Intel to Retaliate to AMD Phenom II Overclocking Feat, Plans Demonstration at CES '09

[Rash-Un-Al]

The Phenom II will have a lower "instructions per clock cycle" then the i7 and Core 2 so it already has to clock higher.

Actually, on average (across a large suite of applications), current Phenoms are at an 8-9% disadvantage when compared with Core 2 at the same clock.

AMD has stated, at 3.0 GHz, a Phenom II will be 30-40% faster than the current fastest Phenom at 2.6 GHz. Let’s take the conservative approach. If Phenom II is only 30% faster than current Phenoms, and you eliminate the frequency advantage of Phenom II, at the very same clock Phenom II will be approximately 12.6% faster than Phenom.

In other words, as already stated by analysts and hinted by AMD, Phenom II is likely to be even with or slightly faster than Yorkfield processors, clock-for-clock.

This is also supported by very conservative documented estimates of an average 5% increase in same-clock IPC resulting from enlarging of cache and 3% from core improvements – a combined 8.1% IPC improvement.

I caution that this is only a very conservative prospect. If you consider the scenarios in which an application (or game) will benefit from 3 times the cache, the IPC improvements will be far greater than described above. (Keep in mind, K10’s average improvement was approximately 15% over K8. Yet in many single-threaded applications and games, it is not uncommon to see 24-40% gains at the same clock).

Also, even if Yorkfield and Deneb find themselves dead-even in most scenarios, Deneb will shine in memory instensive situations, with lower latencies and higher throughputs.

This tells us one thing. Intel didn’t rush to Core i7 as a luxury; it was forced to in order to maintain its leadership. In the coming months, we’ll see sub-$300 Phenom 940/945 processors that rival and beat 1000-dollar QX9650s, and with just as much overclocking headroom.

Owners of current AM2+ platforms will have the last laugh – and here’s why:

• Statistics show the vast majority of Intel Quad owners are sporting Q6600s (limited to 3.6-3.8 GHz, at best, on average while expending far more energy and outputting significantly more heat than next month’s PII 940).
• Intel owners of Yorkfield 45nm Quads either have to spend $1000+ for an unlocked processor or, otherwise, face the fact the less expensive Quads are multiplier-limited.
• AM2+ platform owners will require a mere BIOS update and CPU swap to enjoy unlocked performance (potentially to 4GHz and beyond), while spending fewer than $300.
• Intel Core 2 platform owners wishing to undo being outdone by PII 940/945 PCs will have to spend – at a minimum – $500 for a platform and CPU change and closer to $1000 if seeking a premium MB and having to purchase quality DDR3 – only to outperform PII 940/945s by an even smaller margin than Yorkfields.

The good news is that the above will cause significant downward shifts in pricing, making previously unattainable (or undesirable) purchase prospects possible for many.

 

[KBD]

This tells us one thing. Intel didn’t rush to Core i7 as a luxury; it was forced to in order to maintain its leadership. In the coming months, we’ll see sub-$300 Phenom 940/945 processors that rival and beat 1000-dollar QX9650s, and with just as much overclocking headroom.

I agree with many things you say and your conservative estimates make sense to me. But this paragraph jumped at me. It seems that Corei7 could've waited another 10 months or so, as we recently learned Intel estimates that Corei7 share by Q3(?) of 09 will only be 2%, it will not be as widely adopted as 775 C2D CPUs were when they were released. Hence, your conclusion is correct, the release of i7 on Nov 17 was done purely to further solidify intel's dominance. They suspected that Deneb and AM3 will either overtake C2Q or be on equal footing with it or at least be competitive with C2Q so they had to release i7 early. Another thing intel done wrong in my opinion was create 2 CPU sockets for i7, i think that was a bad idea and anyone who will own an 1166 socket board will not be able to upgrade to Nehalem CPUs and will be forced to buy another board. After the longevity and upgradability of 775 this seems like a bad move.

 

[Wile E]

Actually, on average (across a large suite of applications), current Phenoms are at an 8-9% disadvantage when compared with Core 2 at the same clock.

AMD has stated, at 3.0 GHz, a Phenom II will be 30-40% faster than the current fastest Phenom at 2.6 GHz. Let’s take the conservative approach. If Phenom II is only 30% faster than current Phenoms, and you eliminate the frequency advantage of Phenom II, at the very same clock Phenom II will be approximately 12.6% faster than Phenom.

In other words, as already stated by analysts and hinted by AMD, Phenom II is likely to be even with or slightly faster than Yorkfield processors, clock-for-clock.

This is also supported by very conservative documented estimates of an average 5% increase in same-clock IPC resulting from enlarging of cache and 3% from core improvements – a combined 8.1% IPC improvement.

I caution that this is only a very conservative prospect. If you consider the scenarios in which an application (or game) will benefit from 3 times the cache, the IPC improvements will be far greater than described above. (Keep in mind, K10’s average improvement was approximately 15% over K8. Yet in many single-threaded applications and games, it is not uncommon to see 24-40% gains at the same clock).

Also, even if Yorkfield and Deneb find themselves dead-even in most scenarios, Deneb will shine in memory instensive situations, with lower latencies and higher throughputs.

This tells us one thing. Intel didn’t rush to Core i7 as a luxury; it was forced to in order to maintain its leadership. In the coming months, we’ll see sub-$300 Phenom 940/945 processors that rival and beat 1000-dollar QX9650s, and with just as much overclocking headroom.

Owners of current AM2+ platforms will have the last laugh – and here’s why:

• Statistics show the vast majority of Intel Quad owners are sporting Q6600s (limited to 3.6-3.8 GHz, at best, on average while expending far more energy and outputting significantly more heat than next month’s PII 940).
• Intel owners of Yorkfield 45nm Quads either have to spend $1000+ for an unlocked processor or, otherwise, face the fact the less expensive Quads are multiplier-limited.
• AM2+ platform owners will require a mere BIOS update and CPU swap to enjoy unlocked performance (potentially to 4GHz and beyond), while spending fewer than $300.
• Intel Core 2 platform owners wishing to undo being outdone by PII 940/945 PCs will have to spend – at a minimum – $500 for a platform and CPU change and closer to $1000 if seeking a premium MB and having to purchase quality DDR3 – only to outperform PII 940/945s by an even smaller margin than Yorkfields.

The good news is that the above will cause significant downward shifts in pricing, making previously unattainable (or undesirable) purchase prospects possible for many.

Unfortunately, I just don't believe PII will be a match for Yorkfield clock for clock, regardless of AMD's claims of performance on it. They have exaggerated the performance of their future chips before.

And even if it does manage to match York, it still falls behind i7. Intel could just make a big push towards i7, and leave AMD behind again.

All in all, AMD desperately needs to release a new architecture, and quit rehashing K8 over and over again, in order to truly catch up to, or pass, Intel. I personally long for that day, and hope to God they survive to pull it off.

 

[btarunr]

The new architecture (Bulldozer/SandTiger), according to Nigel Dessau (from the Analyst Day event) is slated for 2011, based on the 32nm process. 2011 sounds way too late for both a new architecture and a newer fab process considering Intel already has 32nm prototypes.

 

[Rash-Un-Al]

They have exaggerated the performance of their future chips before.

AMD’s performance projections concerning Phenom were fairly spot-on. The issue rested in the fact many consumers had a hard time converting AMD’s quad-core projections to per-core, same-clock advantages over the K8 architecture and that launch speeds were far below what was expected. And, to worsen matters, AMD displayed Phenoms operating @ 3.0 GHz, and none came to fruition. However, concerning performance projections claimed by AMD, they were consistent with actual performance.

Unfortunately, I just don't believe PII will be a match for Yorkfield clock for clock, regardless of AMD's claims of performance on it.

Unbeknownst to many, there are already a few benchmarks which dispel this notion, illustrating an average 9% IPC increase, at the same clock (enough to be competitive with and even comparable to Yorkfield at the same clock).

Feel free to take a look...

And even if it does manage to match York, it still falls behind i7. Intel could just make a big push towards i7, and leave AMD behind again.

Tough proposition. Changing platforms (purchasing a new CPU, Motherboard, and Memory) isn't at the top of even enthusiast's to-do list. That push would require dramatic price reductions. I don't think investors would be too happy with that.

All in all, AMD desperately needs to release a new architecture, and quit rehashing K8 over and over again, in order to truly catch up to, or pass, Intel. I personally long for that day, and hope to God they survive to pull it off.

Indeed -- we all benefit from stronger competition.

 

[Tatty_Two]

Botttom line is, this dont mean sh*t if these chips arent readily available to the consumer, bit like saying a formula 1 Grand prix car is the fastest in the world........"well knock me down with a whore's wet wipe" none of us have one so it's a bit immaterial, now if we could get our hands on them, that would be different

 

[KBD]

The new architecture (Bulldozer/SandTiger), according to Nigel Dessau (from the Analyst Day event) is slated for 2011, based on the 32nm process. 2011 sounds way too late for both a new architecture and a newer fab process considering Intel already has 32nm prototypes.

yea, i heard about this also. this is wayyyyy too late. I dont know what the hell is taking them so long, Intel will prolly have their 32nm on the market by 2010, i hope they make good use of that money they got and bring this tech out earlier. K8 has been around since 2003 for god's sake, it takes them 8 years to come up with a totally new arch, thats just crazy.

 

[Rash-Un-Al]

none of us have one so it's a bit immaterial, now if we could get our hands on them, that would be different

Agreed.

January 8, 2009 is when the channel is expected to have stock.

 

[Wile E]

AMD’s performance projections concerning Phenom were fairly spot-on. The issue rested in the fact many consumers had a hard time converting AMD’s quad-core projections to per-core, same-clock advantages over the K8 architecture and that launch speeds were far below what was expected. And, to worsen matters, AMD displayed Phenoms operating @ 3.0 GHz, and none came to fruition. However, concerning performance projections claimed by AMD, they were consistent with actual performance.



Unbeknownst to many, there are already a few benchmarks which dispel this notion, illustrating an average 9% IPC increase, at the same clock (enough to be competitive with and even comparable to Yorkfield at the same clock).

Feel free to take a look...



Tough proposition. Changing platforms (purchasing a new CPU, Motherboard, and Memory) isn't at the top of even enthusiast's to-do list. That push would require dramatic price reductions. I don't think investors would be too happy with that.



Indeed -- we all benefit from stronger competition.

3 tests from a random forum, 2 of which are already known Phenom strong suits, doesn't draw any conclusions at all.

And it is still easier for Intel to lower prices and flood the market with i7 (and it's soon to be mainstream derivatives), than it would be for AMD to make a real dent in the market. If Intel wants to overshadow/overtake AMD, they will. They have the financial ability to do so.