BF2S Forums What's the deal between AM2 and Socket 939?

929 = ddr AM2 + ddr2

haffeysucks wrote:

swedge wrote:

929 = ddr AM2 + ddr2

Oh.  Is that it?  939 can support DDR, while AM2 can support DDR2?

yes am2 is newer than 939, so they are moving from 939 to am2

Last edited by Leprechaun56 (2006-12-06 12:51:04)

AM2 is new, so supposedly they can do a lot with it before maxing out it's potential.

Wait, one last question.  Can DDR RAM fit in DDR2 slots?

you're dealing with different pin counts at that point.

Thanks to everyone that helped, karma will be dished out accordingly.  If I haven't karma'ed you yet, send me a PM so I can get you when my karma comes back.

Bertster7 wrote:

AM2 is rubbish. Don't know what AMD were thinking with that one. It's a revamped version of the old socket 940 for FX chips, remember that one, from the days of socket 754. The need for registered memory is gone and support for DDR2 memory has been added. The question is why get AM2? The setup is no faster, yet is more expensive (you have to buy DDR2 RAM - more expensive, you need a newer MB - more expensive, you need an AM2 CPU - more expensive). Performance increases on comparable 939 and AM2 CPUs are negligible, in some cases they even run slower.

There are some newer CPUs that are not available for 939. But you can get upto an FX60 or an X2 4800+ on 939. If you're looking for anything in that sort of price range you should really be after a C2D anyway.

939 is a solid choice if you are building a system on a tight budget. AM2 is a useless rehashing of old technology that is not even slightly competetive with it's Intel rivals. I didn't know about the AM3 improvements to the memory controller, but they are certainly needed. AM2 does not perform anything like as well as it should considering the higher frequency memory.

The current A64 architecture is not bandwidth limited, so there was not really an expectation that it would be faster.  The move to DDR2 was because that is where the market was going, and overall is a better design, allowing for larger capacity modules than DDR.  Also, DDR production was going to be reducing as manufacturers continued to ramp up DDR2 production, so DDR was going to become less common and more expensive.  Lastly, while the A64 isn't really bandwidth limited now, the move to DDR2 will help when you start reaching quad core level CPUs.

So before you start bashing AMD for moving to DDR2, maybe you should have a better understanding of why they did it.

Bertster7 wrote:

Agent_Dung_Bomb wrote:

Bertster7 wrote:

AM2 is rubbish. Don't know what AMD were thinking with that one. It's a revamped version of the old socket 940 for FX chips, remember that one, from the days of socket 754. The need for registered memory is gone and support for DDR2 memory has been added. The question is why get AM2? The setup is no faster, yet is more expensive (you have to buy DDR2 RAM - more expensive, you need a newer MB - more expensive, you need an AM2 CPU - more expensive). Performance increases on comparable 939 and AM2 CPUs are negligible, in some cases they even run slower.

There are some newer CPUs that are not available for 939. But you can get upto an FX60 or an X2 4800+ on 939. If you're looking for anything in that sort of price range you should really be after a C2D anyway.

939 is a solid choice if you are building a system on a tight budget. AM2 is a useless rehashing of old technology that is not even slightly competetive with it's Intel rivals. I didn't know about the AM3 improvements to the memory controller, but they are certainly needed. AM2 does not perform anything like as well as it should considering the higher frequency memory.

The current A64 architecture is not bandwidth limited, so there was not really an expectation that it would be faster.  The move to DDR2 was because that is where the market was going, and overall is a better design, allowing for larger capacity modules than DDR.  Also, DDR production was going to be reducing as manufacturers continued to ramp up DDR2 production, so DDR was going to become less common and more expensive.  Lastly, while the A64 isn't really bandwidth limited now, the move to DDR2 will help when you start reaching quad core level CPUs.

So before you start bashing AMD for moving to DDR2, maybe you should have a better understanding of why they did it.

I'm not bashing for the move to DDR2. That's a good idea. The AM2 implementation of it isn't very good though. The memory controller on the AM2 chips is a bit rubbish. Nothing to do with bandwidth limitation, just poorly optimised for taking advantage of the extra available bandwidth of the memory.

They've implemented a solution that moves to newer technology without any of the performance increases usually associated with such a transition. They could have made it faster, it just seems like they're not even trying these days. Usually such a change would be accompanied by a partial redesign of the pipeline to optimise the CPU for taking advantage of more frequent memory access. Not incorporating such changes is just laziness on AMDs part and they need to suffer for it. Like Intel with Netburst.

The memory controller is fine and synthetic benchmarks show just how well it does perform.

As for making changes to the chip to utilize it isn't really possible without overhauling most of the chip.  Because the A64 has a pretty damn good BPU it doesn't miss a prediction very often.  With the rather large L1 cache, and more than sufficient L2 cache (it doesn't duplicate the L1 cache data in L2 like the P4 did), more often than not the data needed was in cache.  The A64 also has a good register system for logging previously accessed data, and with short pipelines, the penalty for a branch miss isn't that severe.

You're basically criticizing them for having a very efficient design that does not require a lot of bandwidth.  Does it necessarily make sense to make changes that may reduce efficiency just so it looks like adding DDR2 did make some kind of a performance difference?

Agent_Dung_Bomb wrote:

Bertster7 wrote:

Agent_Dung_Bomb wrote:

The current A64 architecture is not bandwidth limited, so there was not really an expectation that it would be faster.  The move to DDR2 was because that is where the market was going, and overall is a better design, allowing for larger capacity modules than DDR.  Also, DDR production was going to be reducing as manufacturers continued to ramp up DDR2 production, so DDR was going to become less common and more expensive.  Lastly, while the A64 isn't really bandwidth limited now, the move to DDR2 will help when you start reaching quad core level CPUs.

So before you start bashing AMD for moving to DDR2, maybe you should have a better understanding of why they did it.

I'm not bashing for the move to DDR2. That's a good idea. The AM2 implementation of it isn't very good though. The memory controller on the AM2 chips is a bit rubbish. Nothing to do with bandwidth limitation, just poorly optimised for taking advantage of the extra available bandwidth of the memory.

They've implemented a solution that moves to newer technology without any of the performance increases usually associated with such a transition. They could have made it faster, it just seems like they're not even trying these days. Usually such a change would be accompanied by a partial redesign of the pipeline to optimise the CPU for taking advantage of more frequent memory access. Not incorporating such changes is just laziness on AMDs part and they need to suffer for it. Like Intel with Netburst.

The memory controller is fine and synthetic benchmarks show just how well it does perform.

As for making changes to the chip to utilize it isn't really possible without overhauling most of the chip.  Because the A64 has a pretty damn good BPU it doesn't miss a prediction very often.  With the rather large L1 cache, and more than sufficient L2 cache (it doesn't duplicate the L1 cache data in L2 like the P4 did), more often than not the data needed was in cache.  The A64 also has a good register system for logging previously accessed data, and with short pipelines, the penalty for a branch miss isn't that severe.

You're basically criticizing them for having a very efficient design that does not require a lot of bandwidth.  Does it necessarily make sense to make changes that may reduce efficiency just so it looks like adding DDR2 did make some kind of a performance difference?

I'm criticising them for not altering that technology to take advantage of benefits from other components. I'm not suggesting they implement a huge pipeline like you might find in a P4 (although I think the K8 pipe is upto 17 stages now for floating point calculations), but with different resources it is inefficient not to change the CPU design to take advantage of them. Rather than simply trying to redo the system they've got in rather a halfarsed manner they should have redesigned the CPUs to take advantage of the higher frequency memory. A complete reworking of Hypertransport would be a start. Get that upto scratch and perhaps we'd see some real benefits from the DDR2 memory. As for the caching, both inclusive and exclusive caches have there pros and cons, duplicating data on L1 and L2 caches can reduce miss rates, which in many scenarios gives better performance.

In any case I totally stand by my statement that no one should be buying AM2 based CPUs because they are totally outperformed by the C2D CPUs for equivalent pricing. 939 setups are still worth a look because of the low cost to performance you get from them.

Last edited by Bertster7 (2006-12-07 09:14:35)

Bertster7 wrote:

Agent_Dung_Bomb wrote:

Bertster7 wrote:

I'm not bashing for the move to DDR2. That's a good idea. The AM2 implementation of it isn't very good though. The memory controller on the AM2 chips is a bit rubbish. Nothing to do with bandwidth limitation, just poorly optimised for taking advantage of the extra available bandwidth of the memory.

They've implemented a solution that moves to newer technology without any of the performance increases usually associated with such a transition. They could have made it faster, it just seems like they're not even trying these days. Usually such a change would be accompanied by a partial redesign of the pipeline to optimise the CPU for taking advantage of more frequent memory access. Not incorporating such changes is just laziness on AMDs part and they need to suffer for it. Like Intel with Netburst.

The memory controller is fine and synthetic benchmarks show just how well it does perform.

As for making changes to the chip to utilize it isn't really possible without overhauling most of the chip.  Because the A64 has a pretty damn good BPU it doesn't miss a prediction very often.  With the rather large L1 cache, and more than sufficient L2 cache (it doesn't duplicate the L1 cache data in L2 like the P4 did), more often than not the data needed was in cache.  The A64 also has a good register system for logging previously accessed data, and with short pipelines, the penalty for a branch miss isn't that severe.

You're basically criticizing them for having a very efficient design that does not require a lot of bandwidth.  Does it necessarily make sense to make changes that may reduce efficiency just so it looks like adding DDR2 did make some kind of a performance difference?

I'm criticising them for not altering that technology to take advantage of benefits from other components. I'm not suggesting they implement a huge pipeline like you might find in a P4 (although I think the K8 pipe is upto 17 stages now for floating point calculations), but with different resources it is inefficient not to change the CPU design to take advantage of them. Rather than simply trying to redo the system they've got in rather a halfarsed manner they should have redesigned the CPUs to take advantage of the higher frequency memory. A complete reworking of Hypertransport would be a start. Get that upto scratch and perhaps we'd see some real benefits from the DDR2 memory. As for the caching, both inclusive and exclusive caches have there pros and cons, duplicating data on L1 and L2 caches can reduce miss rates, which in many scenarios gives better performance.

In any case I totally stand by my statement that no one should be buying AM2 based CPUs because they are totally outperformed by the C2D CPUs for equivalent pricing. 939 setups are still worth a look because of the low cost to performance you get from them.

First of all, you don't just throw together the type of alterations that would be required to see the kind of benefits you seem to expect.  If they could have done that they would have K8L out the door.

What does the HT have to do with anything dealing with the memory?  Access to RAM still occurs over a traditional memory bus, not an HT bus.

Bertster7 wrote:

Agent_Dung_Bomb wrote:

Bertster7 wrote:

I'm criticising them for not altering that technology to take advantage of benefits from other components. I'm not suggesting they implement a huge pipeline like you might find in a P4 (although I think the K8 pipe is upto 17 stages now for floating point calculations), but with different resources it is inefficient not to change the CPU design to take advantage of them. Rather than simply trying to redo the system they've got in rather a halfarsed manner they should have redesigned the CPUs to take advantage of the higher frequency memory. A complete reworking of Hypertransport would be a start. Get that upto scratch and perhaps we'd see some real benefits from the DDR2 memory. As for the caching, both inclusive and exclusive caches have there pros and cons, duplicating data on L1 and L2 caches can reduce miss rates, which in many scenarios gives better performance.

In any case I totally stand by my statement that no one should be buying AM2 based CPUs because they are totally outperformed by the C2D CPUs for equivalent pricing. 939 setups are still worth a look because of the low cost to performance you get from them.

First of all, you don't just throw together the type of alterations that would be required to see the kind of benefits you seem to expect.  If they could have done that they would have K8L out the door.

What does the HT have to do with anything dealing with the memory?  Access to RAM still occurs over a traditional memory bus, not an HT bus.

Because the Hypertransport system is integrally linked to the speed of the FSB, due to the clock rate being used to drive the Northbridge. Changes to the FSB affect the overall memory bus width which in turn has an effect on the memory bandwidth. Whilst the K8 memory controller improves on request times it severely limits the bandwidth, which for DDR was not a problem (in fact it was a great advantage, what with the low latencies you get with DDR memory), with DDR2 though, it is an inelegant solution that does not take full advantage of the nice fast memory it has available. With multiple CPUs I can see the system taking full advantage of DDR2 memory, which could explain why AMD are following the route they are with 4x4 CPUs.

Well, you're in luck because the new HT will use separate PLLs and allow separate clock generation.

Bertster7 wrote:

Agent_Dung_Bomb wrote:

Bertster7 wrote:

Because the Hypertransport system is integrally linked to the speed of the FSB, due to the clock rate being used to drive the Northbridge. Changes to the FSB affect the overall memory bus width which in turn has an effect on the memory bandwidth. Whilst the K8 memory controller improves on request times it severely limits the bandwidth, which for DDR was not a problem (in fact it was a great advantage, what with the low latencies you get with DDR memory), with DDR2 though, it is an inelegant solution that does not take full advantage of the nice fast memory it has available. With multiple CPUs I can see the system taking full advantage of DDR2 memory, which could explain why AMD are following the route they are with 4x4 CPUs.

Well, you're in luck because the new HT will use separate PLLs and allow separate clock generation.

Now that makes a bit more sense.

That's exactly the point I'm making, AMD should have finished all this stuff before releasing AM2. As it stands it just seems that they've rushed it to market to compete with Intel and done a botched job.

Well, like I said.  The AM2 wasn't about designing a new processor.  It's main goal was to get people migrated to DDR2 to prevent AMD users from getting trapped in the death throws of DDR; which would have happened before K8L and quad core processors made it to market.