Bulldozer vs Nehalem

[wickedmonster]

Phenom has lost to C2D in this round but AMD's next gen Bulldozer is waiting in the wings. Upon it's release (2009-10), it'll probably go up against Nehalem. Details about both chips are scant, but Nehalem will probably be a better performer than Penryn which means Bulldozer needs to be better than K10 by at least 20-30% to match Penryn's IPC. Things are looking tough for the green team but they were in dire straits before when the P4's slaughtered the Athlon XPs.

What do you guys think AMD needs to do to make Bulldozer perform competitively with Nehalem/Penryn? Better L2 cache? MCM? Let's give AMD suggestions to make them better!

 

[nruo]

you don't have a clue about Nehalem performance (in 2008), let alone 2010 processor. So now AMD can start again with the "just wait for xxx which will bit Intel". lets first see the next generation - nehalem vs AMD 45NM beore you jump to pure speculations (which seems to be the only feild AMD can win)

 

[Ycon]

AMD cant compete at new vs. old core, so how could they compete at old vs. new architecture?

 

[epsilon84]

It's a bit too soon for anybody to be predicting the performance of Nehalem or Bulldozer.

Besides, I hope AMD survives long enough to actually release Bulldozer...

 

[Slobogob]

This is fiction. And given that AMD has yet to launch a processor on time, i highly doubt we will see "Bulldozer" this decade.

 

[jkflipflop98]

I can tell you firsthand that Nehalem is a monster. It will make you crap your pants.

 

[pogsnet]

These things does not exist in labs yet so we cant tell yet.

Even AMD cannot take the crown of the fastest as of this time it doesnt mean it cannot take it in the future. Look at Pentium 4? its equivalent performance is almost only half the Athlons. But here comes Core2 to take the crown, though the Core1 is unsuccessful on performance. Phenom is still struggling now at few gaps over C2Quads. Maybe the Phenom2 or as others say Phenom FX can do the trick. lol

 

[Raviolissimo]

Nehalem or Bulldozer.

they're actually calling it Bulldozer ?

read up on an American woman named Rachel Corrie and
you'll see what i mean.

the Nehalem plant is in Israel; Rachel Corrie was killed
by a bulldozer operated by Israel, her offense being
to stand in front of the home of a Palestinian family.

 

[Iain1974]

The word 'bulldozer' has been around and in common useage since before the unfortunate Miss Corrie stood in front of a bulldozer.

 

[bornking]

This thread started 12 months too soon.

 

[frozenlead]

with amd's reputation lately, i think this thread started a little earlier than 12 months too early.
not picking on them or anything..but it's the truth.

 

[Grimmy]

Naaaa.. its not that its too early. Both company's prolly already have it locked away, and on perhaps some kind of time lock. I mean, what else are they going to do with all this time, especially when they have the fastest CPU to help them make a better product. :lol:

 

[WR]

What do you guys think AMD needs to do to make Bulldozer perform competitively with Nehalem/Penryn? Better L2 cache? MCM? Let's give AMD suggestions to make them better!

I'm no designer so this is just a wishlist. Compartmentalize the design - CPU-buffer logic, CPU-core, CPU-cache, GPU/vector-core, GPU-RAM (?), IMC, HT. For the CPU, design huge look-ahead buffers and robust decoding/analysis instead of so much generic read caching. Add raw execution engines to enable multithreading (I'm sure Intel went after this). Add redundant CPU logic units/subunits whenever possible and useful (based on past experience in failure rates; extra units would still be fused off if everything comes out working). Redesign and condense old core components for easier clock scaling and power efficiency. Work on HyperTransport/IMC logic to enable low-latency "MCM" if later called for. Remove separate voltage planes and instead focus on internal ability to turn components on/off.

While thinking through this, I noticed that they may want to put some design effort into how they "glue" components together. Placing an entire performance design on one slab of silicon and microconnecting all the parts seems to be a sure way to cause yields to tank. On the other hand, having every tiny component on separate silicon could cause a manufacturing and assembly nightmare and, depending on the gluing method, unnecessary latency.

Too bad it's too late to be designing like this in time for release in 1H'10.

 

[jimmysmitty]

These things does not exist in labs yet so we cant tell yet.

Even AMD cannot take the crown of the fastest as of this time it doesnt mean it cannot take it in the future. Look at Pentium 4? its equivalent performance is almost only half the Athlons. But here comes Core2 to take the crown, though the Core1 is unsuccessful on performance. Phenom is still struggling now at few gaps over C2Quads. Maybe the Phenom2 or as others say Phenom FX can do the trick. lol

Where do you get your info at? Intel has had Nehalem up and running at the first step 45nm chips for a while now with full plans to release in Q3/Q4 08. The P4 was stomping the Athalon until 2003 when they released the Athlon 64. The only way the Phenom FX will ever see the light of day is if AMD can fix the L3 mess up and fix the thermal problems they have above 2.4GHz since the only thing different for a FX is the higher frequency. And Phenom is not only struggling in the performance area vs C2Qs but als in power usage. My Q6600 still uses less than the 2.3 or 2.4GHz Phenom.

So lets say AMD can release Bulldozer in late 08 or earl 09. This means all of those who bought a Phenom and want the best performance out of their AMD would need to buy another chip in less than a year. I for one do not forsee AMD being able to release Bulldozer that soon since the Phenom may not sell as well as expected. Or at least not well enough to get them the cash needed to be able to produce them in a large enough volume.

 

[Shadow703793]

I cant wait until AMD releases this chip...(sarcasm) Hopefully Intel has some thing better. : lol :

FUD????

 

[ro3dog]

I thnk this thread is interesting.now, as far as I now (net news) Bulldozer will really be a more advance barcy/agena core,which means to me that it will be more mature with new sets of instructions working on a ddr3 motherboard vs.a brand new Intel proc.I;m def not like WR with some really serious knowledge of cpu.And for those who really think AMD will be that far behind, less heart(fanboy-ism) and more brain please

 

[WR]

brand new Intel proc

Those words would make me cautious. Last time, Netburst cost Intel a lot of headache. Fortunately, Nehalem seems to borrow from the highly successful Merom core; the brand new parts are the caching system and memory interface. As long as they don't ruin both, it should be solid.

Bulldozer will really be a more advance barcy/agena core

It seems as dramatic a shift as from Penryn to Nehalem. I hope it is, as AMD needs something like that to compete. I also hope they simulate and then test it - tired of their dumping half tested products on us lately (2900-series power consumption, K10 clock ceiling).

 

[BaronMatrix]

Phenom has lost to C2D in this round but AMD's next gen Bulldozer is waiting in the wings. Upon it's release (2009-10), it'll probably go up against Nehalem. Details about both chips are scant, but Nehalem will probably be a better performer than Penryn which means Bulldozer needs to be better than K10 by at least 20-30% to match Penryn's IPC. Things are looking tough for the green team but they were in dire straits before when the P4's slaughtered the Athlon XPs.

What do you guys think AMD needs to do to make Bulldozer perform competitively with Nehalem/Penryn? Better L2 cache? MCM? Let's give AMD suggestions to make them better!

BullDozer is a TOTAL REDESIGN @ 45nm. It will have SSE5 and probably much more IPC. Montreal will be MCM with more L3.

 

[teh_boxzor]

i think AMD will have the next round, why? because they already put out a native quad-core and are tweaking it to make it more efficient and faster, whereas intel is shooting for their first native quad core with Nehalem. Intel will probably run into problems like AMD is right now.

Where the hell do they come up with these names? do they just pick them out of a hat?

 

[monsterrocks]

I think the ramp from 65nm to 45nm will be much kinder and easier on AMD than that of the 90nm to 65nm. So I think Bulldozer is going to be an amazing processor. The only problem is that I think Nehalem will be a little better; mainly because AMD saw large performance gains whilst putting an inbuilt memory controller. So I think that intel, who already has great core architecture, is gonna see some great gains from the inbuilt memory controller. Of course, we know almost nothing about these two processors, so until then, I am just going to have to pray that AMD can get enough cash to help their R&D team so they can be a competitor again.

 

[pausert20]

We do know that Intel already has them in validation in their labs. They showed an A0 stepping part at the IDF in San Francisco. I hope that AMD can get better performance and yield out of their transition from 65nm to 45nm. But currently if we extrapolate what has occurred from the 90nm to 65nm I only see worse yields and increased thermal issues do to leakage. They need to come up with their own version of High K. But from everything that has so far come out they will only be using 45nm immersion with Low K. That means they will not see the advantages that Intel has with their High K metal connect process.

 

[ro3dog]

AMD and IBM along with another company are jointly working on high K.AMD is far ahead of what people are giving them credit for,no one know for sure if Phenom is a bust for them to say its no good .The native core on 65nm and the merge did take a toll on AMD,but now imagine how far they will be when Intel goes native core.Native core is here to stay and AMD is well on there way.Like they say "What does'nt kill you makes you stronger

 

[pausert20]

Ro3dog, Sorry I think you are being overly optimistic on AMD's behalf and overly pessimistic on Intel making a Quad core monster which is what I believe the Nehalem is going to be. If they got it working enough to have Windows operating on it showing 16 cores in a 2 processor system in the first release of silicon (A0) stepping and don't plan to release the part until the end of next year. They will have it working great by then.

 

[ro3dog]

I believe AMD too showed native cores working and everyone was saying how AMD will make this strong come back and distroy c2d but it was not the case(atleast right now).Pessimistic about Intel,no,they will still have strong products as buffer till Nehalem is matured.It will a great time for consumers,for the compition between AMD and Intel will be big.Just cannot count them out

 

[WR]

BullDozer is a TOTAL REDESIGN @ 45nm. It will have SSE5 and probably much more IPC.

Throwing those words around like it's simple to redesign and increase IPC.

While it wasn't even a complete redesign, look where K10 stands. Hardly noticeable real-world IPC improvements; good multicore scaling as with K8, but newfound single-thread troubles and lackluster frequency headroom.

Netburst was a complete redesign; look where that put Intel. Complete redesigning is very hard. History has favored conservative refinement because you only have so much R&D time.

i think AMD will have the next round, why? because they already put out a native quad-core and are tweaking it to make it more efficient and faster, whereas intel is shooting for their first native quad core with Nehalem. Intel will probably run into problems like AMD is right now.

I think AMD's mistakes with its first try of quad-core are isolated. Did the Athlon64 exhibit such trouble going x2? Well, they did claim they planned ahead while designing the core. Did Intel's Core Duo fall behind Pentium M?

I don't think we could foresee AMD's problems until their odd transition to 65nm. If at 90nm you go from 1 to 2 cores, then with an equally proficient 65nm process, you should be able to go from 2 to 4 of the same cores. Not only is AMD's 65nm process not up to par with previous node shrinks, but AMD made it worse by further complicating the design.

I think the ramp from 65nm to 45nm will be much kinder and easier on AMD than that of the 90nm to 65nm.

I hold my doubts because of lackluster information on how AMD will address leakage (not simple gate leakage but source to drain I.off) which plagued their transition to 65nm and gets worse with further shrinks. If they do it as well as Intel, then their early transition to immersion lithography can help them leapfrog Intel in process quality.

But immersion lithography alone is not kind. Nor is IBM's focus on quality at all cost, since AMD must take costs into higher concern for desktop/mobile production.