What do you think of AMD processors?

God thanks AMD just for existing!
This is last wall stopping Intel from monopoly...

 

In case you guys didn't see enough of him yesterday at superbowl half-time, Will of BEP is also an Intel employee.
Core77.com • View topic - Lady Gaga, William and more
and click Play on the first video. It's somewhere between awkward and hilarious, but it's great!! One of these days they've gotta have him pull out a powerpoint to give a presentation!!

(Both AMD and Intel work just fine... they each have different purposes.)

 

at price i bought amd p520 w/ radeon 4250 256mb it was way better than comparable intel. the main advantage is not really the cpu but the integrated graphics. intel 4500mhd and even the newer intel HD cannot hold its own against the radeon 4250. since the graphics on both are the bottleneck it doesnt matter which cpu you have. but amd is cheaper.

 

AMDs mobile CPUs are terrible, and the gap is only getting wider. That is what I think.

 

you are incorrect sir.

you need to re-boot and upgrade to kevin jack 3.0

 

What about Sandy Bridge integrated graphics?
Also [and I seriously don't know] but does AMD CPUs have multiple threads?

 

AMD cpus have multiple threads. They do not have hyperthreading.

They also have crap for instruction sets in comparison to intel. If you want performance you do not go to AMD.

 

To be fair the performance is decent enough to justify the prices.

 

Hyperthreading has limited usefulness, and practically none to most people. For gaming and other real-time tasks it provides practically no benefit.

 

It is called poor mans Dual Core.
It will boost system responsiveness due to the extra threads present because nowadays all modern OS are multitasking/multithreading.
I won't say it is useless because it is a very good and cheap way of squeezing extra performance out of the silicon, while adding extra cores is more costly.

 

I have just revitalize an old D410 and agree with the above. It is that extra thread(in this case lack of as it is a Pentium M 1.86) that has noticeable lag comparing to an even slower clocked Duo Core.

However, I also agree with Trottel in the context that if there is already more than one core, hyperthreading's usefulness greately diminished. As when the situation makes all cores(so long it is more than one) busy, those hyper-core can help very little.

 

On a computer with an N270 and without noteworthy GPU acceleration (read: an older netbook) Hyperthreading makes the difference between "to be or not to be" able to watch 720p videos encoded in h264.

 

Whoever came up with that moniker is a moron. Many people believe that their i3 is practically a quad core, which could not be further from the truth. Mutlitasking and multithreaded apps do not see a boost in performance just because a single processor core has hyperthreading. The processor core must first be pegged at full utilization before hyperthreading can do diddly. Then, whatever task you are doing cannot be bottlenecked by a single thread. In some cases, hyperthreading can even decrease performance.

I didn't say it was useless. Hyperthreading is awesome if you are doing the right tasks. I used my i7 860 for folding@home. Crunching large files or lots of numbers is just no something most users spend much time doing, if at all.

But it's not the lack of an extra "thread" appearing in the task manager, it is the lack of a second core. A Core 2 Duo is more efficient clock for clock and has two cores, so of course it would blow the Pentium M away. But if you take an i3, i5, or i7 and disable hyperthreading, more often than not, the user is going to be none the wiser.

It did not even cross my mind that there was hyperthreading on some Atoms. That is probably the platform that would show the most benefit the most amount of time. A very weak single core like that should be able to make at least some use of hyperthreading for any application that is even semi dual-threaded and remotely intensive.

 

Actually all Atoms support Hyperthreading.

True!
I would even say that the Atom platform would be nearly useless for the mass market if it wouldn't have HT. Due to its design it needs two clock cycles to do a floating-point operation, which means that a 1.6GHz-Atom has a similar performance like an 800MHz-Celeron in many situations.
So it might be slow, but because of HT and most applications still being only able to use one thread the system stays responsive - which is what most people expect (and often confuse with speed).

 

Do you mean isn't? I would agree depending on the consumer, but my point was that if you have the money to spend and you're looking purely for performance intel is the best way to go.

 

AMD really took a hit on their budget when they bought ATI , and i think thats one of reason they fall behind on CPU development?

but now they managed to catch up with the race , and their processors are pretty decent even for high end built i would recommend them versus less budget friendly intel

 

I was not comparing with C2D but the generation after Pentium M, the Core Duo which is basically slightly improved Pentium M with 2 core. And it is the lack of 'thread'(or core) that makes the difference for responsiveness. Once there is one additional core/thread, the overall system experience becomes better even the raw CPU speed(per core) is slower. This can even be seen on Pentium 4 with hyperthreading.

Which is why I said I agree with you, once there is more than one 'thread/core' the gain for further core/thread diminished for typical desktop. But that transition from 1 -> 2 is very noticeable.

EDIT:
This is mentioned by another poster in the case of N270. The difference between single core vs single core(dual thread) is that the later is much better at context switching(invalidate the registers, the instruction set pipeline etc. in the case of single thread) that is key for latency(say in the case of playing 720p video without lag).

 

The only difference being that the minimal application Hyper Threading is used in in the year 2011 is dual and quad core processors.

 

It breaks down to this:

AMD = cheaper, value-conscious

Intel = over-all faster, can be much more expensive

That being said, most of my CPU's (until dual core CPU's came out) have been AMDs in the past 13 years or so. There's nothing wrong with them per say, they just tend to be slower than Intel's offerings, and have been for 5-6 years (since the Athlon 64 came out years ago) but they also sell for significantly less. You get what you pay for, etc.

 

the market is drifting from desktop though. I think sales of notebook has now outnumber desktop(i.e. annual new shipment) and the trend will continue.

 

Remember though that AMD let their CEO go very recently because of his "lack of direction" in the mobile and ultramobile market. With new leadership, we may see good progress in the company with mobile processors and chips - at least let us hope so, so that there will be an alternative to Intel and competition for them.

 

there is one, which is called ARM. Even the mobile space is changing and Intel is facing the old issue again of 'what if the slower guy is fast enough'. ARM cannot compete with even Atom in terms of computation power but its power profile beats it. For mobile device like smart phones, it is good enough and Moore's law cuts both way in that it will be getting faster as well.

It is interesting to see it unfold.

 

I don't buy that, especially since AMD's board consists of distinguished industry veterans from failed companies. I don't see what they expected him to do in those markets with all the things he's had to fix after the previous CEO. Was he supposed to borrow more money and put AMD back into the red or deeper, so he can put resources into competing with nVidia's Tegra line?

 

so, i have been wondering for some time...

if amd offers cheaper, and intel offers faster, then for the same price, amd should be faster, so wouldn't it mean that it would be better to go with amd for the majority of performance levels, meaning until up to amd's fastest cpu, and then go for intel only for cpus that are even faster than amd's fastest?

for example... i'm assuming that mobile i5 and i7 are faster than any of amd's current mobile offerings, but how about i3?

why get an i3 when you can get an amd cpu with equal performance for less?

from my limited intuitive "experience", amd doesn't seem to be as good at multitasking, and performance is not as consistent (sometimes fast, sometimes slow)

 

AMD isn't really cheaper, example the AMD P340 2.2GHz vs. i3 370m 2.4GHz.

Passmark:

i3 370m: 2225

AMD P340: 1300.

Now used, the P340 cost more than the much faster 370m

Intel Core i3 Mobile Processor i3-370M 2.4GHz 3MB CPU, - eBay (item 260725372361 end time Feb-19-11 16:21:56 PST)

AMP340SGR22GM NEW AMD MOBILE ATHLON II SERIES P340 2.2G - eBay (item 290530130953 end time Mar-04-11 08:44:12 PST)

 

Performance isn't the only factor (well, it is, but you have to look at "total" performance, and not just processing ability). For the same processing level, AMD tended to be hotter than Intel (in the mobile sector). This was compounded by the fact that the only major manufacturer that would offer AMD at the time was HP/Compaq in their consumer models, and HP has never been very stellar on cooling in their consumer notebooks. This meant that the AMD notebooks that were offered would run very hot, and thus performance would be degraded. This is often no longer the case, but people have long memories, and as they say, "once bitten, twice shy."

 

AMD laptops overall are cheaper for similar performance. You really don't get into i3 territory until 500+ and many manufacturers its 600+. Not everyone needs it, for alot of people an AMD dual core is enough. I think the higher pricing for AMD is just due to the minuscule OEM market compared to Intel's...I wouldn't base whether a brand makes worse quality stuff based off upgrade pricing..

 

Passmark is a joke and any numbers it spits out you might as well pull from your anus. The numbers it comes up with don't translate into anything in real life. And ebay is not exactly the benchmark for microprocessor prices.

 

Before I begin, I want to point out that the passmark tests are nice because they show the multi-tasking capability of a processor and therefore put out accurate overall system performance ratings.

Now to render the results of my analysis and thorough research of this topic.. AMD holds the best value for performance for their processors, but they run hot. I would not buy an AMD-powered laptop due to the high heat output and power consumption, but AMD is an excellent choice for most desktop users. However, thanks to the recently-released Sandy Bridge processors, you can buy a near-top-of-the-line processor for the incredible prices of $333. This processor boasts crazy-good benchmark ratings right out of the box, but it's overclocking capabilities make it outright insane. Here are some links to back up my point:

Incredible performance (and price).
PassMark CPU Lookup
PassMark - Intel Core i7-2600K @ 3.40GHz - Price performance comparison

Current price.
Amazon.com: Intel Core i7-2600K Processor with 8 MB Cache, 3.40 GHz for Socket LGA1155 - Unlocked Boxed: Electronics

Read here for user reviews. Take note of people's overclocking and temperature reports.
Newegg.com - Intel Core i7-2600K Sandy Bridge 3.4GHz (3.8GHz Turbo Boost) 4 x 256KB L2 Cache 8MB L3 Cache LGA 1155 95W Quad-Core Desktop Processor BX80623I72600K

Unfortunately, the chipset problems make the i7-2600k a poor choice for right now, but intel will surely get this solved ASAP in the interest of not completely failing as a company.

Now I am going to make my case for AMD as a great desktop processor at an even better price. I will list the processors primarily being compared and a couple of notes about them first.

AMD Phenom II X4 965 BE (quad-core). This processor essentially gets beat in all of the comparison links I will provide, but keep in mind it's extremely low price of $160 when considering how it rates against the Intel processors.
Newegg.com - AMD Phenom II X4 965 Black Edition Deneb 3.4GHz 4 x 512KB L2 Cache 6MB L3 Cache Socket AM3 125W Quad-Core Processor HDZ965FBGMBOX

AMD Phenom II X6 1090T BE (six-core). This is the processor that I would recommend for essentially ANYONE who wants a really nice desktop but doesn't feel the need for top-of-the-line. Keep in mind it's beautiful price of $200 when viewing the performance comparisons.
Newegg link:
Newegg.com - AMD Phenom II X6 1090T Black Edition Thuban 3.2GHz 6 x 512KB L2 Cache 6MB L3 Cache Socket AM3 125W Six-Core Desktop Processor HDT90ZFBGRBOX
Passmark link for price/performance ratio and performance comparison:
PassMark - AMD Phenom II X6 1090T - Price performance comparison

Intel i5-760 (quad-core). At $210, this is Intel's closest price match to the 1090T in terms of price. The comparison I want you to remember this for is the one using the i5-750. Of course the 760 will have slightly better performance than the 750, but not enough to make much of a difference in these performance analyses.
Newegg.com - Intel Core i5-760 Lynnfield 2.8GHz 8MB L3 Cache LGA 1156 95W Quad-Core Desktop Processor BX80605I5760

Intel i7-920 (quad-core). This is a $320 processor that gets compared to the $200 1090T in the first comparison article I will list. Note how close they are in performance despite the drastic difference in price.
Amazon.com: Intel Core i7 Processor i7-920 2.66GHz 8 MB LGA1366 CPU BX80601920: Electronics

This first article compares performance results of the Phenom II X6 1090T, i7-920, Phenom II X4 965, and overclocked versions of the first two in addition to the out-of-the-box results. The 1090T actually is (surprisingly) neck-and-neck with the i7-920 in most tests when they are overclocked.
http://www.techreaction.net/2010/05/06/amd-phenom-ii-six-core-1090t1095t-thuban-review/

This other comparison is a thorough one in which you should take note of the Phenom II X4 965 (cuz of how it rates in spite of it's killer $160 price tag!), the Phenom II X6 1090T, and the i5-750. The reason I ignored the i7 processors in any tests they took the lead on was because they are far more expensive than the AMD processors, whereas the i5-750 is essentially the exact same price as the Phenom II X6 1090T. I encourage you to look through the article yourself, but I will give you the basic results for how they differ significantly in performance.
-i5-750 has a nice lead on the 1090T in memory bandwidth
-i5-750 mild lead in SPECviewperf v10
-1070T significant lead in Cinebench R11.5
-1070T whoops butt in the Povray v3.6 test
-i5-750 performs far better in WinRAR compression
-1070T performs far beter in Handbrake fps test
-i5-750 has a significant edge in AVI to MPEG conversion
-i5-750 is much better in Unreal Tournament 3
-i5-750 is mildly better in Company of Heroes Opposing Fronts
-i5-750 is much better in Far Cry 2
-1090T gains a huge edge in non-gaming applications when overclocked
http://www.legionhardware.com/articles_pages/amd_phenom_ii_x6_1090t_be_1055t,1.html

Remember also that the six-core processor is going to be able to handle multiple processes and programs better for the simple reason that it has 12 threads instead of 8, so general performance of the system is going to be more efficient.
So that was my exhaustive breakdown of AMD vs Intel lol; sorry it was so dang long

 

They are fast but do not have long life as Intel has. AMD cannot offer durability. I thank good quality thing is always good . But it is good for those who can not afford .

 

AMDs currently dont have hypertheading so the 6 core is only 6-threads, not 12. So in terms of efficiency the intels are more efficient because of hyperthreading, (and power consumption)

 

I'm sorry; you are right on that one, JKleiss. I thought I read somewhere that AMD has a similar technology, but I was wrong. They do have better "turbo-boost" ("true core" they call it) technology that allows half (3) of the processor cores to gain a significant boost instead of 0.5 cores like Intel's turbo-boost.

To jounkarry:
I would not know about durability; most people who use AMD (in desktops) say they love them, and brand loyalty is not maintained by unreliable products. However, I can't say you're wrong, either, because my hypothesis here is not strongly based on scientific data. Of course, I'm sure it helps to use nice cpu coolers like the Corsair Hydro series coolers. I would not buy an AMD desktop without using one, because I am uncomfortable with running my CPUs so hot, even if the producer boasts a high maximum temperature (happens to be 62 degrees Celsius for the 1090T, and that doesn't seem very high to me).

 

something something XBOX 360 something something

 

People get way too excited about hyperthreading. It is something akin to more L2 cache or a higher FSB, but people just make it sound like it is the greatest thing ever; two cores for the price of one. Hyperthreading is only useful for a very specific set of tasks, and the more cores we have, the less useful hyperthreading becomes. And people act like more "threads" showing up in task manager actually means something, but it doesn't. If I had a choice between a quad core sandy bridge at 3Ghz, and a single core sandy bridge at 12Ghz, I would chose the single core each and every time. Mutli-threading is not inherently good, but a necessary evil.

They last a long time. Perhaps even decades past their useful life. And no, we really don't know which brand or what specific processors are more durable, but it doesn't matter anyway.

 

Lol xbox360 is pretty horrible, indeed. I only have one because of Halo.

 

I can't find anything on this "true core" that you talk about, at least in comparison to turbo-boost. The closest I can find has to do with a comparison on the use of "real" cores as opposed to logical ones ( here), and that has nothing at all to do with turbo-boost. That comparison is all about how a "real" core has more performance than a logical core, which is obvious because a "real" core doesn't have to share it's resources with anything else, while a logical core has to share the performance of a single "real" core with another logical core. Numbers have shown that a hyperthreaded core (AKA 2 logical cores) has a performance boost of about 10% over a non-hyperthreaded core when both are being pushed to full capacity (so a logical core on that hyperthreaded core would be operating at around 55% of the non-hyperthreaded core). This isn't a bad investment, considering you're getting about 10% more performance for only 5% more die/transistor area (as opposed to 100% more performance for 100% more die area to add another core). So basically, that link just says that, "oh, 2 of our 8-core processors is better than 2 hyperthreaded 4-core processors when you load 16 threads". My answer to that is something along the lines of "well, duh!"

Turbo-boost is really the opposite of hyper-threading; turbo-boost is about overclocking the processor when there's enough thermal room to do so, especially when you can shut down some of the cores. Basically, if you're only going to use one core, you can shut down the other core(s) and use the cooling that would otherwise have been devoted to those cores to cool that single core, allowing that single core to overclock beyond the normally listed clockspeed. This turns the listed clockspeed of all Intel CPUs into a "minimum" guaranteed clockspeed; they'll actually usually automatically run at a higher clockspeed if they can (within the limits of turbo-boost; the turbo-boost on Sandy Bridge is dramatically better in that respect).

 

Hyperthreading does not matter if the physical cores are fully (100%) loaded.

Hyperthreading works only when the CPU is not at full load. For instance an intel CPU loaded at 70% will use the remaining 30% to emulate additional logical cores.

In heavy duty applications such as rendering or video encoding Hyperthreading has a little to no benefit. The CPU is fully loaded and often no room is available to allow the creation/ emulation of extra cores.

Same here
No matter how you think of it, a CPU with 4 cores at 3GHz will lose few cycles to divide the workload between the 4 cores and coordinate between them. 1 core at 12 GHz is surely faster as it doesn't have to deal with this overhead

 

Hyperthreading still matters a little then, because (one of) the executing threads can still stall (due to a cache miss, branch misprediction, or data dependency); this is where I got the 10% number from (from rendering benchmarks which should be fully loading all the cores; if you really want, I can try to go back and find all the links we had when we were hashing out how hyperthreading worked a year or so ago). Basically, the numbers were showing that when you disabled hyperthreading, that same CPU suffered about a 10% performance drop in rendering benchmarks.

 

I see AMD has released the X940 for the mobile space...still 45W TDP but 100mhz faster than the X920 @ 2.3Ghz

Almost tempting...2.4Ghz quad...but I'm still pretty happy with the 2.8Ghz dual @ 35W in my laptop now.

 

can someone please summarize everything that was just said above so a 4 year old (me) can understand it?

so at the same price point, does AMD or intel have better performance for mobile processors?

(i looked at the passmark price/performance chart, it basically shows that intel is better for the majority, and amd for its top 2 processors, but aren't these for desktop cpus?)

 

Actually the opposite is true. Rendering, encoding, and other purely number crunching activities make the best use of hyperthreading. The thing is that when a processor is at full load, it's execution units are not all utilized all the time. For example if something on one thread isn't in cache, the further out the cache that has that information is, the more cycles are going to go by without any work to do as the data is transferred. During this time having a second thread ready to jump in means that those clock cycles don't go to waste. Also if one thread is using some of the execution resources but not others, the other thread can use those unused execution resources even on the same clock cycle. Both threads of a hyperthreaded core are equal. There is no real core and no logical core. They are identical and share execution resources equally.

Also no matter what program you are running or how multithreaded it is or isn't, it will always be able to take advantage of most or all of the processor's capabilities.

 

What about the critical resources like register and data/instruction cache ? With 4 cores, I have 4x the registers whereas in 1 core, those reigisters keeps on getting flushed due to context switching ? I don't know much about the data/instruction cache but I would assume each core would have their own(the small size L0 ones).

One of the reason early day x86 sucks comparing with RISC was the lack of registers if I am not mistaken.

 

I think he confused True Core with Turbo Core which is, afaik, only available on the desktop hexacores, but it seems like a makeshift Turbo Boost. Llano is rumored to have a 2.0 version of Turbo Core, but we'll have to wait for a review to find out how it'll work.

 

Oh, interesting. So Turbo Core works by overclocking up to 3 out of 6 cores by up to 500 MHz when the other 3 (or more) cores are idle. That's at least theoretically better than the turboboost on Gulftown (1/1/1/1/2/2), but Gulftown also uses the older version of Turbo-Boost. Given the considerably enhanced Turbo-Boost 2.0 that's come out with Sandy Bridge, it'll be interesting to see how Turbo Core 2.0 stacks up.

 

AMD is a good alternative to Intel, since one company should not control all the market like a monopoly.

 

It's actually completely opposite. A single core running multiple threads has much more of a penalty for context switching than multiple cores. It has to completely flush the cache to switch between threads at the penalty of hundreds of cycles many times.

I would probably pick a 12GHz as well, but if I was doing significant work with encoding video or other efficiently parallel loads the 3GHz will be better.

 

Man, you guys must have seen some really awesome analogical drawings from those 'how things work' books to be able to understand that stuff about hyper-threading so well.

On AMD CPUs however, I like how the Athlons OC so well because the CAS timings of the RAM don't tighten when downclocked with the FSB At least on my chipset.

 

QFT. This was why Pentium 4 hit the speed brickwall and was replaced with Core 2 Architecture.

 

The penalty for switching threads is not hundreds of clock cycles, is it? That is the worst case scenario kind of latency that would occur if the CPU needed some data from the main memory, not switching out the data it needs to execute. The latency of L1 cache is just a few cycles, and the latency for L2 cache is about a dozen. It's also not flushing the cache, which would be horrendous, but the architecture state. Anyway, that is what hyperthreading is there for, since it helps do work without any latency.

I found some good data on Nehalem and Core 2 cache and memory latencies here: The Dark Knight: Intel's Core i7 - AnandTech :: Your Source for Hardware Analysis and News

So even if you switch between two threads a million times per second on a 3Ghz Nehalem processor with hyperthreading disabled, you are only losing 4 million clock cycles, or .133% of the total, which is insignificant. Even if it had to pull the data out of its slowest cache, it would still only be about a 1% loss in this hypothetical situation. Open up your task manager and see how many threads there are. For instance, mine says 797 right now. I don't care how many cores and how much hyperthreading you have, if there was a serious delay for switching between threads, any CPU intensive tasks I do would be noticeably affected by the number of other random things I have my computer doing. This is also why there is no penalty for running some program with say 8 perfectly even threads on a processor with only 2 cores. If all those threads were jockeying with each other, there would be a serious performance hit and we would see a more than linear increase in the amount of work that got done with more cores.

Not exactly. It went Pentium D (2 cores with hyperthreading) to Core 2 Duo on the desktop, and Pentium 4 with hyperthreading to Pentium M on the laptop. Adding more cores/threads did not play any role.

 

Firstly Hyperthreading isn't a software implementation, it is a hardware + software implementation.
A HyperThreading capable processor has 2 Cache but 1 execution unit compare to a Non-Hyperthreading processor which has 1 Cache and 1 Execution unit.
The 2 Cache acts as 2 Virtual Processors but shares 1 Execution Unit.
Which means the Execution Unit switches between the 2 caches to produce a Pseudo 2 Core Processor

The reason why Pentium 4 Northwood, Prescott was dropped because it had a SuperScalar 24/32 stage Pipeline but the disadvantage of that was if the processor Branch Prediction Unit predicted wrongly the pipeline has to be flushing inc-curing huge performance penalty. The superscalar design no longer yield much improvement per Clock Increase. So they took back the Pentium 3 Architecture 14 stage pipe-lining approach at the same time their Israel Team designed the the Pentium M which was a heavily modified Pentium 3. And the rest is history.