THG-Preview of upcoming AM2 4800+ with benchmarks

[classy]

Well its slower no question, which even at this point you would think it should be faster. I don't really have an opinion, but it looks like the high latency of DDR2 just saps any real tangible performance gain.

Socket AM2

 

[Furen]

Remember that the socket AM2 system was running on a pre-production motherboard with very relaxed timings (hell, it could even have a 2T command rate) while socket 939 was running at 2-2-2 DDR400. That said, I was quite surprised that neither PCMark's nor Sandra's memory bandwidth benchmarks saw any bandwidth improvement.

 

[BrownTown]

yeah, the fact that a completly synthetic benchmakr like the Sandra memmory bandwidth one showed only 1% immprovement in bandwidth means there is a problem somewhere. Even at DDR2-667 it should be 66% more bandwidth in the ideal case. So, it almost looks as though something is screwed up and its actually only running at DDR2-400, or that some other problem is occuring that is not letting the bandwidth get where it should be. The fact that the AM2 lost so bad indicates to me that something is wrong with the BIOS/MOBO, or that the chips memmory interface isnt running at the correct speed.

 

[Viditor]

Also remember that THG is using an older ENG sample...quite probably without the mem-controller revamp.

Edit: THG has the memory base clock at 100.5 MHz (look at CPU-Z)... it was pointed out to me that this means Tom's is running the memory as DDR2-400 for the test.

 

[BrownTown]

that would certinaly be consistant with the benchmark results with the bandwidth being identical...

 

[JackPack]

Originally posted by: ViditorEdit: THG has the memory base clock at 100.5 MHz (look at CPU-Z)... it was pointed out to me that this means Tom's is running the memory as DDR2-400 for the test.

100 MHz would be DDR2-200. Looking at the benchmarks, that doesn't seem to be the case.

At this point, I would not put much faith in CPU-Z.

 

[n7]

I would never trust THG, but TBH, i also didn't expect much improvement.

How many times has it been proved than high memory frequencies provide minimal benefits for A64s...yet somehow no one remembers this...

 

[Hacp]

repost and its probably due to the ddr2 bug

 

[PingSpike]

Its Toms Hardware so who the hell knows? But I wasn't expecting much on this either.

 

[RichUK]

Its a pointless upgrade IMO if you already have a dual core s939 platform. Although this review is a bit borked due to the DDR2 memory controller bug, don?t expect to see much more performance from a fully working production chip (Unless you feel sandra mem benchies equate to anything). AMD?s are not that dependant on memory performance (speed or latency), so moving to DDR2 will not yield much of an improvement in the early stages of AM2 over using current slower DDR technology.

I patiently await 65nm AMD's, as this WILL bring something new to the plate, and IMO will be a more worthwhile upgrade.

EDIT: Something that does interest me is this future 35W X2 3800+

 

[BrownTown]

it should be noted that the ICs in DDR2-400 run at 100mhz, so if CPU-Z was looking at the IC frequency like it would for DDR1 then it would read 100. Either way it should send red flags everywhere.

 

[Bull Dog]

Bottom line: it seems that Tom's had yet another borked review designed to make people think that AMD is doomed to die and fade away into oblivion.

 

[BassBomb]

the new AM2 isnt a new architecture right?

 

[dexvx]

Originally posted by: Bull Dog
Bottom line: it seems that Tom's had yet another borked review designed to make people think that AMD is doomed to die and fade away into oblivion.

Why? Most people were expected a nil difference. Were you expecting a 20% boost from this or something?

 

[ayabe]

Personally I think AMD should just forgo DDR2 and wait for DDR3 so we can benefit from some real performance increases. Let Intel deal with DDR2 and if DRAM makers don't like it then tough titties.

 

[Furen]

Originally posted by: BrownTown
it should be noted that the ICs in DDR2-400 run at 100mhz, so if CPU-Z was looking at the IC frequency like it would for DDR1 then it would read 100. Either way it should send red flags everywhere.

I dont think there's any way for CPU-Z to look at the clock of the mem cells themselves. It normally looks at the external clock which, if running at DDR2 400, would still be 200MHz.

 

[dajo]

I don't understand why is this new platform being introduced. Am I missing something here or did that article conclude that there is no performance gain in moving to this new platform?

Unless forthcoming higher ddr speeds with tighter latencies can show some real performance gains I think I'll sit this one out.

So, you have to get a new board, new chip, and, potentially, new ram so that you can basically wind up with the same performance you have now?

Why?

 

[RichUK]

Originally posted by: dajo
I don't understand why is this new platform being introduced. Am I missing something here or did that article conclude that there is no performance gain in moving to this new platform?

Unless forthcoming higher ddr speeds with tighter latencies can show some real performance gains I think I'll sit this one out.

So, you have to get a new board, new chip, and, potentially, new ram so that you can basically wind up with the same performance you have now?

Why?

BINGO!!!!

65nm AM2 is realistically the next gen AMD processor even if it is still based on K8.

 

[Hacp]

Originally posted by: dajo
I don't understand why is this new platform being introduced. Am I missing something here or did that article conclude that there is no performance gain in moving to this new platform?

Unless forthcoming higher ddr speeds with tighter latencies can show some real performance gains I think I'll sit this one out.

So, you have to get a new board, new chip, and, potentially, new ram so that you can basically wind up with the same performance you have now?

Why?

Main reason is DDR2. DDR prices WILL eventually rise, and AMD guys will be at a price disadvantage.

2nd reason is that the new mobos will support AMd's future 65nm process.

 

[hans007]

Originally posted by: JackPack

Originally posted by: ViditorEdit: THG has the memory base clock at 100.5 MHz (look at CPU-Z)... it was pointed out to me that this means Tom's is running the memory as DDR2-400 for the test.

100 MHz would be DDR2-200. Looking at the benchmarks, that doesn't seem to be the case.

At this point, I would not put much faith in CPU-Z.

ddr2 is quad pumped.

that said, ddr2 at 400mhz is CL3 so the latency is no worse than ddr400 cl3. so i guess since they were comparing it to pc3200 cl2 ram, that is where you performance difference is.

overall though, still pretty disappointing. i dont think the athlon 64 is remotely bandwidth starved so i doubt ddr2-800 will do much for it. i mean a dual channel athlon 64 is barely faster than a single channel one at same clock, so its not a huge bandwidth problem.

 

[carlosd]

it's amazing THG saying that:
"AMD is going to introduce low-powered processors for power and noise sensitive environments. The single cores will maintain a 35 W power envelope, while the dual cores won't exceed 65 W. The latter is the mainstream thermal design that Intel intends to introduce with its upcoming Conroe processors in Q3. What we find amazing is that AMD must be able to reach these envelopes using its 90 nm technology, while Intel had to switch to 65 nm in order to maintain this course."

and also...
"Three years ago, AMD could never have afforded to delay a major technology launch for the sake of waiting for the ideal moment (or to make sure the new chips are available in sufficient quantities). Today, the Athlon 64 processor family is widely considered superior, which puts AMD into a role as leader rather than as a struggling competitor."

 

[kknd1967]

Just a quick update since so many ppls were complaining about THG's DDR2 set up for AM2
In this new benchmark, as expected, DDR2-800 still does not help much anyway

Link

AMD Athlon 64 4800+ Socket AM2 (2.4GHz/1MB L2 x2)
AMD Athlon 64 4800+ Socket AM2 (2.4GHz/1MB L2 x2)
nForce4 Ultra Socket 939 Motherboard
nForce4 Ultra Socket AM2 Motherboard
Corsair DDR2 800 512MB x 2(CL 5-5-5-15)
Geil DDR400 512MB x 2 (CL 2.5-3-3-7)
Gigabye Geforce 7900GTX 512MB (650MHz/1.6Ghz)
Magic-Pro MagicPower 550W
Windows XP Professiona w/ SP2
nVidia Forceware 84.21
nForce Driver for AMD 6.70





Socket AM2 VS Socket 939 @ 4800+

System Test

Socket AM2
Athlon 64 X2 4800+ Socket 939
Athlon 64 X2 4800+
PC Mark 05
Total 5569 5572
CPU 4926 4928
Memory 4429 4298
Graphic 6684 6664
Super PI
1M Test 35s 35s
4M Test 3m02s 3m01s
Sandra 2005
ALU 20339 20314
FPU 7631 7628
SSE2 9882 9877
Integer 45620 45831
Floating 49670 49653
Memory Bandwidth Int 6197 5818
Memory Bandwidth Float 6137 5820
Cache Combined Index 6683 6524
Cache Speed Factor 7.1 7.5
Science Mark 2
Total 1292.30 1248.68
Moleciular 1233.33 1233.34
Primordia 1189.26 1188.27
Cryptography 1236.22 1235.85
Stream 1574.51 1248.43
Memory 1461.23 1387.22
Flop 1284.82 1268.14
DivX 6.1
Mpeg 4 Encoding 5:43 5:40
Xvid 1.1
Mpeg 4 Encoding 1:52 1:46
C.C Winstone
Overall 37.1 37.2
Business Winstone
Overall 34.4 34.6

Graphics Test @ Geforce 7900GTX

Socket AM2
Athlon 64 X2 4800+ Socket 939
Athlon 64 X2 4800+
3DMark 05
Default 9351 9359
CPU Test Total 6124 6071
CPU Test 1 2.7 2.6
CPU Test 2 6.3 6.2
3DMark 06
Default 6853 6856
CPU 1824 1811
CPU Test 1 0.588 0.578
CPU Test 2 0.906 0.908
F.E.A.R
1600 x 1200 4AA 8AF 53.8 54.1
Doom3
1600 x 1200 4AA 8AF 81.4 81.6
Serious Sam 2
1600 x 1200 4AA 8AF 58.5 58.4
Splinter Cell CT
1600 x 1200 4AA 8AF 63.8 63.9

 

[Duvie]

NVM... I thought those were THG numbers.....another post goes into it more...

It obvious shows video performance better then THG did with dd2-667...so there is improvement though the synthetic benches must be borked cause they show NO increase in theoretical...which they should have