Will AMD's Bulldozer plow through Intel's Sandy Bridge?

rflcptr said:
Yeah, Intel essentially licenses the chipset with the server or workstation as its focus, leaving the actual "workstation," "server", or "gaming rig" branding up to the motherboard manufacturers' discretion.

Which is pretty much matched by Intel's hyper-threaded quads. Which leaves the likes of the 980x commanding such a premium.
Click to expand...

Exactly. OEM's can choose to use Core i7's which are known in the retail channels or "Xeon" branding which is only known to corporate and IT sectors.
=
 
One big thing AMD needs nail in servers is virtualization. Not only do they have to do it competitively from a price and performance standpoint, they also need to play a game of catch-up in regards to energy efficiency per metric of performance. Bulldozer's seeming focus (so far) on an abundance of physical execution resources over many cores may hurt that, though. On top of that, the platform will be an enticing alternative to Intel, to get OEMs to dedicate the resources and time for validation.
 
BD seems squarely focused on server performance, particularly the 16c Interlagos versions (since that's the only part even mentioned in the vaguely worded performance expectations by JF of AMD). However, most games and applications use nowhere near 8 threads. Draw your own conclusions about how it will work out on the desktop.
 
Well, any decent general-purpose OS has the thread scheduling facilities to affect better multitasking and performance overall. That should include recognizing and differentiating between idle or loaded physical or logical cores and scheduling threads appropriately, so the CPU can correctly turbo, downclock, or turn off cores as it sees fit. Obviously both companies' hardware has to do those things well. My thoughts on the mainstream desktop, so far, is that AMD's downward trend in marketshare will continue because AMD's approach with more physical cores hasn't done as well as Intel's approach that favors (at the same price) fewer but much more efficient cores. Since both companies with are moving forward with roughly the same respective strategies in that regard, I think my point holds. If AMD's desktop chipsets can meet or exceed Intel's feature set in quantity AND quality AND price AND [most importantly] demand, people may see past potential shortcomings in the CPU and AMD may have a bit of redemption. I'm not really holding my breath there, but as an aside, I wish Intel would enforce Intel gigabit LAN in the chipset spec or lower the licensing costs so more boards can get that throughput goodness. On the mobile front, Intel has substantially raised the bar in terms of performance and [the BIG DEAL] energy efficiency (mobile quad, ~7 hour discharge, and isn't a Macbook Pro?! :p), so AMD has an uphill fight ahead, for sure.
 
Last edited:
pxc said:
BD seems squarely focused on server performance, particularly the 16c Interlagos versions (since that's the only part even mentioned in the vaguely worded performance expectations by JF of AMD). However, most games and applications use nowhere near 8 threads. Draw your own conclusions about how it will work out on the desktop.
Click to expand...

Which is funny cause the client parts launch first, it's been there other way around for them for a while. haha.
 
Which is pretty much matched by Intel's hyper-threaded quads. Which leaves the likes of the 980x commanding such a premium.
Click to expand...

while hyper-threading is great it doesn't match the performance of actual cores. For the applications which are able to handle 6+ threads at a time the 6 core phenom's actually edge out the quad cores with 8 threads.

while hyper-threading does help quite a bit, it doesn't 100% make up for it.
 
ebduncan said:
while hyper-threading is great it doesn't match the performance of actual cores. For the applications which are able to handle 6+ threads at a time the 6 core phenom's actually edge out the quad cores with 8 threads.

while hyper-threading does help quite a bit, it doesn't 100% make up for it.
Click to expand...

In certain cases Hyperthreading actually hurts performance due to context switching (In specific workload scenarios it's actually preferred to disable it). On the server side AMD is still very competitive. Top CPU based supercomputer is still an AMD based installation (Dept. of Energy's Jaguar:Cray XT5-HE Opteron 6-core 2.6 GHz). AMD K10.5 performs better under Linux than it does under Windows. I wish review sites included some Linux benchmarks. Just check out the difference in Blender 2.5 under linux vs windows:

http://www.anandtech.com/show/2978/amd-s-12-core-magny-cours-opteron-6174-vs-intel-s-6-core-xeon/7

Tables get completely turned. 60% of servers outhere run Linux (conservative estimate), AMD is still very competitive in that sector.

I have a feeling AMD won't have any issue on the server side of things with Bulldozer. Desktop could be a different story.
 
Last edited:
ebduncan said:
while hyper-threading is great it doesn't match the performance of actual cores. For the applications which are able to handle 6+ threads at a time the 6 core phenom's actually edge out the quad cores with 8 threads.

while hyper-threading does help quite a bit, it doesn't 100% make up for it.
Click to expand...
No doubt, a physical core will edge a virtual one. :)
 
sirmo said:
In certain cases Hyperthreading actually hurts performance due to context switching (In specific workload scenarios it's actually preferred to disable it). On the server side AMD is still very competitive. Top CPU based supercomputer is still an AMD based installation (Dept. of Energy's Jaguar:Cray XT5-HE Opteron 6-core 2.6 GHz). AMD K10.5 performs better under Linux than it does under Windows. I wish review sites included some Linux benchmarks. Just check out the difference in Blender 2.5 under linux vs windows:

http://www.anandtech.com/show/2978/amd-s-12-core-magny-cours-opteron-6174-vs-intel-s-6-core-xeon/7

Tables get completely turned. 60% of servers outhere run Linux (conservative estimate), AMD is still very competitive in that sector.

I have a feeling AMD won't have any issue on the server side of things with Bulldozer. Desktop could be a different story.
Click to expand...
I can't recall the exact article on AnandTech, but one indicated that it took a 4 socket AMD to match an Intel 2 socket machine in virtualization capability. Then that power consumption issue (unfriendly to the datacenter) kicks in.
 
rflcptr said:
I can't recall the exact article on AnandTech, but one indicated that it took a 4 socket AMD to match an Intel 2 socket machine in virtualization capability. Then that power consumption issue (unfriendly to the datacenter) kicks in.
Click to expand...

Most VPS/Cloud providers run Linux or some form of Unix. Anandtech mainly benchmarks under Windows. I am not a big fan of Phoronix but at least they benchmark everything under Linux and the numbers are completely different usually in favor of AMD. I don't think desktop CPUs need to be benchmarked under Linux but server CPU benchmarks should.
 
rflcptr said:
One big thing AMD needs nail in servers is virtualization. Not only do they have to do it competitively from a price and performance standpoint, they also need to play a game of catch-up in regards to energy efficiency per metric of performance. Bulldozer's seeming focus (so far) on an abundance of physical execution resources over many cores may hurt that, though. On top of that, the platform will be an enticing alternative to Intel, to get OEMs to dedicate the resources and time for validation.
Click to expand...

We actually perform better in virtualization.

In most virtualized environments, customers tend to use "1 VM per core" as the rule of thumb. That means in a typical 2P world, we can get twice the VM density per server (24 VMs vs. Intel's 12 VMs).

In addition, we have 4 memory channels and larger memory footprints for virtualization (most of it is done on 2P 2U servers.)

Also, when you compare the cost per VM from a hardware perspective you see that AMD is in a much better position.

Intel does well in things like the VMmark benchmark, but that is not a realistic benchmark because it shows ~100VMs running on a 2P platform. That never happens in real life. Typically customers are putting 5-10 VMs on a typical box and 10-20VMs if you are doing large-scale consolidation projects.

From a power point we are very similar. Intel shows better in spec_power benchmarks because they have really low idle power (bulldozer will have some things that close the gap). But how often are virtualization servers sitting idle? Rarely, if ever., If you take out the idle power measurements, we are almost the same.
 
JF-AMD said:
We actually perform better in virtualization.
Click to expand...
Let's dissect the claim.
Intel does well in things like the VMmark benchmark, but that is not a realistic benchmark because it shows ~100VMs running on a 2P platform. That never happens in real life. Typically customers are putting 5-10 VMs on a typical box and 10-20VMs if you are doing large-scale consolidation projects.
Click to expand...
By your own admission, Intel is capable of running much more than 1 VM per physical core.
In most virtualized environments, customers tend to use "1 VM per core" as the rule of thumb. That means in a typical 2P world, we can get twice the VM density per server (24 VMs vs. Intel's 12 VMs).
Click to expand...
While 100 VMs may or may not be realistic, your assertion of comparative VM density is incorrect. What some customers do with the machine and what it's capable of can be wildly different.
In addition, we have 4 memory channels and larger memory footprints for virtualization (most of it is done on 2P 2U servers.)
Click to expand...
In a purely 2P world (assuming we ignore quad Beckton and quad G34), we've got dual 1366 against dual G34. Last I checked that should be 18 and 24 available DIMMs or 144 gigs and 192 gigs (@ 8 gigs per DIMM), respectively. Going by your greatest "large-scale consolidation" figure of 20 VMs, that leaves 7.2 gigs and 9.6 gigs per VM (ignoring the host OS). How often are these VMs pegging the lower bound of 7.2 gigs? I wouldn't virtualize a medium to large-scale database, so it's out of the question.
Also, when you compare the cost per VM from a hardware perspective you see that AMD is in a much better position.
Click to expand...
Assuming [incorrectly] that the only metric is of how many physical cores the machine has.
From a power point we are very similar. Intel shows better in spec_power benchmarks because they have really low idle power (bulldozer will have some things that close the gap). But how often are virtualization servers sitting idle? Rarely, if ever., If you take out the idle power measurements, we are almost the same.
Click to expand...
Close the gap on Nehalem ... or Sandy Bridge? But by the time Bulldozer arrives, AMD will be competing against Sandy Bridge servers.
 
Last edited:
rflcptr said:
Close the gap on Nehalem ... or Sandy Bridge? But by the time Bulldozer arrives, AMD will be competing against Sandy Bridge servers.
Click to expand...

hmmmm, you believe the 2P Sandybridge EP launches at or before BD?
 
JF-AMD said:
hmmmm, you believe the 2P Sandybridge EP launches at or before BD?
Click to expand...

What I've seen so far points to Intel Q3 and AMD "tentative" Q2 launches. There's still the very real possibility of a late Q2 / early Q3 AMD launch. Even if AMD keeps an early Q2 schedule, we're talking about Sandy Bridge following very quickly.
 
From what I can remember, BD server launch is one quarter after the client launch, which would put it in Q3 2011. AFAIK, LGA 2011, which is intel's higher performance server socket (I think..) should launch around the same time. However, in terms of deployment costs, BD can drop in to current systems whereas Sandy Bridge cannot. This is of course, for the server side. On the client side, BD is supposed to be Q2, but the higher end Sandy Bridge is still Q3 or Q4
 
Last edited:
rflcptr said:
In a purely 2P world (assuming we ignore quad Beckton and quad G34), we've got dual 1366 against dual G34. Last I checked that should be 18 and 24 available DIMMs or 144 gigs and 192 gigs (@ 8 gigs per DIMM), respectively. Going by your greatest "large-scale consolidation" figure of 20 VMs, that leaves 7.2 gigs and 9.6 gigs per VM (ignoring the host OS). How often are these VMs pegging the lower bound of 7.2 gigs? I wouldn't virtualize a medium to large-scale database, so it's out of the question.
Click to expand...

VPS providers and even Amazon S3 prefer smaller "slices". Anywhere from 256Mb to 2Gb. AMD with more true cores absolutely has the edge there. VPS and Cloud is a pretty big market. Bulldozer is most likely going to increase this edge.
 
kirtar said:
From what I can remember, BD server launch is one quarter after the client launch, which would put it in Q3 2011. AFAIK, LGA 2011, which is intel's higher performance server socket (I think..) should launch around the same time. However, in terms of deployment costs, BD can drop in to current systems whereas Sandy Bridge cannot. This is of course, for the server side. On the client side, BD is supposed to be Q2, but the higher end Sandy Bridge is still Q3 or Q4
Click to expand...

I thought Bulldozer was going to require socket AM3+, and not be compatible with any current AMD boards?
 
rflcptr said:
What I've seen so far points to Intel Q3 and AMD "tentative" Q2 launches. There's still the very real possibility of a late Q2 / early Q3 AMD launch. Even if AMD keeps an early Q2 schedule, we're talking about Sandy Bridge following very quickly.
Click to expand...

We'll be in production in Q2 and launching in Q3. Too bad I am not allowed to bet with you, I believe the "very" part of your statement will not be accurate. I guess we'll see in Q3.
 
Well, today officially marks the day that AMD has been lapped twice. First by Nehalem and now by Sandy Bridge. And AM3+ BD still has no release date.
 
If you look at AMD's share price history. There has been two massive peaks (when the share prices have rocketed above $40). This happened once in April 2000 and again in Jan 2006.

The patten seems to be repeating again for 2011..
 
I tripled my investment on both of those peaks. I will buy in again probably for 1000 to 2000 shares if the price hits around $6. I am expecting the price to decline in Q1 after AMD posts a loss.
 
If partners are planning on releasing 900 series boards before Bulldozer it would be nice to see a good performance preview by Q2 at the latest with boards going on sale simultaneously. It would potentially give users incentive to hang on to their AM3 CPU's and not jump ship to Sandy Bridge. I'm not gonna lie Sandy Bridge is excellent and Q3 is a long way away with performance as good as SB for as cheap as it is.

We've known the performance of every major Intel arch months before launch since Conroe. I think Anandtech's Sandy Bridge preview was done in September or November but I could be wrong and it gave a pretty comprehensive picture of performance.

AMD needs something like this. I'm gonna be hanging onto my 1090t/800 series setup with hopes of picking up a 990FX board in the near future and eventually bulldozer. If performance is not where it needs to be LGA 2011 here I come.

Bulldozer needs to trump Sandy Bridge and potentially Ivy Bridge. If it can make itself comfy between the two great, but if it's slower than SB it'll be a sad day.
 
drescherjm said:
I tripled my investment on both of those peaks. I will buy in again probably for 1000 to 2000 shares if the price hits around $6. I am expecting the price to decline in Q1 after AMD posts a loss.
Click to expand...

And it will go lower before "their launch" in Q3. Why? Because by the end of Q2, Intel will have a "foothold" on the mainstream market with SB. What do I mean. Well, its Tax Time! Between January-May, refunds a plenty are gonna go SB, specially at the cheaper prices. People and lots of them, have been sitting on C2Q and C2D waiting for this type of upgrade. Why put off for a possible 10% better or worse performance for 6 months??? (Really, do you think AMD is gonna best SB by more than that? Or maybe less?) Even if BD beats SB for lauch in July, whats around the corner? LGA 2011. AMD missed a huge Holiday season(i3,i5, and i7 taking this market) and now tax refund time. Intel made/will make a killing. I can't see people in the nice summer months really spending much in the way of comp gear.

The Athlon64 beating the snot out of P4 days is over and that was Intel's plan with the whole TICK/TOCK program. AMD will no doubt make their money in the server platform. But as you and I as consumers, nope. I am not a fanboy by no means as I have had AMD from K2-300, Athlon 1000mhz, to todays X2 250 Regor in daughter's comp. I have also ran multiple Core 2 processors from mine to my nephews. Amd is a good low budget processor for home use AGAIN. (Like they were from the 486 days to the pentium 1-2 days.) Also to note is Intel's production/resources are double of what AMD is. This isn't the tech stock boom of 2000. (Where I also cashed in on AMD, but since have not seen a reason to invest.) 6 months behind you better come out looking like a Shinobi Master, not a Karate instructor. Maybe AMD will stay on low budget builds and make its money there.

As Cramer would say "NOT A GOOD BUY." ;)
 
SnowBeast said:
And it will go lower before "their launch" in Q3.
Click to expand...
Possibly, but not for the same reasons. Product introductions are always good for a bump, but AMD continues to shed server market share and margins are continuing to collapse after a brief run of "good" results (propped up by one time sales of equipment and licensing royalties). There will likely be at least 2 intervening quarterly results that could affect the price.

AMD's stock is pretty unpredictable, but AMD is in a very different position now than it was in April 2000, notably significantly more debt and recovering against much stronger competition. :p
 
As Cramer would say "NOT A GOOD BUY."
Click to expand...

I would expect him to say:

SELL, SELL, SELL!

But I do not follow Cramer all the time. AMD is and has been a high risk stock for as long as I have watched them. However it has yielded me the best return rate of any stock I have owned.
 
Last edited:
The only figure AMD is confirming is the year 2011. But there are rumours of it being April.

The is one thing that I am surprised about is that nobody has reported seeing 'closed doors' bulldozers viewing during CES. Which is strange, if they really are going to have these out [in volume] by April, there should be some leaks by now...
 
jbz7890 said:
Any idea when the bulldozer cpus will come out?
Click to expand...
There are only vague calender periods so far, despite how close the date supposedly is. AMD's official word is "Q2" for the desktop and "Q3" for the server version.

It wasn't shown or even demoed at CES like the Llano was, which is also supposed to launch in Q2. I wouldn't hold my breath for BD until a date is more firm.
 
pxc said:
There are only vague calender periods so far, despite how close the date supposedly is. AMD's official word is "Q2" for the desktop and "Q3" for the server version.

It wasn't shown or even demoed at CES like the Llano was, which is also supposed to launch in Q2. I wouldn't hold my breath for BD until a date is more firm.
Click to expand...

I thought Llano was pushed back to Q3? I think it should be a pretty interesting part.
 
TheBlueChanell said:
I thought Llano was pushed back to Q3? I think it should be a pretty interesting part.
Click to expand...
I'm just going by what AMD posts and Dirk Meyer says. I don't hear the voices in Charlie's head. :p

I like the idea of having a somewhat powerful OCL/DC coprocessor on die. Not really interested in the graphics functionality, but it should be good enough for many buyers. I mean, look at the even slower stuff Intel sells and it has like 1/2 of the total graphics market.
 
pxc said:
I mean, look at the even slower stuff Intel sells and it has like 1/2 of the total graphics market.
Click to expand...

Intel is very similar to Microsoft in the corporate world. No one ever got fired for buying Microsoft and same can be said for Intel.

Combine this with all the non-enthusiast end users who buy Intel because that's all they know, it's easy to see why Intel has such a high market share of their inferior on board GPU solutions.
 
Feels like a weird time for the CEO to leave. Fusion, HD6000 series just came out, and even a 10% clock for clock efficiency boost with BD would be a win because of cheap per integer core cost. Desktop fusion will also debut late this year. The future looks brighter than at any point in recent AMD history.

I wonder if he has a personal reason? I hope it's not politics, AMD isn't that well off.
 
Dan_D said:
If you ever want to know how a ship is doing, watch the rats. :D When they jump ship it's usually best to follow. :cool:
Click to expand...

Well, I'm looking at SB S1195 and maybe also S2011 as AM3+ looks like garbage to me and that's why I'm now looking at Intel as AMD's line is far to consumer orientated for my liking.

When they had S754 (single channel) and S939 (dual channel) that was great!:D

AM2/+ not so great, but AM3 = crap.:mad:
AM3+ (dual channel) and another socket with quad channel would be a better solution.:cool:

AMD's socket 1207 could be used for BD FX dual channel (2* 2133+) and quad channel (4* 1333MHz).;)
 
Last edited:
You must log in or register to reply here.
Back
Top