes (it's not enough for such system, though) it means two times performance gain (with today's price for 128 MBytes PC133 SDRAM of $44), and in case of 256 MBytes it will be 40% higher. And this is true considering that the mainboard for DDR will cost equally to the normal one, but today it's wrong...
If a case costs around $500, then, first, percentage of programs exacting to the system memory bandwidth is lower, and average speed increase can be considered to be 4%. Then, expediency of usage of DDR is estimated to be $20 and 45% of a normal SDR memory price under the condition of decrease in prices for DDR-boards down to the level of usual mainboards. But it won't happen in the nearest time.
So, we can get on with the price for DDR memory which is 0-30% higher than that for SDR. It means that the price for brand DDR memory should fall down to at least $62 and $126 for 128 and 256 MBytes correspondingly from today's $89 and 180 in order this type of memory could become an alternative for a buyer. I don't think it can happen before the summer ends. And until that time you can forget about DDR in a home system.
KT133 vs KT133A - is it worth to wait for a successor?
The majority of users won't take home an integrated chipset. As a rule, with it you will have to refuse an AGP slot, expendable slots and to decrease a system performance. Of course, such systems have their own market, mainly cooperative, but they are still meant for a home computer.
I will touch upon only two chipsets - KT133 and KT133A. Especially considering that during the nearest two months there won't appear more chipsets.
The KT133A differs from its predecessor only in full-value support for 133 MHz FSB. It results in gain from 0.5 to 8%, in different applications (5-6% in average). The KT133A performance more often is closer to the AMD760 with DDR memory than to its predecessor. And all this is desert of a 133 MHz bus.
The market still lacks for Athlon's for 133 MHz bus, though they were announced very long ago. It seems that AMD are waiting for mass appearing of the KÒ133A based boards. According to the data received, such processors are to appear in February, but only starting from 1 GHz.
Sooner or later, you will meet such concept as change of CPU multiplier. The default multiplier of all modern CPUs is strictly fixed, though in theory many mainboards have got such possibility. AMD processors allow changing it nowadays.
While in overclocking mode, cooling issue is of the greatest concern. A cooler should be chosen carefully. Beside a physical damage of a CPU you can also lose important data because of failure while overclocking.
Note that modern Duron can be overclocked in average by 150 - 250MHz (new models are worse than old ones), while Athlon's - by 70 - 200 MHz (here it's vice versa: new ones are better). Overclocking with FSB is 2 times effective than with a multiplier (FSB frequency sometimes can reach 140-155 MHz decreasing at the same time a CPU multiplier), since in the first case we are accelerating the whole system including the memory. With core frequency increase the Duron performance slows greater than of its older brother, because of the smaller size of L2 Cache.
Whether it's worth to overclock it's for you to decide. Although 650MHz Duron is sufficient for the majority of modern applications, for some programs overclocking is useful. For example, for 3D StudioMAX, MPEG4 encoding/decoding, speech recognition etc.
The ideal motherboard should possess the following features:
Such board must cost around $120-130 (in implementation without RAID), and by the end of March - $110-120.
The rule is simple: the bigger and more powerful - the better. However, the more powerful cooler, the more noisy and heavier it is. Turbine coolers didn't justified the hopes. If you look properly, you can find a more effective cooler, even with thermo paste, for less money.
An ideal cooler, in my opinion, must meet the following requirements:
Such cooler must cost at least $15.
Often, consumption current(A) is indicated right on the fan. The higher the value, the better.
In practice, we often meet the following coolers:
Special attention should be paid to safety of a CPU's crystal. Too tight installation of a cooler can lead to mechanical damage of a die, too soft one can result in fall of efficiency of cooling, and therefore can cause overheat. A fan must be intended exactly for Socket A!!!!
Special stress is to be laid on a general ventilation inside the case. It's desirable to have one-two additional fans beside the fan mounted in the power supply unit.
In principle, without overclocking you can take any PC133 memory. But the memory can refuse working at 133 MHz if the read/write latencies are less then the rated. The higher the memory quality, the higher frequencies and less latencies.
Ideal is 160 MHz in latency mode of 2-2-2. In practice, at 150 MHz and 2-2-2 the majority of modules with Micron and Hyundai chips, some modules on NCP and M-tec, Mira chips are able to work. This possibility depends on a certain lot.
How to determine this ratio? To my mind, the performance should be divided not by a CPU price but by a price for the whole system block. Otherwise, the cheapest CPUs turn to be the most beneficial, even if a system features 512 MBytes RAM, SCSI, video editing board, and possibilities for video and photo editing.
Moreover, a circle of tasks and implemented programs as a rule is different for cheap and expensive computers. That's why a set of programs for determination of performance factor will be different. I will show you how to give a preference to one or another CPU calculating a "price/performance" factor by different ways:
the rational one includes:
All the mentioned calculations are simplified, for example, they ignore such parameter as overclockability.
It's possible that for calculations you should take into consideration a price for a set consisting of a processor, motherboard, a cooler, memory, a video card and HDD.
If this article will interest you, the author
will share his ideas on the rest of components of a home PC.
Write a comment below. No registration needed!