New Am2 Preview on Anandtech's Page

[DrMrLordX]

Originally posted by: Absolute0
I'm kinda pisses off cuz for many months now, all logic has pointed to AM2 being about 5% faster than s939, and these results bolster that argument. However, for months, i've heard dozens of people insist that AM2's performance will ramp up "significantly" and that we'd all be surprised. I don't think that is happening, and at this rate unless AMD experienced a miracle break through, all those people have been insane.

I'd only be pissed if I planned on getting an AM2 machine. Remember that buying one is optional. We can always buy Conroe or Merom-based machines instead.

AMD's front will likely be silent until 2007, or whenever it is they make the move to 65nm. It is also rumored that they'll release AM3 in 2007. AM2 may have a very short product life.

The only interesting AM2 product I see is the 35W X2-3800+. Anyone else interested in this thing?

 

[Markfw]

I'm waiting for cheaper dual-core 940 or quad-core 940, The quad-cores will be on 65 nm, and I think they will make them for socket 940, as server people wouldn;t want to spend thousands, just to upgrade their cpu's (like my $850 mobo and ram, and $400 SCSI controller)

I may do a conroe to add to my fleet, just to see what they are like, but I have no reason to upgrade other than that, X2's do fine at the speed they are now.

 

[Absolute0]

Originally posted by: rise4310

Originally posted by: Absolute0
I'm kinda pisses off cuz for many months now, all logic has pointed to AM2 being about 5% faster than s939, and these results bolster that argument. However, for months, i've heard dozens of people insist that AM2's performance will ramp up "significantly" and that we'd all be surprised. I don't think that is happening, and at this rate unless AMD experienced a miracle break through, all those people have been insane.

where'd you see that info

i don't recall anyone here saying that but i may have missed it.

Other forums, people saying how AM2 will surprise us so much and talking down to us for being ignorant, when it seems that these people (a handful) have been wrong the whole time...

 

[Brunnis]

It's a little interesting that the AM2 platform couldn't take 3-3-3 timings... Was that a BIOS limitation, or would it simply not work correctly? If the latter is true then that, along with the low bandwidth, could indicate some problem still remaining with the memory controller. Time's running out...

3-3-3 timings with an actual 10GB/s of usable bandwidth and AM2 performance would probably look slightly different...

 

[Absolute0]

i like how for the past week, everybody was saying how AM2 performance would improve a ton, and didn't expect Conroe's numbers to go up, and AM2 is due out sooner.

 

[RussianSensation]

Originally posted by: Absolute0
I'm kinda pisses off cuz for many months now, all logic has pointed to AM2 being about 5% faster than s939, and these results bolster that argument. However, for months, i've heard dozens of people insist that AM2's performance will ramp up "significantly" and that we'd all be surprised. I don't think that is happening, and at this rate unless AMD experienced a miracle break through, all those people have been insane.

No, I think what they meant by "ramping up" is usually a reference to increases in clock speed (ie. S478 to 775 Northwood --> prescott, or as AMD transitions to 65nm on socket AM2). Ramping up and being surprised about performance simply means that you'll be suprised how high AMD will be able to take 2-year-old A64 architecture that started at 1.8ghz (to say 3.6ghz). Plus the reasons why they switched to socket 940 was to eliminate people from putting DDR1 A64s into AM2 DDR2 sockets. Other than that, I don't recall any credible source making any mention that A64 will be changed internally as a processor to make it more efficient.

Plus I don't see what the big deal is about AMD losing the performance crown? So what? The only thing that matters to consumers is performance for the $. Mobile XPs had it, 2.4C had it, A64 had it, now Intel's turn. Until AMD brings $200 dual core processors, they aren't even in the running to advance multi-core processing. And if Intel can bring us $200 dual cores, then I am all for it.

 

[Starbuck1975]

I am a little bit confused about the AMD path for the next one to two years.

I have been debating performing an upgrade to socket 939 dual core, but the technology seems to be moving faster then my budget can afford.

My last system build is roughly 3 years old, anchored on a Socket A Athlon XP 1800+ and a Radeon 9700Pro.

I typically do not upgrade incrementally...2 to 3 years is usually my upgrade entry point, which typically requires a new video card, RAM, processor, MOBO & power supply...very rarely am I able to salvage parts from one build to another, save the sound card, hard drive, DVD burner and DVD-ROM drives.

If AM2 is not going to provide a leaps and bound performance jump with DD2, should I just enter the market now on an Athlon 64X2 system...by the time my next upgrade point arrives, AMD will probably go through two if not three more architecture revisions.

 

[GimpyFuzznut]

AMD is basically releasing the same chip they have had out for months. It is no surprise that there is no real performance gain. We will probably have to wait until the next generation of processors come out from AMD to see any actual improvements. I only hope that they don't plan on moving towards a newer platform when that generation comes - otherwise AM2 will be a total bloody waste. (ie if you buy a motherboard for AM2 now, will it be outdated for the next gen. of chips that actually do more than what has been done fo the last year). But like as stated, quad-core etc is planned for 940 so that's good news.

 

[Madwand1]

I can't figure out which processor they used. D'oh! Help me out if you can.

Why does it matter? Because part of the reason for going to higher bandwidth memory is because it becomes a greater bottleneck as CPU speeds ramp. So this is material, and the differences will be highlighted in the higher end processors, not the lower end ones.

Comparing some of the benchmarks with previously stated results, I would guess 4800+, for others, 3800+. In either case, I think this is a bit of an oversight -- still faster processors ought to be used in the end for comparison for the above reason. Of course, it is useful for the masses and I who don't buy the most expensive processors to know about the performance of lower end ones.

 

[Ika]

hopefully AMD is pulling a Nintendo, except that they aren't giving a hint that "theres something else we haven't revealed yet"....

 

[phaxmohdem]

I don't know about your guyz's game plans but here is how I'm gunna roll... I've been on this FX-51 for a freckle past 2 years now and as much as I'd love to dive head first into Dual core now, I'm going to bite the bullet and wait for K8L or K10, or whatever AMD's next gen platform is before I buy. That way, if AMD's stuff is good enough I can go with that, and if its underwhelming, I should be able to get a Conroe setup hella cheap. Either way, I'm glad that there is some action/excitement in the CPU arena again.

 

[d3lt4]

I'm not surprised, as I wasn't expecting any real performance gain. I think Intel will be performance crown again for about a year, but who cares. THey have been for the last 15 years except for the k8's era. Next year AMD will most likely be back on par for performance with k8l, and hopefully this year they will have some really cheap chips.

That 35w x2 3800 looks sweet! can't wait to see how it overclocks and performs. unfortunately that's about the only interesting thing from AM2 that I see.

 

[Extelleron]

Originally posted by: chilled
They need to test it with differnt DDR2 speeds. I can't get DDR2 800 for the price of DDR 400 yet. I'm willing to bet that the performance at slower memory speeds could be a few percent less than what we've seen today.

Meh, DDR2-533 or DDR2-667 MIGHT be actually faster, or at least the same, as DDR2-800 in many cases on A64 processors. Since the A64 processors arent bandwidth starved and arent going to benefit much from added bandwidth, DDR2-533/667 at lower latencies than DDR2-800 might actually be better for it.

 

[Hard Ball]

OK, let me see here:

AM2 infrastructure:
--shows little improvement over previous socket
--requires new and costly memory upgrade
--due to be replaced by DDR3 compliant version in as little as a year
--whose chips at this point (I mean later this quarter) pales in comparison to the competition.
--and is the last ditch effort by AMD before the new K8L core comes out.

Question:
Why would anyone contemplating buying something like this?? This at best, would be another S754, and probably wouldn't even have that kind of longevity.

 

[Absolute0]

More like a Socket 754 that can barely outperform socket 462 LOL

 

[AnandThenMan]

Originally posted by: Hard Ball

Why would anyone contemplating buying something like this?? This at best, would be another S754, and probably wouldn't even have that kind of longevity.

Well, if you don't think 3 years for a platform is not long enough, then give Intel a try. With Intel, you practically have to buy 3-4 motherboards a year just to be able to keep updating to their latest. And that is not overstating things.

edit: stupid mistakes

 

[weskurtz]

From what I have been reading, the real issue here is not the AM2 performance, but more about the lack of performance from Conroe. From what I have read the only real cometitive performance will come from cache hungry apps. IE. Conro will win in a benchmark of Super PI. But in the real world, the two architectures will be closer in performance than what most people think. Maybe I am wrong but I think that K8 and Conroe will be in more of a tie than either one being a clear winner.

 

[designit]

Originally posted by: Absolute0
More like a Socket 754 that can barely outperform socket 462 LOL

But can't be worst than Conroe's cache thrashing.
Read:

When you share, you also fight, just like two kids sharing a single toy will fight for it. Thrashing is well known problem in paged memory multi-tasking systems, when it happens, too many tasks compets for memory and the system spends most of its time in paging activity between the memory and the disk. From a user's view, the system basically comes to a halt. With shared cache multi-core designs like INTEL Yonah, multiple cores fight for cache. Cache thrashing will become a common phenomenon inside the INTEL CPUs.

help me.......master..... I am freez.......ing...burrrrrr. system halt... system halt....syssssssssssssssssssssssssss....... a moment of silence please.

 

[DrMrLordX]

Originally posted by: Hard Ball

Why would anyone contemplating buying something like this?? This at best, would be another S754, and probably wouldn't even have that kind of longevity.

The only reason I can see is if the x2-3800+ 35W CPU proves to be an excellent overclocker. Even then, Conroe . . .

 

[Accord99]

Originally posted by: designit
help me.......master..... I am freez.......ing...burrrrrr. system halt... system halt....syssssssssssssssssssssssssss....... a moment of silence please.

As opposed to AMD which has to talk to each other to maintain cache coherency, except over the much slower and lower bandwidth SRQ. And unfortunately since the caches are separate, it can't be allocated smartly to improve performance in single-threaded apps or share data to improve efficiency in multi-threaded apps.

 

[dmens]

Originally posted by: designit
When you share, you also fight, just like two kids sharing a single toy will fight for it. Thrashing is well known problem in paged memory multi-tasking systems, when it happens, too many tasks compets for memory and the system spends most of its time in paging activity between the memory and the disk. From a user's view, the system basically comes to a halt. With shared cache multi-core designs like INTEL Yonah, multiple cores fight for cache. Cache thrashing will become a common phenomenon inside the INTEL CPUs.

help me.......master..... I am freez.......ing...burrrrrr. system halt... system halt....syssssssssssssssssssssssssss....... a moment of silence please.

LOL, that sharikou (PhD!!!) guy is totally clueless. He claims merom/yonah will suffer from L2 cache thrashing simply because the L2 cache is shared and accessible by either core. But since the L2 is physically tagged... so much for that!

He uses a hyperthreading example to justify his claim even though neither yonah nor merom has SMT, and I doubt he has any clue how exactly HT is negatively impacted by the particular example given in the MSDN blog post. With two processes on one core and the helper thread taking control of large chunks of the caches, the worker thread is forced to refetch continuously. But with just a shared L2 and no SMT, it is *no dfferent* than a physically partitioned L2.

Even his analysis of thrashing is wrong, he makes it sound like the CPU livelocks, even though with the given worker/helper scenario, it is not possible.

 

[dexvx]

Originally posted by: dmens
He uses a hyperthreading example to justify his claim even though neither yonah nor merom has SMT, and I doubt he has any clue how exactly HT is negatively impacted by the particular example given in the MSDN blog post. With two processes on one core and the helper thread taking control of large chunks of the caches, the worker thread is forced to refetch continuously. But with just a shared L2 and no SMT, it is *no dfferent* than a physically partitioned L2.

Its a known fact that cache thrashing was a huge problem for Netburst CPU's that had HT. This was mainly due to the small L1 prefetch, or in Northwood's case, a small L2 and a first generation implementation of HT. Overall, the shared cache solution's downfalls of cache thrashing is greatly overcompensated by a non-bridged L2 solution's higher latency, duplicity of data, and half effectiveness on single-threaded apps.

 

[designit]

Originally posted by: Accord99

Originally posted by: designit
help me.......master..... I am freez.......ing...burrrrrr. system halt... system halt....syssssssssssssssssssssssssss....... a moment of silence please.

As opposed to AMD which has to talk to each other to maintain cache coherency, except over the much slower and lower bandwidth SRQ. And unfortunately since the caches are separate, it can't be allocated smartly to improve performance in single-threaded apps or share data to improve efficiency in multi-threaded apps.

peeeeehhh.
I gues you dont know much about AMD key Architecture. How much do you know about SRI and Crossbar Switch?
enlighten yourself.
http://www.amd.com/us-en/Processors/Pro...ion/0,,30_118_9485_13041^13043,00.html

 

[Zebo]

Originally posted by: n7
I'm actually impressed that overall, performance is slightly better, or at least not worse.

I'm also glad AT had to test with 1 GB dimms.

Testing with 512 MB dimms is kinda silly now, since the large majority of people get 2x 1024 MB these days.

AMD's A64 architechture is not bandwidth starved, so these results don't surprise me.

I'd assume DDR2-533 @ 3-3-3 or 667 @ tighter timings will be around the same as DDR2-800 @ 4-4-4, etc.

No it's not better. Socket 939 rev E supports DDR 500 at max spec. They tested 939 at mid spec, DDR400 while testing AM2 at max spec DDR2 800 and it barley even won..but 0-2% in RL. If you threw a max spec 2-2-2 DDR 500 setup in there, good night AM2.


If Conroes numbers hold up to legit testing I see $40 chips on the AMD horizon again like back in t-bred days.

 

[Zebo]

Originally posted by: d3lt4
I'm not surprised, as I wasn't expecting any real performance gain. I think Intel will be performance crown again for about a year, but who cares. THey have been for the last 15 years except for the k8's era. .

You're totally clueless AMD has held performance crown since very first athlon except Northwood C era which lead for 8 months last 6 years and only because A64 was late.