newcastle and clawhammer?

whats up with the newcastle cores? why are they half the L2 (512K) of the clawhammer (1M)? are they supposed to be a cheapo version of the A64? i did notice at the same rated clock speed 2.0g they have different PR, 3000+newcastle, and 3200+ clawhammer!

so i take it the clawhammer is still the a better cpu? and the newcastle is for the low budget man correct?

The Clawhammers are being phased out, it seems. The NewCastles will have a dual-channel memory controller in the near future to offset the lack of cache. I think the reason that AMD is doing this is to differentiate the normal A64's from the FX's. With the full 1MB of cache, they're virtually one in the same. Castrating them keeps costs low for AMD and gives enthusiasts more motivation to buy the high-end.

ClawHammer at 2000 MHz is considered the same PR rating as 2200 MHz NewCastle according to the AMD model number.

3200+ 2000 MHz L2 1 MB (ClawHammer)
3200+ 2200 MHz L2 512 KB (NewCastle)


2800+ 1800 MHz L2 512 KB (NewCastle)
3000+ 2000 MHz L2 512 KB (NewCastle)
3200+ 2000 MHz L2 1 MB (ClawHammer)
3200+ 2200 MHz L2 512 KB (NewCastle)
3400+ 2200 MHz L2 1 MB (ClawHammer)

Recently, I did some analysis, I came to a conclusion that for NEW build system, a 754 system w/ 250 GB chipset can be as cost effective as a Nforce2 + mobile Barton, with 754 being $100 higher but it delivers potentially 15-25% better performance after CPU overclocked. Detailed analysis see link below.

http://www.ocforums.com/showthread.php?s=&postid=2770023#post2770023

I think this took AMD a bit by surprise...

It looks like the on-die memory controller and the reduced latency for access to main memory reduces the dependency on huge L2 caches to maintain performance.

The 512MB cache chips should be cheaper than the 1MB versions due to the slight performance difference, but mostly due to the reduced die size or the use of clawhammer dies with flawed cache memory. Either way, it helps AMD by improving chip yields from each wafer and thus reduces the cost of production.

I am not sure if the Newcastle supports the higher speed 1GHz Hyper Transport bus, but this would make it somewhat better if paired with a K8T800 Pro based board.

800 MHz HT bus is the rated specification of HT bus over all the processors. The 1 GHz HT bus may be just a "turbo" or "speedup" implementation by the chipset and motherboard makers.

Helps AMD, hurts the consumer. I'd be much happier with a 2 GHz 1 MB L2 part like the one I have than a 2.2GHz 512k part. AMD's short-pipeline architecture has always been able to get by just fine with relatively low amounts of cache, but the more, the merrier. http://www.xtremesystems.org/forums/showthread.php?s=&threadid=35186
The nF3 250 and K8T800 Pro both support a 1 GHz HT, and it shouldn't be long until processors roll out that will inherently support it, too. Although, I should think that by changing the LDT from 4x to 5x on a suitable mobo, you'd end up reaching a 1 GHz even with an older processor.

hitechjb1 said:

800 MHz HT bus is the rated specification of HT bus over all the processors. The 1 GHz HT bus may be just a "turbo" or "speedup" implementation by the chipset and motherboard makers.

Click to expand...

Really? I thought that 1GHz was slated to be spec pretty soon?

Gautam said:

The Clawhammers are being phased out, it seems. The NewCastles will have a dual-channel memory controller in the near future to offset the lack of cache. I think the reason that AMD is doing this is to differentiate the normal A64's from the FX's. With the full 1MB of cache, they're virtually one in the same. Castrating them keeps costs low for AMD and gives enthusiasts more motivation to buy the high-end.

Click to expand...

If ClawHammer is/will be being phased out, then what core would a dual channel 1 MB L2 A64 (aka 939) based on?

I meant s-754 Clawhammer. The 939 Clawhammer just looks an awful lot more like a Sledgehammer than anything else. I'm doubting if these code names even much at all anymore, as the 939's and 754's are quite different processors.

no, the dependancy on the L2 cache is not inherently tied to the system ram and therefore not so much adversely affected by it, but more so dependant on the L1's ability and therefore its inability to fit everything in

the inclusive nature of the L1 vs L2 cache on the intels force the L2 to be be otherwise bigger, as otherwise adversely affecting performance


and hence the exclusice nature of the L1 vs L2 cache on the amd's would therefore mean that the bigger L2 cache would not give as much of a performance boost as it would on the intels, but would perform better than the intels cache system assuming if L2>L1 by at least 400%, and also assuming that the cache size is equal

intel gets around the problem by using L1 with lower latency, and therefore better performance, alleviating the performance difference when compared to the exclusive nature of the amd cache system

so therefore the integrated memory controller affects performance more than it should

the reduction in cache size for the newer amd chips are more of a cost reduction reason rather than performance related reason

less cache = less chip area = cheaper chips


cheaper chips = more chips = less overhead costs = more market = more profit

fafnir said:

the reduction in cache size for the newer amd chips are more of a cost reduction reason rather than performance related reason

less cache = less chip area = cheaper chips


cheaper chips = more chips = less overhead costs = more market = more profit

Click to expand...

To finish it up...

less cache = more profit



So if i read this correctly, AMD is turning the mid to highend ranges of CPU's to 512k cache and dual channel, and keeping the "extreme" end to dual channel and 512k?

If the dual channel offsets the performance loss for the lack of cache well, i say this's a good decision by AMD.

Chowdy said:

To finish it up...

less cache = more profit



So if i read this correctly, AMD is turning the mid to highend ranges of CPU's to 512k cache and dual channel, and keeping the "extreme" end to dual channel and 512k?

If the dual channel offsets the performance loss for the lack of cache well, i say this's a good decision by AMD.

Click to expand...

i myself see the dual channel as a better benifit on the larger L2.

henceforth the clawhammer 754 is getting fased out and everything is going to 939 the regular athlon64 is only going to have the 512mb cache while the Athlon fx series will be the only chip with the 1mg cache.

while i do think its a good idea to have a cheapo version with the 512mb catche for the people on a budget! but that also means that the hardcore PC junkie is going to have to pay top doller for the FX if he wants to have the larger 1M cache? witch i think sucks a big fatty!