Friday, January 20th 2012

AMD Vishera Packs Quad-Channel DDR3 IMC, G34 En Route Desktop?

AMD might be a little sore that its "Zambezi" FX processor family based on its much-hyped "Bulldozer" architecture didn't quite meet the performance expectations of a ground-up new CPU architecture, but it doesn't want to take chances and build hype around the architecture that succeeds it. From various sources, some faintly-reliable, we have been hearing that the next-generation of high-performance desktop processors based on "Piledriver" architecture, codenamed "Vishera", will pack five modules or 10 cores, and will be structured essentially like Zambezi, since Piledriver is basically a refinement of Bulldozer architecture. The latest leak comes from the Software Optimization Guide for AMD 15h family (read here), which was picked up by CPU World while most of us were busy with CES.

CPU World compiled most of the features of what it suspected to be AMD referring to its future processors based on the Piledriver architecture, that's "Vishera" (desktop high-performance), "Terramar" (high-density server), and "Sepang" (small-medium business server) parts. The three are not the first chips to be based on Piledriver, AMD has a new mainstream desktop and notebook APU in the works codenamed "Trinity", which is en route for a little later this year. Trinity basically has an identical CPUID instruction-set as Vishera, Terramar, and Sepang, confirming their common lineage compared to today's "Bulldozer" architecture. The most catchy detail is of Vishera featuring 4 DDR3 channels.

The plot thickens where "HyperTransport Assist feature" is listed as being supported on Vishera. HT Assist is a feature found on AMD's enterprise socket G34 processors, which facilitates better inter-die communication between the two dies of a typical socket G34 Opteron processor. The G34 (LGA1972) package is a multi-chip module of two quad-core, six-core, or four-module dies, which combined have four DDR3 memory channels, and a number of HyperTransport links to communicate with neighbouring sockets and the system's chipset. Could this be the first indication that AMD wants to take on Intel LGA2011 HEDT (high-end desktop) using Vishera chips in the G34 package? It will be a while before we find out.

Apart from using common silicon between client and enterprise platforms, AMD does have a history of colliding the two.Source: CPU World
Add your own comment

229 Comments on AMD Vishera Packs Quad-Channel DDR3 IMC, G34 En Route Desktop?

#2
Aquinus
Resident Wat-man
by: xenocide
Nowhere in that article does it say 20%.
I'm going to say the same thing I've been saying about everything said about Kepler.
"Got a source?"
Posted on Reply
#3
Inceptor
by: nt300
am taking 20% based on the upcoming Trinitys base Piledrier CPUs.
I think that he was just making a guess based on his own thinking, rather than citing concrete information about Vishera's performance that he read somewhere.
An inference, if you will, from the as yet unconfirmed performance boost that Trinity has over Llano, and then applying some special-opinion sauce to that to speculate on the performance of the Vishera core optimizations.

There, that sounds better.
Posted on Reply
#4
Super XP
Wow, 20% in a nice rounded number. It does say somewhere Trinity should be about 20% performance via CPU over LIano.
Posted on Reply
#5
Bvanofferen
What matters most AMD vs Intel

So Clock 4 Clock performance is crazy to consider when purchasing a PC. Cores, Cache and Ghz for the $ it costs. If you over clock then watts and voltage, unlocked or locked. Memory controller speed and channels all count too. Core architecture performance is almost always quickly optimized by software. Instruction sets get utilized and better drivers get published. Last but not least PLATFORM! I think the main let down with bulldozer is that AMD fans were expecting Zambezi to thoroughly out perform the 2600k. AMD showed a couple of tests with the 7970's that gave better frame rates then sandybridge and handbrake is better on FX. But the 2600k out performs FX quite a bit in soo many areas. SOOOOO WHAT. It costs more!!! And The AM3+ Platform has way more potential then 1155. And 1155 costs more. Upgradability, price for performance and mufti-threaded architecture way beyond sandybridge. That's why they code named it "BULLDOZER". FX can park 7 cores. Power consumption??? Is out performing sandybridge with another $1000 chip really the only thing that matters? For less then 1/10 of 1% of pc's it does. That's right less then 1/10 of 1% of pc's run more then the 8150 can currently process. We don't need another $1000 chip. We need $200 chips on a platform that can give us 10 good years of pc power. Spend $2000, or so, on a PC every 3-4 years with intel to stay on top or spend $2000 once and a couple hundred every 3-4 years to stay on top with AMD. 1155 Sandybridge chips will be far behind am3+ chips in two years. Guess Intel boys will have to use it as a coaster and take out another mortgage to upgrade. American consumerism is way better off with AMD. And so is my rig :) PS my FX-6100 rig gets 8900 on passmark at 5GHZ with a $20 CAFA70 cooler. I paid $250 for FX-6100 and TA990fxe. I wonder what a 10 core pile driver chip will get in my rig in a few years. 15000 seems reasonable.
Posted on Reply
#6
Patriot
by: Bvanofferen
So Clock 4 Clock performance is crazy to consider when purchasing a PC. Cores, Cache and Ghz for the $ it costs. If you over clock then watts and voltage, unlocked or locked. Memory controller speed and channels all count too ....
fanboi drool... unsuported ... wishful thinking... misguided logic ... combining server and desktop roadmaps ... I paid $250 for FX-6100 and TA990fxe. I wonder what a 10 core pile driver chip will get in my rig in a few years. 15000 seems reasonable.
I think you should consider decaf ...
Either troll or hapless fanboi...

Anyhow, I hope AMD does pull this one out... FX are not bad chips, just fall pathetically short of the hype. For many applications you will not notice if your running on them or not.

But the fact is currently a 2500k will beat an fx rig in cost and in performance.
IB doesn't look so promising so AMD may be able to narrow the gap here.

I am seriously wondering why people pull this 10core number out of their ass...
On desktop Piledriver is 8 not 10... on server it stays the same as IL 4,8,12,16 core variants... though who knows they may trim it down a bit.

That said I have an i5 laptop and a 2p MC for my daily drivers and a 4p MC for folding (find me a machine faster than harbringer for folding).
AMD has work to do, but they will survive. SB-E is not that great power/perf and IB is looking to be a poor oc...
Posted on Reply
#7
Vulpesveritas
OOOKay sooo my main issues here are the following:
1. The spec's listed are Pre-Vishera, the 10 core was a "komodo" FM2 socket designation, before AMD's new CEO got his position, and canceled Komodo in favor of an 8 core AM3+ socket compatible design. And hence dual-channel RAM.
http://www.anandtech.com/show/5491/amds-2012-2013-client-cpugpuapu-roadmap-revealed
2. Phenom II IPC is only 20% behind sandy Bridge. As Trinity appears to have the same IPC as Phenom II, despite lacking L3 cache, it can be probable to assume that Piledriver will perform to near sandy bridge IPC, while clocking much higher.
3. This says nothing of the Resonant clock mesh AMD has already stated will be present on Piledriver, increasing power efficiency/clock.
4. Outdated information is outdated.
Posted on Reply
#8
Melvis
All i want is a (i dont care how many cores it has) AMD CPU to perform around the same as a 2600K (like Bulldozer was meant to) then id be very happy, that to much to ask for?
Posted on Reply
#9
xenocide
by: Vulpesveritas

2. Phenom II IPC is only 20% behind sandy Bridge. As Trinity appears to have the same IPC as Phenom II, despite lacking L3 cache, it can be probable to assume that Piledriver will perform to near sandy bridge IPC, while clocking much higher.
The IPC won't be the same, it will still be pretty far behind Sandy Bridge. In this topic it is revealed Piledriver will have about the same IntGHz as Stars. This might put it closer to SB, but definitely won't be at the same level. According to that chart SB's FP/GHz is 25% higher and Int/GHz is also 25%+ higher than PD-based Trinity. This is just like Pentium 4, AMD can't make the architecture a ton better, so they are cranking the clock rates as high as they can go.

by: Vulpesveritas
3. This says nothing of the Resonant clock mesh AMD has already stated will be present on Piledriver, increasing power efficiency/clock.
According to most articles on the subject RCM will only really allow for higher clocks with a lower power draw, it doesn't increase performance by itself, just performance per watt. It's part of the reason PD is going to be clocked so high.
Posted on Reply
#10
Bvanofferen
You still can't rule out a 10 core piledriver/steamroller am3+ or am4 (quad channel) chip that works on am3+. So outdated is jumping the gun, most likely delayed to keep am3+ the performance platform rather than FM2. An FX-8120 vs i5-2500k is proof of price performance in AMD favor. Anyone care to comment on the upgradable platform advantage AMD has over Intel? Or are we going to talk about coffee. Bulldozer wasn't meant to outperform the 2600k in every app. New architecture rarely outperforms high end current architecture. But in sub $200 chips FX has a big advantage over Intel. And the 8150 runs right where it should between 2500k and 2600k unless you do video conversion then FX has the advantage.
Posted on Reply
#11
Vulpesveritas
by: xenocide
The IPC won't be the same, it will still be pretty far behind Sandy Bridge. In this topic it is revealed Piledriver will have about the same IntGHz as Stars. This might put it closer to SB, but definitely won't be at the same level. According to that chart SB's FP/GHz is 25% higher and Int/GHz is also 25%+ higher than PD-based Trinity. This is just like Pentium 4, AMD can't make the architecture a ton better, so they are cranking the clock rates as high as they can go.



According to most articles on the subject RCM will only really allow for higher clocks with a lower power draw, it doesn't increase performance by itself, just performance per watt. It's part of the reason PD is going to be clocked so high.
Okay well, 1. Integer performance is, at least to my knowledge, what matters most of the time in a CPU. With integer performance being the same as STARS without L3, then it may be 10-20% faster clock for clock when it is added( note- i said as fast or almost, i.e. slightly slower clock for clock but not much) , and with stock clocks being 20-30% higher at stock vs SB/IB, it should compete nicely.
And yes that RCM makes it clock higher/watt. It may mean it becomes a high end OC chip, if it can clock into the upper 5ghz range on water, and will give AMD an overall fully competitive mainstream chip. Also, if I remember correctly part of the turn to BD architecture was because PII couldn't be improved much more, and they wanted better power efficiency.
1st gen failed at that, though I'm hoping Piledriver gives them a competitive edge, and means we can recommend AMD again, and the fanboy war debates with no good answer one way or the other can commence.
Posted on Reply
#12
xenocide
by: Vulpesveritas
Okay well, 1. Integer performance is, at least to my knowledge, what matters most of the time in a CPU. With integer performance being the same as STARS without L3, then it may be 10-20% faster clock for clock when it is added( note- i said as fast or almost, i.e. slightly slower clock for clock but not much) , and with stock clocks being 20-30% higher at stock vs SB/IB, it should compete nicely.
And yes that RCM makes it clock higher/watt. It may mean it becomes a high end OC chip, if it can clock into the upper 5ghz range on water, and will give AMD an overall fully competitive mainstream chip. Also, if I remember correctly part of the turn to BD architecture was because PII couldn't be improved much more, and they wanted better power efficiency.
1st gen failed at that, though I'm hoping Piledriver gives them a competitive edge, and means we can recommend AMD again, and the fanboy war debates with no good answer one way or the other can commence.
Stars didn't have L3 Cache either I believe. The addition of said Cache will only give a modest 3-5% performance boost anyway, definitely NOT 10-20%. I think Clock for Clock SB will still be better. Piledriver--and Vishera by extension--is basically shaping up to have the performance of Phenom II (when it comes to per Coreper thread) but a much better IMC and hopefully better power consumption this time around.

I have no doubt they will reach 5GHz on Water, probably be able to do that on Air if the stock clocks are so high. I just don't see Piledriver shaping up to be better than or even on par for Sandy Bridge, let alone Ivy Bridge (which is something like 8% faster in most applications while overclocking a little less and generating slightly more heat). The big issue is AMD really needs to figure out how to get their "Tock's" on the shelves correctly. It seems like the first version of the last couple CPU's AMD has made (Phenom I, Bulldozer) have been riddled with problems, but they usually have a decent second round. If they want to be competative their first go has to work as well as Intel (Sandy Bridge etc.).
Posted on Reply
#13
Bvanofferen
A thought I have, that comments would be much appreciated on, is latency in FX Cache being looser then Phenom and sandybridge. Is this causing the low IPC and clock per clock performance ratio? The way I understand latency is that to achieve the best throughput on ram looser latency can achieve much higher Clock speeds and achieve better overall results but the clock per clock results go down. This idea occurred to me because the results I'm having with my fx-6100 is 6000-6400 Physics score @ 4.6-5.0ghz with corsair CAFA70 on 3dmark11 with a single 6870. This is better overall result then Thurban and the 2500k would need a much better cooler (water) to substantially out perform my results. I'm pretty sure the 2500k would only hit 4.2-4.4 with the same cooler. So if I'm right FX IPC is less yet better because of the higher overall results when overclocking on air. Comments?
Posted on Reply
#14
eidairaman1
by: Bvanofferen
A thought I have, that comments would be much appreciated on, is latency in FX Cache being looser then Phenom and sandybridge. Is this causing the low IPC and clock per clock performance ratio? The way I understand latency is that to achieve the best throughput on ram looser latency can achieve much higher Clock speeds and achieve better overall results but the clock per clock results go down. This idea occurred to me because the results I'm having with my fx-6100 is 6000-6400 Physics score @ 4.6-5.0ghz with corsair CAFA70 on 3dmark11 with a single 6870. This is better overall result then Thurban and the 2500k would need a much better cooler (water) to substantially out perform my results. I'm pretty sure the 2500k would only hit 4.2-4.4 with the same cooler. So if I'm right FX IPC is less yet better because of the higher overall results when overclocking on air. Comments?
FX requires such a high clock speed to achieve what a Core i or Phenom II can with lower clocks, because Single Thread IPC is higher on the phenom II and core i. Only time a FX will pull ahead is in heavily threaded tasks but thats primarily in servers
Posted on Reply
#15
Bvanofferen
Is having an FX chip that can overclock to higher levels with low ipc, just as good or even better when the clock level outweighs the low ipc? It seems that the FX can pull ahead on air unless your only using 4 threads at all times. When i game (dragon age II) All six cores light up.
Posted on Reply
#16
eidairaman1
by: Bvanofferen
Is having an FX chip that can overclock to higher levels with low ipc, just as good or even better when the clock level outweighs the low ipc? It seems that the FX can pull ahead on air unless your only using 4 threads at all times. When i game (dragon age II) All six cores light up.
not if they get hotter than the other chips, this chip is pretty much like what Prescott was to Intel, it Clocked High but was hot.

IPC rules over clock speed
Posted on Reply
#17
xenocide
by: eidairaman1
not if they get hotter than the other chips, this chip is pretty much like what Prescott was to Intel, it Clocked High but was hot.

IPC rules over clock speed
As well as causes massive power consumption which puts a lot more strain on the system. I think you shouldn't need to overclock the crap out of a CPU for it to be competative personally.
Posted on Reply
#18
Aquinus
Resident Wat-man
by: eidairaman1
because Single Thread IPC is higher
IPC = Instructions per clock, if Bulldozer is slower at the same clock speed then the IPC would be lower not higher.
Posted on Reply
#19
eidairaman1
by: xenocide
As well as causes massive power consumption which puts a lot more strain on the system. I think you shouldn't need to overclock the crap out of a CPU for it to be competative personally.
i was trying to explain that to the other guy.

Both Phenom 2 and Core i architectures are superior than the bulldozer arch, Overclocking is only a plaything.

by: Aquinus
IPC = Instructions per clock, if Bulldozer is slower at the same clock speed then the IPC would be lower not higher.
I did say IPC on the Phenom II and Core i are higher than bulldozer hence the abysmal performance numbers tested even here.

Re-read the post.
Posted on Reply
#20
Aquinus
Resident Wat-man
by: eidairaman1
Re-read the post.
Read that out of order, my bad.
Posted on Reply
#21
eidairaman1
by: Aquinus
Read that out of order, my bad.
Ok thanks, and for the poster that was askin about bulldozer and anyone else who wants to know

Intel CPUs:

Clock Speed Based (include 1st Gen Netburst) (Low IPC)

8086
8088
286-486DX
Pentium 1
Pentium 4 (P4 Based Xeon, Celeron)
Pentium D

P6 Based (Include Core) (High IPC)
Pentium Pro
Pentium 2 (Celeron)
Pentium 3 (P3 Based Xeon, Celeron)
Core
Core 2
Core i (P6 with 2nd Gen Netburst)

unknown Intel Part
Ivy Bridge-E


AMD CPUs:

Clock Speed Based

8080
D8086
Am286-Am486 (Slighty higher IPC than Intel parts)
K10 Phenom 1 (Including Phenom 1 Based Opterons, Athlon, Sempron)
Bulldozer (Including Bulldozer based Opterons)

High IPC Parts

Am586
K6-K6III
K7 (Geode, Duron, Sempron, Athlon, Athlon XP)
K8 (Athlon 64, Athlon 64 FX, Athlon 64 X2, Opteron, Sempron)
K10.5 (Phenom II, Opteron, Athlon II, Sempron)

Unknown AMD Parts
Piledriver
Opteron 3200
Posted on Reply
#22
Bvanofferen
Thanks for the feedback everybody. Being able to overclock the crap out of a chip shouldn't be it's main feature and performance competitive edge. Which is also what I've found with my fx-6100. It seems this is what AMD is doing though i.e. Pairing a 8150 at normal speed with a h100 style water cooler. It's like a billboard saying "I need to be blue screened until every last mhz is found!"
Posted on Reply
#23
eidairaman1
by: Bvanofferen
Thanks for the feedback everybody. Being able to overclock the crap out of a chip shouldn't be it's main feature and performance competitive edge. Which is also what I've found with my fx-6100. It seems this is what AMD is doing though i.e. Pairing a 8150 at normal speed with a h100 style water cooler. It's like a billboard saying "I need to be blue screened until every last mhz is found!"
thats what marketing touted it as. It runs for what it does. I mean honestly are you happy with it?
Posted on Reply
#24
Aquinus
Resident Wat-man
by: eidairaman1
I mean honestly are you happy with it?
I would be happy with it in a server, it also won't break the bank if you get it for a server too, in comparison to 1,300+ USD for an 8-core Xeon chip.
Posted on Reply
#25
eidairaman1
by: Aquinus
I would be happy with it in a server, it also won't break the bank if you get it for a server too, in comparison to 1,300 USD+ 8-core Xeon chips.
thats what they were derived from of course
Posted on Reply
Add your own comment