AMD Zambezi news, info, fans !

[Pestilence]

First post, Anxiously awaiting BD.

BLT has the 8150, 8120 and 6100 models on there site but

This product is not in stock, and is not yet on order with the manufacturer but should be shortly

8150 - 266.28

http://www.shopblt.com/cgi-bin/shop/...r_id=590972573

8120 - 221.73

http://www.shopblt.com/cgi-bin/shop/...r_id=590972573

6100 - 188.32

http://www.shopblt.com/cgi-bin/shop/...r_id=590972573

Conclusion i draw from those prices are as follows..

2700K>8150>2500K>8120

[Oese]

would make complete sense..

and tbh is not too bad.. i dont want to pay more then 300$ for my new high-end amd so it must not be faster then 2600K

[Pestilence]

would make complete sense..

and tbh is not too bad.. i dont want to pay more then 300$ for my new high-end amd so it must not be faster then 2600K

Agreed. Thinking very hard about ordering one today

[rog]

AMD will try to keep the discussion away from performances as long as possible. You'll see next week what I'm talking about

There is no launch next week, so what's your point? Or do you refer to the developer central meeting next week? I think this is all about Fusion.

[BeepBeep2]

I thanked you for that post, simply because of the car analogy usage lol

You could also do the "blasphemous" move and swap out the MT for an AT, then equip it with a 3500 Stall so it launches immediately in the power band A lot of the problem could easily be the fact that the Neon (sans SRT-4) was not intended for any sort of performance, so an overhaul of the computer's programming could do a world of difference (remap spark curves, AF:R table, tune the knock-retard table). Factory tune is sort of like the whole Intel Compiler use with AMD fiasco, where if you just run it as is you're not getting all the performance you could be heh

As for the cat: Schrödinger's Skun Cat Theory - The cat has neither been skinned, nor does it have skin!
(Definition for our foreign friends, -- ^^^^^ -- since it's not a commonly used term)



In the US I don't see it being that much for a system build, even for a A8-3850 (which I think the 3550 in that chart might be representative of what the 3850 is).
A8-3850: ~$130
Mobo: $65-140
RAM: 2x4GB - $30-40 (DDR3-1600)
PSU: $40-50 (for a quality 600W with MIR, which is overkill)
Case: $40-90
HDD: $40-55
Optical: $20
-------------------
Total: $525 (at most)

Monitor, keyboard, mouse, general purpose speakers: ~$155 (monitor would be name brand ~19-21 inch, but relatively inexpensive mouse/keyboard/speakers)
-------------------
Complete System: $680

I am actually building one of these for a customer soon.

A8-3850 - $130
Mobo with DVI+VGA+HDMI, USB3 = $100
DDR3 (2x2GB DDR3-1600) $30
PSU - Corsair Builder Series 430w = $42, $29 after MIR
CASE - Lian Li Lancool PC-K58 = $60
1TB Seagate HDD = $50
Optical w/ Lightscribe = $22
OS that you forgot - Win7 HP = $100
Keep current monitor as they bought one within the last few years, keep KB/Mouse as they have one = $0

Price: $534

[Formula350]

I am actually building one of these for a customer soon.

A8-3850 - $130
Mobo with DVI+VGA+HDMI, USB3 = $100
DDR3 (2x2GB DDR3-1600) $30
PSU - Corsair Builder Series 430w = $42, $29 after MIR
CASE - Lian Li Lancool PC-K58 = $60
1TB Seagate HDD = $50
Optical w/ Lightscribe = $22
OS that you forgot - Win7 HP = $100
Keep current monitor as they bought one within the last few years, keep KB/Mouse as they have one = $0

Price: $534

I completely forgot about an OS, but even still, as you pointed out a person can save if they can reuse any old hardware. If not... there's always Linux for free lol

I think those specs will make a really good all-around system, more-so than the G/GX chipset motherboards with a similarly clocked Athlon II X4. There's quite a bit of upgrade potential with these, which I hope AMD can stick with that socket for a few years so people can upgrade bits and pieces here and there

[maltrabob]

AMD will try to keep the discussion away from performances as long as possible. You'll see next week what I'm talking about

Could you, please, stick to "speech is silver, silence is golden" unless you really want to tell us something useful. Ta.

[informal]

First post, Anxiously awaiting BD.

BLT has the 8150, 8120 and 6100 models on there site but



8150 - 266.28

http://www.shopblt.com/cgi-bin/shop/...r_id=590972573

8120 - 221.73

http://www.shopblt.com/cgi-bin/shop/...r_id=590972573

6100 - 188.32

http://www.shopblt.com/cgi-bin/shop/...r_id=590972573

Conclusion i draw from those prices are as follows..

2700K>8150>2500K>8120

Thanks for posting this. Looking at the price of 8150 I wonder what will happen to 1100T after FX launches. If the results leaked so far are accurate then 8150 will loose quite a bit of benchmarks to 1100T.How will they justify 266$ price when you can buy older generation of their own products that can perform the same or better while costing 26% less(AMD lists 210$ for 1100T on their website as price for distributors). If 1100T goes down in price after FX launches(which always happens with new product launches),then picture is even worse for FX. Then you will have even more affordable Thuban X6 ( <200$ ?) at 3.3Ghz that OCs easily to 3.8Ghz+ on air. Price/perf. ratio of FX will then be even worse IMO.
Note this is all based on what we know via the leaks and what people in the know (such as xsecret and chew*) hinted so far.
Nothing makes sense to me.

PS And yes,I know 32nm means higher OC for FX but who cares about that when you can OC Thuban as well.With IPC deficiency you will end up at the same spot with Thuban @4Ghz even if you push FX 81xx to 4.5Ghz+ ...

[radaja]

Heres x2 opteron 6220 8c at 2.75GHz from sisoft,although cpu world
has them listed as 3.0Ghz cpu's.its from sept 7th 2011
Dont know if this is useful or not

2x8c opteron 6220@2.75Ghz

Scores
Processor Arithmetic Benchmark
score-138.538 GOPS

Processor Multi-Media Benchmark
score-315.004 Mpix/s

[Oese]

Thanks for posting this. Looking at the price of 8150 I wonder what will happen to 1100T after FX launches. If the results leaked so far are accurate then 8150 will loose quite a bit of benchmarks to 1100T.How will they justify 266$ price when you can buy older generation of their own products that can perform the same or better while costing 26% less(AMD lists 210$ for 1100T on their website as price for distributors). If 1100T goes down in price after FX launches(which always happens with new product launches),then picture is even worse for FX. Then you will have even more affordable Thuban X6 ( <200$ ?) at 3.3Ghz that OCs easily to 3.8Ghz+ on air. Price/perf. ratio of FX will then be even worse IMO.
Note this is all based on what we know via the leaks and what people in the know (such as xsecret and chew*) hinted so far.
Nothing makes sense to me.

PS And yes,I know 32nm means higher OC for FX but who cares about that when you can OC Thuban as well.With IPC deficiency you will end up at the same spot with Thuban @4Ghz even if you push FX 81xx to 4.5Ghz+ ...

part of the problems we have with clear reasoning about this all is the totally new architecture. bd has 4 cores, that are stronger as one thuban core, but to be reallystrong need at least two threads. so in some scenario they outperform thuban whilst in others maybe thuban hold some ground. bd is designed to cure its weak one-threaded performance by matter of clock speed, which at the moment might not be enough due to problems in the process.

whatever, where thuban is strong, like multithreaded workloads, i expect bd also to be quite strong. All in all i guess it's a clear update over thuban, but having problems to compete with intels one-thread-strong-ipc sandybridges...

anyway, i can see now, that bd will for most cases be a good processor. only the 1-core/multicore benchmark logic will not fit it very well..

[informal]

Well it can't hurt. Thanks for posting this radaja! Weird thing is the clock Sisoft reads. The model should be 8C 3Ghz base clock and 3.5Ghz Turbo,at least this is the spec according to Gateway server configuration menu (now pulled).

For 8C zambezi with Turbo for integer cores (500Mhz presumably?) that ran at 2.8Ghz/3.3Ghz:
Scores
Processor Arithmetic Benchmark
score-56.64 GOPS

Processor Multi-Media Benchmark
score-147 Mpix/s

Base clock difference between Zambezi and that opteron is 7%. Similar goes if you count the all core Turbo clocks :3.5/3.3=1.06 or 6%. So let's take 6.5% as mean value.
So 3Ghz zambezi with 3.5Ghz should score :
Processor Arithmetic Benchmark
score-60 GOPS

Processor Multi-Media Benchmark
score-157 Mpix/s

Scaling of Interlagos (since it's 2P system in question) is probably very good,close to perfect. Assume 1.95x,that takes a few % from perfect scaling due to software-hardware limitations.
For 1P 3Ghz Opteron then we should have:
Processor Arithmetic Benchmark
score-138.538/1.95=71 GOPS

Processor Multi-Media Benchmark
score-315.004/1.95=161 Mpix/s

Now compared to Zambezi platform @ 3Ghz,this Opteron is:
71/60=18% faster in Processor Arithmetic Benchmark
161/157=2% faster in Processor Multi-Media Benchmark

Practically Integer performance is somehow off in Zambezi's case (3Ghz+Turbo of 500Mhz) while SIMD is almost the same as with 1P 3Ghz Opteron (3Ghz + Turbo of 500Mhz).
This leads us to conclusion that at least there might be something wrong with integer performance. SIMD is not promising though since Opteron scores the same.
The only problem with above "math" is that Sisoft sees the Opteron system @ 2.75Ghz instead of 3/3.5Ghz. This may be error in detection or the samples may actually run at 2.75Ghz,we don't know. But ,Opteron's performance should be final as the chips are shipping to partners.

I'm going to search for 2P Lisbon 6140 (2.6Ghz) sisoft results and compare them to this 2P Valencia system.

[Pestilence]

Thanks for posting this. Looking at the price of 8150 I wonder what will happen to 1100T after FX launches. If the results leaked so far are accurate then 8150 will loose quite a bit of benchmarks to 1100T.How will they justify 266$ price when you can buy older generation of their own products that can perform the same or better while costing 26% less(AMD lists 210$ for 1100T on their website as price for distributors). If 1100T goes down in price after FX launches(which always happens with new product launches),then picture is even worse for FX. Then you will have even more affordable Thuban X6 ( <200$ ?) at 3.3Ghz that OCs easily to 3.8Ghz+ on air. Price/perf. ratio of FX will then be even worse IMO.
Note this is all based on what we know via the leaks and what people in the know (such as xsecret and chew*) hinted so far.
Nothing makes sense to me.

PS And yes,I know 32nm means higher OC for FX but who cares about that when you can OC Thuban as well.With IPC deficiency you will end up at the same spot with Thuban @4Ghz even if you push FX 81xx to 4.5Ghz+ ...

Is this processor really an 8 core or more of a 4 core with 8 threads? I've heard both sides and i don't know what to believe. I also don't expect the IPC to be a hell of a lot better then Thuban but i do expect it to hit over 4.5Ghz with the 8150. Have a feeling the 8120's overclock ceiling is going to be very low.

[undone]

@informal
IMO those interlagos sisoft result have many traps in them, be careful to try to reach any conclusion.

[radaja]

@informal,yea thats what i thought too about the clock speed,everywhere
i have checked says 3.0GHz yet that sisoft shows 2.75Ghz?

[informal]

Well now we have 2 of them:
Details for Device 2x AMD Opteron 6220
2x AMD Opteron 6282 SE

To save you the boring math and long story,the results kinda match with only one thing standing out. Multi-Media Benchmark performance of the 6220 (per core!) is kinda low compared to 6282 (per core). Now I'm pretty sure I know at what clock the 6282 runs. It's either 2.5Ghz base or 2.6Ghz base. I assume it's 2.5Ghz. So the Multi-Media Benchmark for 1P 6282SE would roughly be around : 585/1.95=300 mpix/s which is only 5% lower than what supposedly 2P 3Ghz Valencia gets . If we take that 2.5Ghz 1P 6282SE scores 300 then 3Ghz 1P 6282SE would get ~360 and then this would imply 8C performance at 3Ghz is around 180Mpix/s . Compare to Zambezi's(8C 3Ghz) 157Mpix/s and 6220's (1P 8C 3Ghz) 161Mpix/s and you see the difference : it's better by 15%/12% roughly.

So to conclude : Zambezi's result deviate from Opterons ,per core and per clock,by: 15-18% in integer benchmark and 3-15% in Multimedia(AVX) benchmark (depending on the opterons used for comparison). One small caveat though : I have noticed that Lisbon platform scores around 30% better at the same clock versus Thuban ! I think this may be due to NUMA (2P Lisbons). So if this is translated to Interlagos too,then none of the above is relevant and Zambezi's original (bad) Sisoft results may even be genuine...

edit: oh and one more thing I forgot to mention. Interlagos at the same core count and roughly same clock is noticeably slower in integer benchmark than Lisbon/MC (generation). Interlagos can clock higher though but IPC difference is very high (think more than 15%). and yes it is noticeably faster in Multimedia one but integer represents a big chunk of both server and desktop workloads,so it is kind of a letdown IMO. All that talk about integer cores being faster etc. and then we get this... I guess it still may not be final platform/BIOS but it doesn't look promising in integer workloads.

[BeepBeep2]

I completely forgot about an OS, but even still, as you pointed out a person can save if they can reuse any old hardware. If not... there's always Linux for free lol

I think those specs will make a really good all-around system, more-so than the G/GX chipset motherboards with a similarly clocked Athlon II X4. There's quite a bit of upgrade potential with these, which I hope AMD can stick with that socket for a few years so people can upgrade bits and pieces here and there

On the other hand...
System with Phenom II X4 925 overclockable to ~3.5 @ 250 HTT + ASUS AM3+ mobo with UEFI, USB3, Bulldozer ready + 8GB DDR3-1600 CL9 @ 1666 250 HTT + HD6670 (would perform better GPU wise) is $580.

I just realized this as newegg tends to run a lot of combo deals...that system has an upgrade path. I know the person I am building for quite well, who knows...maybe in a year or so they will want to upgrade to Bulldozer?

[freeloader]

Seems like we have a total stinker headed our way. I will be thoroughly peeved at AMD if they've had us in the dark this long for what amounts to garbage.

[Oese]

Well now we have 2 of them:
Details for Device 2x AMD Opteron 6220
2x AMD Opteron 6282 SE

To save you the boring math and long story,the results kinda match with only one thing standing out. Multi-Media Benchmark performance of the 6220 (per core!) is kinda low compared to 6282 (per core). Now I'm pretty sure I know at what clock the 6282 runs. It's either 2.5Ghz base or 2.6Ghz base. I assume it's 2.5Ghz. So the Multi-Media Benchmark for 1P 6282SE would roughly be around : 585/1.95=300 mpix/s which is only 5% lower than what supposedly 2P 3Ghz Valencia gets . If we take that 2.5Ghz 1P 6282SE scores 300 then 3Ghz 1P 6282SE would get ~360 and then this would imply 8C performance at 3Ghz is around 180Mpix/s . Compare to Zambezi's(8C 3Ghz) 157Mpix/s and 6220's (1P 8C 3Ghz) 161Mpix/s and you see the difference : it's better by 15%/12% roughly.

So to conclude : Zambezi's result deviate from Opterons ,per core and per clock,by: 15-18% in integer benchmark and 3-15% in Multimedia(AVX) benchmark (depending on the opterons used for comparison). Once small caveat though : I have noticed that Lisbon platform scores around 30% better at the same clock versus Thuban ! I think this may be due to NUMA (2P Lisbons). So if this is translated to Interlagos too,then none of the above is relevant and Zambezi's original (bad) Sisoft results may even be genuine...

edit: oh and one more thing I forgot to mention. Interlagos at the same core count and roughly same clock is noticeably slower in integer benchmark than Lisbon/MC (generation). Interlagos can clock higher though but IPC difference is very high (think more than 15%). and yes it is noticeably faster in Multimedia one but integer represents a big chunk of both server and desktop workloads,so it is kind of a letdown IMO. All that talk about integer cores being faster etc. and then we get this... I guess it still may not be final platform/BIOS but it doesn't look promising in integer workloads.

well, one integer thread in bulldozer is not as strong as one integer thread in thuban, but tuban has six and bulldozer 8. so if single-thread integer performance is kinda low on bd, it doesnt surprise me... bd needs turbo to solve that.

[informal]

I will be thoroughly peeved at AMD if they've had us in the dark this long for what amounts to garbage.

Well it may not be garbage in server segment. On the contrary it may be pretty good ,even versus SB-E.
On desktop though,if nothing changes,many websites will label it as a letdown in their reviews.

well, one integer thread in bulldozer is not as strong as one integer thread in thuban, but tuban has six and bulldozer 8. so if single-thread integer performance is kinda low on bd, it doesnt surprise me... bd needs turbo to solve that.

IMO this doesn't make sense. We are now in 2011. AMD knew what kind of cores it will have to fight this year and in the years to come. Core generation already had pretty big advantage in integer workloads(some due to compiler advantages some due to better uarchitecture). Now I'm supposed to think that even though AMD knew the gap will be around 35-40% ,at similar clocks,they will go with 15% slower design (IPC) versus their own familiy 10h and hope to make up all this by clocking it sky high? They would need 5.5-6Ghz to match QC SB with SMT which will never happen. Bulldozer can barely touch 4Ghz with Turbo on and this is not on all cores. Something is not right here...

[Manicdan]

In the US I don't see it being that much for a system build, even for a A8-3850 (which I think the 3550 in that chart might be representative of what the 3850 is).
A8-3850: ~$130
Mobo: $65-140
RAM: 2x4GB - $30-40 (DDR3-1600)
PSU: $40-50 (for a quality 600W with MIR, which is overkill)
Case: $40-90
HDD: $40-55
Optical: $20
-------------------
Total: $525 (at most)

Monitor, keyboard, mouse, general purpose speakers: ~$155 (monitor would be name brand ~19-21 inch, but relatively inexpensive mouse/keyboard/speakers)
-------------------
Complete System: $680

Use the cheaper motherboard and case, and they knock it down to $555! That's lower than the Lynx with E2 APU prediction

i was looking at things like laptops where the price is already summed up based on all the specs and the market kinda dictates prices much more than individual component competition. and keep in mind that the top end it mentions dual gpu, so thats another 50-70$, then beep pointed out you missed the os for 100$ right there. bringing the total to $800+
things seem to be about 20% over those prices (which are system prices, not user needs upgrade by pc knowledgeable friend prices)

[Oese]

Well it may not be garbage in server segment. On the contrary it may be pretty good ,even versus SB-E.
On desktop though,if nothing changes,many websites will label it as a letdown in their reviews.



IMO this doesn't make sense. We are now in 2011. AMD knew what kind of cores it will have to fight this year and in the years to come. Core generation already had pretty big advantage in integer workloads(some due to compiler advantages some due to better uarchitecture). Now I'm supposed to think that even though AMD knew the gap will be around 35-40% ,at similar clocks,they will go with 15% slower design (IPC) versus their own familiy 10h and hope to make up all this by clocking it sky high? They would need 5.5-6Ghz to match QC SB with SMT which will never happen. Bulldozer can barely touch 4Ghz with Turbo on and this is not on all cores. Something is not right here...

i'm not sure... maybe some prefetch tricks should add up to turbo to solve this issue, and maybe there or with cache something is still missing.. but who knows..

[FlanK3r]

Well it may not be garbage in server segment. On the contrary it may be pretty good ,even versus SB-E.
On desktop though,if nothing changes,many websites will label it as a letdown in their reviews.



IMO this doesn't make sense. We are now in 2011. AMD knew what kind of cores it will have to fight this year and in the years to come. Core generation already had pretty big advantage in integer workloads(some due to compiler advantages some due to better uarchitecture). Now I'm supposed to think that even though AMD knew the gap will be around 35-40% ,at similar clocks,they will go with 15% slower design (IPC) versus their own familiy 10h and hope to make up all this by clocking it sky high? They would need 5.5-6Ghz to match QC SB with SMT which will never happen. Bulldozer can barely touch 4Ghz with Turbo on and this is not on all cores. Something is not right here...

u right, if is it true, then :-(...AMD must know power of SB and Q1/Q2 netx year of IB. They need +- performance in single thread as Nehalem. It will be OK, still slower in ST, but very impressive in module architecture in multithread (MT). What is the logic for lower single thread performance in new generation architecture for next years than 2.5 years old Phenom II?

[PerryR]

IMO this doesn't make sense. We are now in 2011. AMD knew what kind of cores it will have to fight this year and in the years to come. Core generation already had pretty big advantage in integer workloads(some due to compiler advantages some due to better uarchitecture). Now I'm supposed to think that even though AMD knew the gap will be around 35-40% ,at similar clocks,they will go with 15% slower design (IPC) versus their own familiy 10h and hope to make up all this by clocking it sky high? They would need 5.5-6Ghz to match QC SB with SMT which will never happen. Bulldozer can barely touch 4Ghz with Turbo on and this is not on all cores. Something is not right here...

Unless AMD is preparing to release an embarrassingly fail of a product, which will have an ironic moniker for years, something indeed doesn't add up. I wish AMD would end the quiet period, and just state how it performs.

[Dimitriman]

u right, if is it true, then :-(...AMD must know power of SB and Q1/Q2 netx year of IB. They need +- performance in single thread as Nehalem. It will be OK, still slower in ST, but very impressive in module architecture in multithread (MT). What is the logic for lower single thread performance in new generation architecture for next years than 2.5 years old Phenom II?

In my opinion, a ing bad choice of wanting to look strong on multithreaded server workloads and not particularly caring about single threaded performance, much less on desktop segment.

BD is a good server chip which will scale wonderfuly in multithreaded work but if it has 15% less ipc than thuban it will simply be garbage on desktop save for a few ocasions.

[maltrabob]

Unless AMD is preparing to release an embarrassingly fail of a product, which will have an ironic moniker for years, something indeed doesn't add up. I wish AMD would end the quiet period, and just state how it performs.

Something like AMD FX => Amazing Modern Design F-up eXtremely? Time will tell.