PDA

View Full Version : I am completely detached from AMD...please help me get back on the wagon



user1397
May 31st, 2011, 04:07 PM
So I used to have a pretty good idea on comparable processors between AMD and intel, but since around the release of the core 2 duo I've stopped looking at AMDs that much and now I basically have no idea how their procs compare.

For example:

AMD Athlon XP = intel Pentium 4

so what about now? like:

core i3 = ?
core i5 = ?
core i7 = ?

and so on (I'm looking for rough comparisons, hence no particular submodels between procs)

3Miro
May 31st, 2011, 04:19 PM
The current top of the Linux AMD processors are the Phenoms II.

Athlon II - low end machines (cores vary from 2 to 4)
Phenom II X4 - good performance
Phemon II X6 - highest end, it competes with i7, although AMD has not yet released an answer to the second generation i7

X4 to X6 the difference is big, not just in number cores, but also X6 give less heat and are overall more efficient.

Phenom II x6 usually beats Intel from the same price range (especially on the multi-threadis as it has 6 real cores and not 4 real + 4 hyperthreading), however, Intel has faster and more expensive CPUs.

Dustin2128
May 31st, 2011, 04:22 PM
So I used to have a pretty good idea on comparable processors between AMD and intel, but since around the release of the core 2 duo I've stopped looking at AMDs that much and now I basically have no idea how their procs compare.

For example:

AMD Athlon XP = intel Pentium 4

so what about now? like:

core i3 = ?
core i5 = ?
core i7 = ?

and so on (I'm looking for rough comparisons, hence no particular submodels between procs)
I'd compare a core 2 to maybe one of the first few athlon dual cores? It's a fairly old processor.... i3-i7 match up with the phenom II line, having largish level 3 caches, and the Athlon IIs match with low end pentium dual cores and stuff- no L3 cache at all.

psusi
May 31st, 2011, 04:31 PM
I've been an AMD fan for many years, ever since my dual Athlon MP 1.2 GHz system. This February though, I had to go with Intel when upgrading my aging system. Their core i5/6/7 was already quite a bit ahead of AMDs offerings, and then the sandybridge just left AMD in the dust. My core i5 2500-K is incredibly fast, and power efficient. With just the stock cooling fan I'm able to run it at 4.3 GHz without a problem.

slooksterpsv
May 31st, 2011, 04:33 PM
The i7's I compare them to the Llano cores, yes still awaiting the arrival on NewEgg and in PCs, but you can't go wrong when you have a CPU and an AMD Radeon HD 6550M embedded on the CPU (APU rather).

Yeah in terms of "processing power" the i7 will be faster to a degree. But CPU speed is all hyped at the moment, the main focus, in my opinion, has shifted to Hard Drives as they are what's holding back the rest of the hardware from being used to their potential.

Say I'm wrong or right, but cost-effective disk drives that perform well and are large in size are non-existent at the moment (unless you know how to RAID).

<3 AMD <3

Dustin2128
May 31st, 2011, 04:34 PM
I've been an AMD fan for many years, ever since my dual Athlon MP 1.2 GHz system. This February though, I had to go with Intel when upgrading my aging system. Their core i5/6/7 was already quite a bit ahead of AMDs offerings, and then the sandybridge just left AMD in the dust. My core i5 2500-K is incredibly fast, and power efficient. With just the stock cooling fan I'm able to run it at 4.3 GHz without a problem.
How much did it cost you?

psusi
May 31st, 2011, 05:08 PM
How much did it cost you?

Had to go back through the order history.. looks like on 1/18 I blew about $650 on the cpu, motherboard ( Asus P8P67 Pro ), 4 GB of DDR3-1600 ram, a nice new case, and a blue ray burner.

screaminj3sus
May 31st, 2011, 05:36 PM
Right now intel has better performance across the board. If you go amd wait until bulldozer. Right now I'd get intel.

3Miro
May 31st, 2011, 05:52 PM
Right now intel has better performance across the board. If you go amd wait until bulldozer. Right now I'd get intel.

Intel has the better performance yes, but not better performance per dollar. Depending on how big of a factor the budget is, AMD can be very competitive. I would much sooner go for Phenom II 4 cores than an i3. Intel even has Core 2 Duo CPUs for 200 dollars and you can get a Phenom II X6 for the same money. 6 cores and L3 cache vs 2 cores and only L2 cache, what is Intel thinking (or the people that are buying it)?

Also note that AMD moherboards also seem to be cheaper on average, although since they use different chipsets we may be comparing apples and oranges.

mips
May 31st, 2011, 05:54 PM
right now intel has better performance across the board. If you go amd wait until bulldozer. Right now i'd get intel.

+1

Dry Lips
May 31st, 2011, 07:20 PM
I've got an AMD Sempron 145! (2.8Ghz) That would match... Intel Atom?




Say I'm wrong or right, but cost-effective disk drives that perform well and are large in size are non-existent at the moment (unless you know how to RAID).
Are software RAIDs (say 0 or 1) faster than normal HDDs?

KingYaba
May 31st, 2011, 07:23 PM
Right now intel has better performance across the board. If you go amd wait until bulldozer. Right now I'd get intel.

I wouldn't buy at all. I would wait for Bulldozer to come out which should be some time pretty soon.

slooksterpsv
May 31st, 2011, 07:31 PM
I've got an AMD Sempron 145! (2.8Ghz) That would match... Intel Atom?


Are software RAIDs (say 0 or 1) faster than normal HDDs?

Sempron? That would compare to a low-end Pentium, high-end Celeron. Semprons are so much better than Atom (crapatom as I call them).

As for a SW RAID I don't know... anyone know that one?

del_diablo
May 31st, 2011, 09:24 PM
I've got an AMD Sempron 145! (2.8Ghz) That would match... Intel Atom?

It is actually quite easy to find out: Can you run Conkers Bad Furday via Project64 on it without large frameskips?
If yes, it outruns it by a really large margin.
1.6GHz atom is roughly the performance of.... 1.7GHz Pentium 4 or something?

psusi
May 31st, 2011, 09:52 PM
Are software RAIDs (say 0 or 1) faster than normal HDDs?

Of course; that is the whole point of raid0.

slooksterpsv
May 31st, 2011, 10:12 PM
It is actually quite easy to find out: Can you run Conkers Bad Furday via Project64 on it without large frameskips?
If yes, it outruns it by a really large margin.
1.6GHz atom is roughly the performance of.... 1.7GHz Pentium 4 or something?

I do not believe that, personally. I had an Eee PC with an Intel Celeron 2.4GHz way underclocked to 680MHz, I upped it to 900MHz and it felt faster than a 1.66GHz Intel Atom. That's just me though.

Actually I may be right too according to these benchmarks =P:
http://www.tomshardware.com/reviews/Intel-Atom-Efficient,1981-15.html

Dry Lips
May 31st, 2011, 10:25 PM
Of course; that is the whole point of raid0.

I thought raid0 was about making two HDDs appear as one (say 1TB+1TB=2TB),
whereas raid1 was a way of having a backup if one disk fails? (1TB+1TB=1TB).

How much increase in speed do you get from using software-raid then?

szymon_g
May 31st, 2011, 11:31 PM
1.6GHz atom is roughly the performance of.... 1.7GHz Pentium 4 or something?

single core 1.6 ghz Atom has performance lower than 1.2 ghz Pentium 3...
while 1.33 athlon xp (1500+) was faster (not in everything) than 1.8/2.0 ghz Pentium 4.



I thought raid0 was about making two HDDs appear as one (say 1TB+1TB=2TB),
whereas raid1 was a way of having a backup if one disk fails? (1TB+1TB=1TB).

How much increase in speed do you get from using software-raid then?

raid0- stripping- should make reading/writing faster than from 1 hdd
raid1- mirroring- should make a "backup" (ok, ok, i know- raid isn't a backup) of one hdd, so if one hdd fails, system still can be up & running (and failed hdd can be replaced)

realistically- you could gain up to 70% in read/writing speed /using raid0/. but in most cases- you wouldn't notice a difference /i.e. you would notice difference only if you would work with large files etc/.

on linux, i suggest using raid10 :)

disabledaccount
May 31st, 2011, 11:52 PM
raid0- stripping- should make reading/writing faster than from 1 hdd
raid1- mirroring- should make a "backup" (ok, ok, i know- raid isn't a backup) of one hdd, so if one hdd fails, system still can be up & running (and failed hdd can be replaced)

realistically- you could gain up to 70% in read/writing speed /using raid0/. but in most cases- you wouldn't notice a difference /i.e. you would notice difference only if you would work with large files etc/.

on linux, i suggest using raid10 :)Atom is a crap - everyone knows this.... - unfortunatelly - because Intel micro-ATX are very energy efficient. But on the other side there is ION...

Myth: Raid0 is for linear read only - that would be true if prefetch/preload would't exist. Of course speed gain depends on load characteristic, but Raid0 is almost always faster than single HDD. Secondary: Linux MD raid1 == Raid0 while reading (almost, - there are differences under different load conditions).

Linux Raid10 is a champ, especially in "override" configurations, eg Raid10 on 2 disks - due to specific distribution of stripes it can drastically reduce seek times - and this is the key for real-life performance when using using classic HDDs.

szymon_g
June 1st, 2011, 12:05 AM
Atom is a crap - everyone knows this.... - unfortunatelly - because Intel micro-ATX are very energy efficient. But on the other side there is ION...

... but still, even in ION, atoms are beaten by amd fusion APU (i don't know how well it is supported under linux, though)


Linux Raid10 is a champ, especially in "override" configurations, eg Raid10 on 2 disks - due to specific distribution of stripes it can drastically reduce seek times - and this is the key for real-life performance when using using classic HDDs.

you can also use odd amount of drivers (for something like raid1e - i.e. benefits of raid0 and raid1 on odd number of drivers)

PhillyPhil
June 1st, 2011, 12:08 AM
Right now intel has better performance across the board. If you go amd wait until bulldozer. Right now I'd get intel.

Well, that's just completely inaccurate. Intel has better performance at the *top end*, above the Phenom II X6. Below that they both have processors spread across the performance range, with the only difference being that in general the AMDs have more cores and cost less.
I've been an AMD fan for many years, ever since my dual Athlon MP 1.2 GHz system. This February though, I had to go with Intel when upgrading my aging system. Their core i5/6/7 was already quite a bit ahead of AMDs offerings, and then the sandybridge just left AMD in the dust. My core i5 2500-K is incredibly fast, and power efficient. With just the stock cooling fan I'm able to run it at 4.3 GHz without a problem.

February? Should have waited a few months for bulldozer ;)

@OP cpubenchmarks is always a good site for at least getting a general idea about the comparison of the current cpu selection.
As you can see there, the AMD Phenom II X6 1100T sells for $200 and outperforms some i7 models that sell for a much higher price.
http://www.cpubenchmark.net/

Zerocool Djx
June 1st, 2011, 12:42 AM
The current top of the Linux AMD processors are the Phenoms II.

Athlon II - low end machines (cores vary from 2 to 4)
Phenom II X4 - good performance
Phemon II X6 - highest end, it competes with i7, although AMD has not yet released an answer to the second generation i7

X4 to X6 the difference is big, not just in number cores, but also X6 give less heat and are overall more efficient.

Phenom II x6 usually beats Intel from the same price range (especially on the multi-threadis as it has 6 real cores and not 4 real + 4 hyperthreading), however, Intel has faster and more expensive CPUs.


Ahhh,.. but wait there is more,.. behold the Opteron series 12 core processor!

http://www.google.com/products/catalog?q=AMD+12+core&oe=utf-8&client=firefox-a&cid=12264112351612238583&os=tech-specs

Personally,.. I think AMD is rivaling Intel atm. they got a name yes, but I am by far satisfied with AMD and how far they come. Intel just don't have it in them to push the envelope anymore.

And here is the mother board this beast goes into:

http://www.google.com/products/catalog?q=AMD+12+core&um=1&ie=UTF-8&tbm=shop&cid=13045808405302804920&sa=X&ei=LXzlTey4M6Xt0gGZu_WkBw&ved=0CNcBEPMCMAk

Here is the budget friendly version:

http://www.google.com/products/catalog?q=AMD+12+core&cid=2467799182244657517&ei=an3lTYXOGonUwwX3puzkCw&sa=image&ved=0CA0Q8gI

psusi
June 1st, 2011, 01:48 AM
@OP cpubenchmarks is always a good site for at least getting a general idea about the comparison of the current cpu selection.
As you can see there, the AMD Phenom II X6 1100T sells for $200 and outperforms some i7 models that sell for a much higher price.
http://www.cpubenchmark.net/

When I was shopping I went by the benchmarks from tomshardware, and iirc, the sandybridge core i5 and i7 put the phenom II to shame.

I can tell you that my 4 core i5 compiles things about 6x faster than my old Athlon64 5000+ black edition, despite having a very similar clock speed ( 3.2 vs 3.3 GHz ) and 2 cores.

slooksterpsv
June 1st, 2011, 02:40 AM
When I was shopping I went by the benchmarks from tomshardware, and iirc, the sandybridge core i5 and i7 put the phenom II to shame.

I can tell you that my 4 core i5 compiles things about 6x faster than my old Athlon64 5000+ black edition, despite having a very similar clock speed ( 3.2 vs 3.3 GHz ) and 2 cores.

See a lot of people do compile times for performance ratings, which I don't believe is fully accurate, same with encoding, video, gaming, etc. I think it's a combination of everything you can throw at it. This is me still, but I believe a true benchmark would have the CPU doing multiple things with multiple threads all at the same time. Cause single task for highest grade of x isn't always going to show how powerful a CPU is.

That's just my opinion. The single-tests show nothing, but threaded-multiple operation tests shows a better overall picture.

LowSky
June 1st, 2011, 02:59 AM
I have a Phenom II 550 x2.. Unlocked to 4 cores and overclocked to 3.5Ghz.
I paid $100 for that processor. Explain to me why I need to buy a Intel Core i7 at $500 to run word processing or record a TV show or play Civilization 5.


Todays computers the bottle neck is storage. I purchased a SSD and things became much faster overall.

The benchmarks for cpu testing is kinda crazy. Most of the tests are over the top and nearly never done by normal users. Even gaming is basically totally relying on graphic cards these days. Not to mention the test applications are usually developed to favor one chip designer over another form the start.

psusi
June 1st, 2011, 03:42 AM
See a lot of people do compile times for performance ratings, which I don't believe is fully accurate, same with encoding, video, gaming, etc. I think it's a combination of everything you can throw at it. This is me still, but I believe a true benchmark would have the CPU doing multiple things with multiple threads all at the same time. Cause single task for highest grade of x isn't always going to show how powerful a CPU is.

That's just my opinion. The single-tests show nothing, but threaded-multiple operation tests shows a better overall picture.

In general I agree, though technically compiling is a multi threaded multiple operation. You have several instances of the compiler, the assembler, and the linker all running at once. It also is the load I care most about since it is the CPU bound load that I do the most.


I have a Phenom II 550 x2.. Unlocked to 4 cores and overclocked to 3.5Ghz.
I paid $100 for that processor.

I suppose that is $100 cheaper than a core i5, and probably almost as fast, though it does rely on using cores that failed the manufacturer test, which is unreliable, and the core i5 uses about 40% less power, giving it more room for overclocking ( 4+ GHz with stock cooler ).

IIRC, Intel is 2 generations of fab process adhead of AMD. The drastically smaller gate size is a significant advantage.

3Miro
June 1st, 2011, 03:44 AM
When I was shopping I went by the benchmarks from tomshardware, and iirc, the sandybridge core i5 and i7 put the phenom II to shame.

I can tell you that my 4 core i5 compiles things about 6x faster than my old Athlon64 5000+ black edition, despite having a very similar clock speed ( 3.2 vs 3.3 GHz ) and 2 cores.

Are you serious? Athlon64 came out shortly after Pentium 4 and before Pentium D. That was about 6-7 years ago. You are comparing it to a modern CPU? I recently got a used Atholon64 computer for 30 dollars off craigslist.

The i5 and i7 outperform the Phenom II x4, but the x6 beats them on multi-threaded applications, which is what I use for work.

As I said before, overall Intel are faster, but unreasonably expensive for the extra performance that you are paying.

@Zerocool Djx: Opterons are also awesome, but too expensive for me.

Dustin2128
June 1st, 2011, 04:15 AM
I agree that hard drives are the main bottleneck in computers right now. As for why people need CPU power above what a Phenom II x6 can offer... I dunno. Maybe serious movie and 3D editing, but that's about it.

PhillyPhil
June 1st, 2011, 04:29 AM
When I was shopping I went by the benchmarks from tomshardware, and iirc, the sandybridge core i5 and i7 put the phenom II to shame.

I can tell you that my 4 core i5 compiles things about 6x faster than my old Athlon64 5000+ black edition, despite having a very similar clock speed ( 3.2 vs 3.3 GHz ) and 2 cores.

You can't compare an Athlon to an i5, of course the i5 is faster.

What model i7 are you talking about? There are many different ones.
Here's a list (probably incomplete?) of i7s in the ranking order cpubenchmarks gives them, and the top couple of Phenom II x6s:

i7:

995x
990x
980x
970
2600k
2600
2920xm
2820qm
2600s
x880
2720qm
975
965
960
875k
880
950

Phenom II x6 1100T

2630qm
2635qm
940
870

Phenom II x6 1090T

930
920
860
870s

I tried to find comparisons at tomshardware, but could only find an x4 vs an i7, not an x6: http://www.tomshardware.com/reviews/overclock-phenom-ii,2119-8.html

I have no particular problems with Intel btw. The cpus in the three computers I use and own are an atom, an i5, and an Athlon II x2.
But IMHO AMD is much better value for money, and it's simply false to say Intels entire range is better than AMD, when it's actually only the top end that can beat all AMD comers.

psusi
June 1st, 2011, 03:24 PM
As I said before, overall Intel are faster, but unreasonably expensive for the extra performance that you are paying.


Again, if you are comparing it to an AMD sold at a discount because it is broken, which you then are able to apparently get working. Not really a fair comparison. Also this is at the stock clocks, and sandybridge has much more overclocking room. Once you factor those in, the picture looks quite different.


I have no particular problems with Intel btw. The cpus in the three computers I use and own are an atom, an i5, and an Athlon II x2.

The one problem I know have with Intel is their cpu fans. They REFUSE to spin down. No matter what the PWM signal says, they are hard wired to have a minimum speed of about 30%. I miss the absolute quiet of my amd system.



But IMHO AMD is much better value for money, and it's simply false to say Intels entire range is better than AMD, when it's actually only the top end that can beat all AMD comers.

Of course AMD positions themselves to be a better value for the money on the low end; they don't have much choice. I thought the question was which is better in the mid to high end?

3Miro
June 1st, 2011, 04:39 PM
Again, if you are comparing it to an AMD sold at a discount because it is broken, which you then are able to apparently get working. Not really a fair comparison. Also this is at the stock clocks, and sandybridge has much more overclocking room. Once you factor those in, the picture looks quite different.


Are you serious or just trolling? I have been using AMD CPUs for years and never had a broken or defective one.

I don't do over-clocking, it is a very bad idea for scientific computing (and it probably is a bad idea in general). Here are some people that do over-clock AMD CPUs:

http://www.youtube.com/watch?v=z6Hf6d404QY

Grenage
June 1st, 2011, 04:53 PM
Are you serious or just trolling? I have been using AMD CPUs for years and never had a broken or defective one.

I don't do over-clocking, it is a very bad idea for scientific computing (and it probably is a bad idea in general). Here are some people that do over-clock AMD CPUs:

http://www.youtube.com/watch?v=z6Hf6d404QY

I assume he's talking about a Phenom x2, which is a x4 with two disabled cores that didn't make the cut in production. Sometimes you can enable them and they work, sometimes they are even stable. A lot of them are perfect x4 units.

psusi
June 1st, 2011, 04:57 PM
Are you serious or just trolling? I have been using AMD CPUs for years and never had a broken or defective one.


I was referring to the fact that AMD decided that two of the cores were broken, and sold it at a discount as a two core cpu. If you switch them back on, maybe they work, maybe they don't, or maybe they appear to work but you get odd random glitches.

3Miro
June 1st, 2011, 04:59 PM
I assume he's talking about a Phenom x2, which is a x4 with two disabled cores that didn't make the cut in production. Sometimes you can enable them and they work, sometimes they are even stable. A lot of them are perfect x4 units.

I was talking about Phenom II x4 and Phenom II x6, those are the top of the line AMD CPUs, read my earlier posts. I don't know what the deal is with x2 and x3, I heard some people try to get more cores out of those, but I am not one of them.

As I stated earlier, Phenom II x6 on newegg is about the same price as some of the Core 2 Duo models. There is no way a Core 2 Duo can outperform a Phenom II x6 (stock clock or overclock).

3Miro
June 1st, 2011, 05:01 PM
I was referring to the fact that AMD decided that two of the cores were broken, and sold it at a discount as a two core cpu. If you switch them back on, maybe they work, maybe they don't, or maybe they appear to work but you get odd random glitches.

I own Phenom II x4 and x6, I have those running on full power for many hours at a time, both compiling Gentoo and doing scientific computing for my job. I never had a CPU glitch.

Grenage
June 1st, 2011, 05:03 PM
I was talking about Phenom II x4 and Phenom II x6, those are the top of the line AMD CPUs, read my earlier posts. I don't know what the deal is with x2 and x3, I heard some people try to get more cores out of those, but I am not one of them.

As I stated earlier, Phenom II x6 on newegg is about the same price as some of the Core 2 Duo models. There is no way a Core 2 Duo can outperform a Phenom II x6 (stock clock or overclock).

I was merely covering what I read as a misunderstanding; the x2 and x3 can net you an x4 if you're lucky. I use both intel and AMD processors, and AMD cost/performance is without a doubt better, but sometimes you want that extra performance.

FuturePilot
June 1st, 2011, 05:31 PM
Intel overall has better performance right now. But if you pick a price and compare what you can get from Intel and what you can get from AMD, AMD will most likely beat whatever Intel is offering at that price. The i7 though is better than the X6. The highest i7 that you can get blows the X6 out of the water, but then you'd have to be willing to pay $1,000 for it. That said, I have a Phenom II X6 1075T and I love it.

psusi
June 1st, 2011, 07:47 PM
I was talking about Phenom II x4 and Phenom II x6, those are the top of the line AMD CPUs, read my earlier posts. I don't know what the deal is with x2 and x3, I heard some people try to get more cores out of those, but I am not one of them.

As I stated earlier, Phenom II x6 on newegg is about the same price as some of the Core 2 Duo models. There is no way a Core 2 Duo can outperform a Phenom II x6 (stock clock or overclock).

My bad, it was LowSky that was comparing the price of an x2 and unlocking the disabled cores.

wizard10000
June 1st, 2011, 09:04 PM
realistically- you could gain up to 70% in read/writing speed /using raid0/. but in most cases- you wouldn't notice a difference /i.e. you would notice difference only if you would work with large files etc/.

The reason for this is that although throughput is higher on a stripe set track to track latency is also higher.

Some years ago I build a hardware SCSI U160 RAID5 setup for my desktop PC using three 10k rpm 9.1G SCSI drives and a fairly expensive Adaptec RAID controller. Since most desktop PC work is opening small files, a single 40GB 7200 rpm BM Deskstar smoked that array in real-world use.

I used to be a huge AMD fan but gave up on them around the Athlon XP as AMD had decided to economize by not including thermal protection on the processor - and that particular chip had a heatsink that could be tricky to get seated properly on the processor. If you didn't get it seated properly it'd let the magic smoke out of the chip in about ten seconds - it never happened to me but did happen to a couple of friends of mine.

I did build a triple-core Phenom for my BIL a couple months back and was lucky enough to get one where the fourth core could be unlocked. Great bang for the buck.

PhillyPhil
June 2nd, 2011, 12:27 AM
Of course AMD positions themselves to be a better value for the money on the low end; they don't have much choice. I thought the question was which is better in the mid to high end?
AMD can match Intel performance at any point in the range up to the X6 1100T. That means there are only 17 x i7 models in Intels entire cpu range that can't be matched by AMD. So it's true to say that Intel performs better at the very top end, but not anywhere else.

Intel overall has better performance right now. No, better top end performance.
But if you pick a price and compare what you can get from Intel and what you can get from AMD, AMD will most likely beat whatever Intel is offering at that price. Right. If I had a set budget for a cpu, $200 or less, I would probably never even consider an Intel chip.
The i7 though is better than the X6. You'll have to be a lot more specific. Which x6 model, and which i7 model? There are certain x6 models which out perform certain i7 models (see my list on page3).
The highest i7 that you can get blows the X6 out of the water, ...
Yes, it certainly does, but lower i7s have worse performance than the top x6.

psusi
June 2nd, 2011, 03:42 PM
The reason for this is that although throughput is higher on a stripe set track to track latency is also higher.

Huh? How do you figure that? Latency isn't going to be any better, but it isn't any worse either.


Some years ago I build a hardware SCSI U160 RAID5 setup for my desktop PC using three 10k rpm 9.1G SCSI drives and a fairly expensive Adaptec RAID controller. Since most desktop PC work is opening small files, a single 40GB 7200 rpm BM Deskstar smoked that array in real-world use.

That's very strange and atypical results. The 10,000 rpm drives have lower latency so usually make for much better real-world performance since latency is more important on small files. I've been a big fan of 10,000 rpm raid0s for years. Those first generation 4.5 gb seagate cheetahs sure made a lot of noise and heat though.

wizard10000
June 2nd, 2011, 04:28 PM
Huh? How do you figure that? Latency isn't going to be any better, but it isn't any worse either.

Easy. I benchmarked it :D

A SCSI initiator can't connect to more than one device at a time, plus SCSI has significantly higher command overhead. On the other hand ATA controllers of that vintage couldn't do command queueing and the SCSI RAID controller could. If drives are identical except for the interface a SATA array can be and often is faster than a SCSI array.


That's very strange and atypical results. The 10,000 rpm drives have lower latency so usually make for much better real-world performance since latency is more important on small files. I've been a big fan of 10,000 rpm raid0s for years. Those first generation 4.5 gb seagate cheetahs sure made a lot of noise and heat though.

Nor really atypical if you consider sector density. These were 9g WD Enterprise drives that I believe may have had 3.5" platters but I'm not certain.

Let's compare a 10k rpm RAID0 set using those 9g drives with a 40gb 7200 rpm single drive. Just for fun we'll assume all the drives had the same number of platters (they don't - the WD drives had three each and the Deskstar had two). Your 18g stripe set has less than half the sector density of the 40g drive so even at 7200 rpm the ATA drive will sustain more than twice the throughput of your RAID array. The SCSI setup may or may not have lower track to track latency though.

A lot of n00bs miss this and I'm not calling anybody a n00b, but folks who worship at the altar of spindle speed are woefully misguided. Dis performance is a combination of sector density, spindle speed and track to track latency - and having a large cache helps too :)

psusi
June 2nd, 2011, 07:03 PM
Easy. I benchmarked it :D

A SCSI initiator can't connect to more than one device at a time, plus SCSI has significantly higher command overhead. On the other hand ATA controllers of that vintage couldn't do command queueing and the SCSI RAID controller could. If drives are identical except for the interface a SATA array can be and often is faster than a SCSI array.

You must have had a crappy controller then. Decent ones DO support the ability to disconnect from one drive and issue commands to others while it is busy. Most drives also support tagged command queuing so you can have several requests in the queues of several drives at once.

Also what do you base the statement about overhead on? ATA drives end up having to write something like 8 or 10 bytes to registers for a command, and SCSI typically uses 10 or 12 byte command blocks. Not much difference.


Nor really atypical if you consider sector density. These were 9g WD Enterprise drives that I believe may have had 3.5" platters but I'm not certain.

Let's compare a 10k rpm RAID0 set using those 9g drives with a 40gb 7200 rpm single drive. Just for fun we'll assume all the drives had the same number of platters (they don't - the WD drives had three each and the Deskstar had two). Your 18g stripe set has less than half the sector density of the 40g drive so even at 7200 rpm the ATA drive will sustain more than twice the throughput of your RAID array. The SCSI setup may or may not have lower track to track latency though.

Throughput yes, but we were talking about latency, which areal density does not help. Even though my raid0 of two old 36 gig 10,000 rpm raptors has lower sequential throughput than my new 1.5 tb 5400 rpm wd green drive, the system still feels faster on it because of the lower latency.


A lot of n00bs miss this and I'm not calling anybody a n00b, but folks who worship at the altar of spindle speed are woefully misguided. Dis performance is a combination of sector density, spindle speed and track to track latency - and having a large cache helps too :)

I have spent the last 15 years wondering why the hell anybody bothers or cares about cache on the drive. The OS has a lot more memory it uses to cache with than a drive ever will, and it doesn't do one bit of good for the drive to cache a block that the OS won't ask for again anyhow because it has it cached itself.

The disk cache is akin to having a huge L2 cpu cache, then a smaller L3 cache instead of the other way around. It's just silly.

3Miro
June 2nd, 2011, 07:06 PM
I have spent the last 15 years wondering why the hell anybody bothers or cares about cache on the drive. The OS has a lot more memory it uses to cache with than a drive ever will, and it doesn't do one bit of good for the drive to cache a block that the OS won't ask for again anyhow because it has it cached itself.


I don't think all OS have this feature.

psusi
June 2nd, 2011, 07:13 PM
I don't think all OS have this feature.

What one doesn't? Windows, Linux and Mac all do.. hell, even DOS 6.0 had an optional disk cache program.

wolfen69
June 2nd, 2011, 07:35 PM
I'll never buy intel. AMD is much more cost effective. Plus, you can't tell me you'll notice any difference between amd and intel. The only differences are going to show up in benchmarking tests, not the real world.

3Miro
June 2nd, 2011, 07:44 PM
What one doesn't? Windows, Linux and Mac all do.. hell, even DOS 6.0 had an optional disk cache program.

I must have missed something.

rJ~
June 2nd, 2011, 11:51 PM
I don't think you can comfortably draw many comparisons between the current CPU lines without knowing what you're going to use them for.

You can see a cheap AthlonII X4 beat an i3 when all four cores are used, or see the i3 beat a more expensive PhenomII X4 in tests where performance per core counts more.

For gaming, the Sandy Bridge based i3, i5 processors look like winners at their price points. Hopefully it'll be more interesting when AMD's Llano and Zambezi come out (Sometime between now and September). Anything below the i3s and you're looking at AthlonII CPUs.
Gaming probably isn't the most interesting benchmark on a Linux forum, but it's usually what I'm looking at when upgrading parts ;)

In relation to my first point, I bought my PhenomII X4 940 because it offered good enough performance and better performance/price in gaming compared to i7 at the time. I didn't pay much attention to benchmarks that were irrelevant to me.

wizard10000
June 3rd, 2011, 01:14 PM
You must have had a crappy controller then. Decent ones DO support the ability to disconnect from one drive and issue commands to others while it is busy. Most drives also support tagged command queuing so you can have several requests in the queues of several drives at once.

Could very well have been a crappy controller. Wish I could remember the model number, but this was > ten years ago and my budget wouldn't allow for much. I do remember that the additional 32mb cache for the controller was almost 200 bucks.


Also what do you base the statement about overhead on? ATA drives end up having to write something like 8 or 10 bytes to registers for a command, and SCSI typically uses 10 or 12 byte command blocks. Not much difference.

JMO but I think 20% command overhead can be significant, especially in an array. On a single drive I'd agree it's academic.


Throughput yes, but we were talking about latency, which areal density does not help. Even though my raid0 of two old 36 gig 10,000 rpm raptors has lower sequential throughput than my new 1.5 tb 5400 rpm wd green drive, the system still feels faster on it because of the lower latency.

I think areal density is relevant - at least if the drive's not too fragmented simply because the heads are gonna spend less time moving around. Maybe we should agree to disagree on this one ;)


I have spent the last 15 years wondering why the hell anybody bothers or cares about cache on the drive. The OS has a lot more memory it uses to cache with than a drive ever will, and it doesn't do one bit of good for the drive to cache a block that the OS won't ask for again anyhow because it has it cached itself.

The disk cache is akin to having a huge L2 cpu cache, then a smaller L3 cache instead of the other way around. It's just silly.

An interesting point. I tried looking it up but wonder if the DRAM on a disk's hardware cache is faster than main RAM on the PC? I think a disk with onboard cache disabled *feels* slower than one with it enabled but there's no way to quantify that. I tried this several years ago playing around with hdparm.

This might amuse you. Way back in the day in my quest for performance I put a Win 3.1 swapfile on a ramdisk to speed up the PC. I got better, though :D

PhillyPhil
June 3rd, 2011, 01:22 PM
The disk cache is akin to having a huge L2 cpu cache, then a smaller L3 cache instead of the other way around. It's just silly.

True.
The disk "cache" is really a disk buffer.

psusi
June 3rd, 2011, 03:14 PM
JMO but I think 20% command overhead can be significant, especially in an array. On a single drive I'd agree it's academic.

Not when you spend less than 1% of the total time even issuing commands. The commands could be 100% longer, when when you only spend 0.5% of the total time issuing commands, the total overhead is still only 1%; 0.5% higher than the commands half the size.

Compared to the time spent actually transferring the data, let alone waiting for the drive to actually get it off the platter, command time is nothing.


I think areal density is relevant - at least if the drive's not too fragmented simply because the heads are gonna spend less time moving around. Maybe we should agree to disagree on this one ;)

For sequential transfers then yes, higher areal density means you cross fewer tracks so spend less time seeking, but again, we are talking random access, where that areal density does not really help reduce the track-to-track seeks, and it certainly does not help the rotational latency.


An interesting point. I tried looking it up but wonder if the DRAM on a disk's hardware cache is faster than main RAM on the PC? I think a disk with onboard cache disabled *feels* slower than one with it enabled but there's no way to quantify that. I tried this several years ago playing around with hdparm.

You can quantify it with fs benchmarks like bonnie++. It doesn't really matter what the speed of the cache is though when the OS never reads from it. It is never going to read from it because to be in the cache it has to have recently been read, but if it has recently been read, then it will be in the OS cache, and since the OS cache is larger, its going to be in that cache long after the drive cache has dropped it.

I'm pretty sure that drive cache is just a vestigial carryover from days gone by when the OS didn't cache, and IDE drives could only handle one command at a time ( shoving writes into the cache and writing them to disk later freed up the disk to process read commands sooner ). These days it's just a waste of money.

slooksterpsv
June 3rd, 2011, 04:07 PM
Besides the OT comments about RAIDs, Drives, Disks, Caches, etc.... well Cache would be relevant to the discussion of AMDs, anyways the overall point is this:

AMD will have some better products over Intel. Especially soon with the Llano cores (I'm so excited!). So remember you don't have to be a fanboy of AMD or Intel or ARM or MIPS or Sparc, etc. Just choose what's going to be best for you - whether that's speed, cost effectiveness, or just down-right energy efficiency.

Either way I'm an AMD fan after a 1.1GHz Duron (low-end, and 2 years older) beats a 2.4GHZ Celeron (low-end, had more RAM too) at gaming.

If we don't get back on topic I'm afraid an Admin may close this thread due to the OT-ness hehe =P

OT Convo though:
I'd rather have a drive with more cache on it than none at all, but I put that kind of stuff to use.

PhillyPhil
June 4th, 2011, 12:52 AM
I'm pretty sure that drive cache is just a vestigial carryover from days gone by when the OS didn't cache, and IDE drives could only handle one command at a time ( shoving writes into the cache and writing them to disk later freed up the disk to process read commands sooner ). These days it's just a waste of money.

You are correct about a smaller cache below a bigger one being mostly a waste of money, but (to repeat myself) the disk "cache" is actually a misnaming of a buffer.

People consistently call it the disk cache, but it's there for buffering reasons, and it's mainly used as a buffer: queuing jobs and doing it's best to match speed on one side of the buffer to speed on the other side. (Yes, I'm aware that the disk buffer allows random access)

A disk "cache" would be mostly useless below large memory. A disk buffer is very useful.

psusi
June 4th, 2011, 04:41 AM
Buffer and cache are used interchangeably. This is because their function is very similar and you almost always see them go hand in hand. If you only wanted to buffer the in flight requests, 1mb would be more than sufficient for that, but they keep adding more to use as a cache as well.

PhillyPhil
June 4th, 2011, 04:49 AM
Buffer and cache are used interchangeably. What I said. A bad habit.
This is because their function is very similar and you almost always see them go hand in hand. Seriously? The functions are only similar in that they both temporarily hold data and make it available for retrieval. They have very different functions.
This is like saying a motorbike and a truck are very similar because they're both vehicles.
If you only wanted to buffer the in flight requests, 1mb would be more than sufficient for that, but they keep adding more to use as a cache as well. Rare to actually be used as cache (accessing previously accessed data again) because as you've correctly pointed out, memory is a much larger cache one level up.

wizard10000
June 4th, 2011, 11:01 AM
Not when you spend less than 1% of the total time even issuing commands. The commands could be 100% longer, when when you only spend 0.5% of the total time issuing commands, the total overhead is still only 1%; 0.5% higher than the commands half the size.

Compared to the time spent actually transferring the data, let alone waiting for the drive to actually get it off the platter, command time is nothing.

You know, you're right. Sometimes I still get caught up in the academic performance vs. real-world performance thing ;)


For sequential transfers then yes, higher areal density means you cross fewer tracks so spend less time seeking, but again, we are talking random access, where that areal density does not really help reduce the track-to-track seeks, and it certainly does not help the rotational latency.

But I am gonna be a little stubborn on this one :D

Assuming the drives aren't fragmented and have the same data on them a drive with twice the areal density has half as many tracks to seek from - actually with VBR it should have less than half as many tracks. With random access I don't think head movement would be reduced by half, but I still think it'd be reduced.


You can quantify it with fs benchmarks like bonnie++. It doesn't really matter what the speed of the cache is though when the OS never reads from it. It is never going to read from it because to be in the cache it has to have recently been read, but if it has recently been read, then it will be in the OS cache, and since the OS cache is larger, its going to be in that cache long after the drive cache has dropped it.

I'm pretty sure that drive cache is just a vestigial carryover from days gone by when the OS didn't cache, and IDE drives could only handle one command at a time ( shoving writes into the cache and writing them to disk later freed up the disk to process read commands sooner ). These days it's just a waste of money.

You've given me some more stuff to ponder here. I've been worshiping at the altar of large disk caches for years but the more I think about it the OS is going to pull from its own cache before requesting data from the disk and if the data doesn't reside in the OS' cache it's not going to be in the disk's onboard cache either.

Thanks for the education. I like it when I learn stuff, even if I do tend to be a little stubborn at times :D

cheers -

angryfirelord
June 5th, 2011, 08:30 PM
So I used to have a pretty good idea on comparable processors between AMD and intel, but since around the release of the core 2 duo I've stopped looking at AMDs that much and now I basically have no idea how their procs compare.

For example:

AMD Athlon XP = intel Pentium 4

so what about now? like:

core i3 = ?
core i5 = ?
core i7 = ?

and so on (I'm looking for rough comparisons, hence no particular submodels between procs)
Well, it's not exactly a straight-line comparison anymore. Even back in the Athlon XP days, there were things that the XP was better at and things that the P4 was good at, although the XP was much cheaper at time. It all depends on how the processor takes advantage of it.

Now, if we are dealing with one or two core applications, the Core i3 can hold its own quite well against a Phenom II X6. This is just simply because on a per core basis, AMD is two generations behind, so Intel has a much higher IPC. But when you start throwing in quad-core optimized tasks (such as video encoding or chess engines), then an Athlon II X4 would be faster than an i3. For the i5s and i7s, they really don't have any competitors. A Phenom II X6 might be able to take on an i5 provided you use the right application, but not an i7.

If you are shopping for a new build and don't need to spend $3000 on gaming hardware, then even the budget chips will provide plenty of horsepower. Celerons are all dual-core these days. I just ordered a new build myself with an Athlon II X2 chip. If you haven't been making new PCs since the XP days, I think you'll be surprised at how much cheap processors can do.

PhillyPhil
June 6th, 2011, 12:00 AM
A Phenom II X6 might be able to take on an i5 provided you use the right application, but not an i7.



For the third or fourth time this thread: this is not true. There are a large number of i7 models (see my list page 2 I think) and the top Phenom X6s *do* compete with the bottom third of i7s.

PhillyPhil
June 6th, 2011, 03:07 AM
Here is a graphical representation of performance (approximately to scale), using cpubenchmark.net scores.
As you can see the range of i5 and i7 models is huge (much larger than my list on page 2). i7s range from dinky little notebook cpus all the way up to $1000 6 core monsters.

The only AMD/Intel comparison that can be made without specifying a model is the i3 vs X6.
For all other comparisons, blanket statements are *wrong*. If you want to say the i7 is better than the X6 then specify models, or at least say ''the top i7 is more powerful than the top X6''.

http://img35.imageshack.us/img35/5976/screenshotmzn.png

Dry Lips
June 7th, 2011, 12:53 AM
I've got a question about core activation.

I've just unlocked the second core on my Amd Sempron 145, which transforms it into a
AMD Athlon(tm) II X2 4450e Processor! (Hey, hey!) \\:D/
However I don't see any improvement when running the weave filter in GIMP with two cores compared to one... Is this normal? :confused:
Otherwise I haven't experienced any problems...