Is it value or not? Choosing your processor.

[Catweazle]

The Inquirer has just published their recommendations for the ‘value points’ in processor purchasing. Their recommendations are quite similar to those promoted by a number of reviewers, but closer inspection leaves me quite unconvinced. The assumptions they make fall somewhat short of the mark in my eyes, and leaves their recommendations more closely allied to cost price than to purchase value. Let’s look at them in more detail, shall we?

Entry level

Like many specialist hardware sites, the Inquirer has sung the praises of the Celeron ‘D’ 336, criticized the Sempron, and touted the Athlon64 3000+ as the entry level chip of choice from AMD. The Celeron, they suggest, will run office applications fine, run some older games, and will provide an upgrade path to dual core Pentium processors to beef up the system later on.

Hang on a minute! That’s not necessarily correct. Seems to me that the argument they’ve used for the Celeron should be the argument used for the Athlon64 instead!

Don’t get me wrong. The Celeron ‘D’ is a pretty good chip, and a considerable improvement on earlier Celeron lines. But it’s not necessarily the path to the future that’s touted here. Not unless you use a motherboard with the i945 or i955 chipsets anyway, because other chipsets won’t accept the dual core processors that are supposed to form the upgrade! A Celeron ‘D’ with a cheapish all-in-one motherboard is a fine system for people who are not going to work the system very hard, but that’s about it.

There’s also performance to consider when making a judgement about ‘value’. Price isn’t everything. I’m in agreement with the Inquirer about the Athlon64 3000+ being the ‘value’ entry level processor from AMD. But at 2.8GHz that Celeron 336 is not a competitor to the Athlon. To get a comparable level of performance you’d need to purchase a Celeron ‘D’ 351, and that brings the prices a lot closer. There’s only a small cost margin between Celeron 551 and Athlon64 3000+, and if you add in the extra cost of late model motherboard and more expensive DDR2 RAM to ensure that upgrade path to dual-core processing later on, then I’m afraid the Celeron falls behind as a ‘value’ purchase!

Dual core processing

Again, the Inquirer has placed cost above all, and produced a somewhat skewed depiction. The Intel Pentium D 820 is a ‘value’ entry to dual core processing, they consider, whilst AMD do not provide a ‘budget’ entry level processor. If you look only at purchase price you might get that impression, but let’s look at the performance of the lower rated dual core chips as well. The Pentium D 820 is a 2.8GHZ single core performance chip. The lowest rated AMD X2 dual core processor, the X2 3800+, gives single core performance about equivalent to an AMD64 3000+.

To get equivalent performance, you need to jump up a level to the Pentium D 830 processor, which comes at a price roughly equivalent to the entry level Athlon X2. And again, you need a motherboard with a particular chipset, together with more expensive (and no better performing) DDR2 RAM modules! Just about any Socket 939 motherboard will accept Athlon X2 dual core processors. At worst, a BIOS update will be all that’s required. People who’ve purchased cheaper Intel systems with a view to changing to dual-core later on could well find themselves without the upgrade path they’d counted on.

When choosing components, don’t get too carried away with price. Too many reviewers seem to. The entry level Intel processors are underpowered compared with their competitors! If it’s a new system you’re planning and it has ‘Intel Inside’, then I’d suggest you plan on getting a system in which the ‘Intel Inside’ is taken up a notch!

[jwenting]

Where is longevity in the equation?
AMDs on average don't last nearly as long as do Intels, inducing a far higher lifetime cost for a system that's not built to be replaced as soon as the next fad hits the market.

Intels can last 5-10 years, AMDs often don't go strong for more than 2-3.
Fine if you're looking to replace your system after 1-2 years, not so nice if you want it to run reliably for 5 years or more.
Same with stability. AMDs have a tendency to run hot (which is a large contributing factor to their shorter MTBF) which also makes them more prone to causing the system to hang or crash under heavy load.
Again not so much a problem if you run the machine for only a few hours a day, but if it has to stay up for weeks or months at a time you want something more reliable.

Of course people wanting such reliability will typically not choose entry level CPUs.

[goldeagle2005]

I beg to differ. I find the AMD more stable than the intel counterparts. My old Athlon XP 2000+ ran like a kitten for 6 years. I'm a hardcore computer user which meant that I left the machine on for 5-6 days at a stretch using the stock heat sink and overclocked to 2GHz, with a downtime of maybe 3 hours for maintenance. I've seen intel CPUs crash and reboot 2-3 times in a day. Not so with my AMD. Also, the intel prescotts are notorious for running hot.

I know people may disagree with this, but this has been my experience.

[Catweazle]

You can't cling to outdated old wives tales forever, I'm afraid. Back in the days of the earliest Athlon XP processors it was fair comment to say that Athlons ran a tad 'hot'. With the original Athlon AMD 'beat' Intel to the 1GHz barrier and beyond, but with a somewhat hotter running chip. Even the first release or two of the Athlon XP processor ran somewhat hotter than the older Pentiums. In some countries (my own Australia a prime example) there were an inordinate number of component 'failures' which resulted from chips being shipped accompanied by heatsink units which were not adequate for local weather conditions, and the company had to adjust its practices accordingly.

Regardless of this factor, I've never seen any evidence to suggest that, in the general run of things, Athlon processors enjoyed any less a MTBF compared with Pentiums.

The release of the Barton core Athlon XP processors saw the situation improve markedly for AMD, and since the release of the AMD Athlon64 and the Pentium Prescott processors the situation has turned a full 180 degrees. Nowadays AMD make the power-efficient cooler running processor, while Intel suffers heat problems. Yes, that's right. If you think that higher temperatures mean a shorter life then choose AMD for sure, because the current Intel Pentiums are hotter running!

It's 'fashionable' and 'elitist' to sneer at AMD processors (that's a general comment, not a personal one) on the basis that they are the 'poor cousin' and thus not as good. But fashionable and elitist though that outlook may be, it's not a correct one.

Intel is working pretty damn hard on their processor lines to correct the power consumption and related heat problems. Futire models will be quite efficient indeed. But they're not there yet!

[JJ122]

I dont know about you all but the semprons aren't worth the silicon they are made on, there is a massive difference in processing power between them and the normal processors, and the price difference isnt worth it. I have used simmilar machines the only difference is the processor, same speed just one is a sempron and the other isnt and it took as much as 3 times the time to load programs.

There is 4 times the L2 cashe on the normal processors over the sempron versions. and its only about a $50 difference

Also, in my experience the AMDs I have used have had many more problems with them then the intels. For months I couldn't get my 3200+ to work at 2.2 GH only at 1.1 reliably.

Plus the AMDs cant hold a flame to the new intels with the net burst tech and the mobile platform they are built on. So there is minimal heat and power consumption

[jwenting]

AMDs come overclocked out of the factory (or used to, not sure if they still do that).
Many of them are marginal for the speed they're sold for, while Intel incorporates far higher tollerances in their system of classing CPUs for a specific speed.

The system is (simplified) as follows:
A production batch is tested for the maximum speed they can hold without failing (or overheating). That's the speed the batch is then sold as. A different batch will have slightly different material properties affecting their maximum speed.
Intel sets their margins wider than does AMD, so their CPUs are less likely to fail at the published speed.
This test is within a speed range of course, determined by the design speed limits of the CPU core.

This system of determining CPU speed causes shifts in available supplies of course.
As Intel gets better at creating their wafers, the average speed of the produced CPUs of each design goes up and thus the supply of lower speed CPUs drops.

[Catweazle]

It's rather staggering to see the amount of misinformation that people base their opinions on.

Whilst I'd agree that it's never been a good idea to purchase Sempron (the newer A64 core based Semprons are far superior to the older Thoroughbred core based Semprons, by the way, although many people are no doubt unaware that different models have even been issued) the comment about '4 times the L2 cache' is not accurate. Compare some Semprons with some A64 chips and you'll see the difference is a factor of 4. Compare others and you'll see it's a factor of 2. AMD alters the amount of L2 cache as part of the process of setting PR levels for their chips, rather than simply altering clockspeed.

'Net burst' is the architecture of the older P4 platform, not the chips which are loosely based on the PentiumM architecture.

And neither Intel nor AMD can be claimed to 'overclock out of the factory'. Whilst jwenting is roughly accurate with his description of how processors are manufactured and tested, his contention that AMD follows some sort of policy of setting tighter 'margins' than Intel is misleading.

All processor cores of a particular type are manufactured in the same production line, regardless of the speed they'lll end up being sold at. They're then tested, and as jwenting has described the 'batches' are sorted according to performance and used in CPU chips with their performance set in accordance with the test results. That's the basic idea of the process anyway. In reality, both Intel and AMD have prgressed to the point where the vast majority of the silicon 'wafers' they create CPUs from produce cores of a high standard, and often batches are allocated to lower performance ratings to meet orders. The process doesn't generally go the other way round, with cores being used for CPUs at a level where they're straining to meet the performance asked of them.

In general, lower 'speed' chips of the same type will often overclock quite well, because they're often 'better' cores set at a lower speed to meet orders. In general, higher 'speed' chips will usually overclock less well, because yjey are pushing the limits of the architecture and the production process. That's equally true of both Intel and AMD.

It's basically demand that determines what speeds chips get set at, rather than the production process which effects how many chips of particular speed ratings are available, until you reach the higher ranks of the product line, where less cores will meet the requirements regardless of which platform you're discussing.!