TuxDave
Lifer
- Oct 8, 2002
- 10,571
- 3
- 71
I dunno why the whole 'things take off at # GHz' that Gary posted. To me that's only worth saying if it scales beyond what is normal so a 10% bump in frequency from 2.4GHz to 2.64GHz somehow gave it a 15% performance gain. To me that would be 'taking off' at some frequency.
If on the other hand like the P4 example, if he just said "It becomes competitive at 2.4GHz" then it all makes sense just as how the P4 was a non-competitive product at frequencies below 1.8GHz.
Arkaign
Lifer
- Oct 27, 2006
- 20,736
- 1,379
- 126
yep, in fact, without other variables (higher FSB, etc) there's no logical way that an increase more than 1:1 of performance v clockspeed is even possible, with the same core.
The increases in P4 performance were due to improvements in the architecture, not magic 1.0/1.1+ mhz/performance returns upon clock speed increase. The P4, over it's lifespan, went from a measly 400fsb/256k L2, to 533 and 800fsb, and 512k and 1m L2 caches, and newer SSE capabilities, and hyperthreading.
Anyways, a P4 Willamette core, from 1.3ghz to 2.0ghz, never showed a rising performance curve vs core speed. Though it did become faster than the Athlon XPs at the time (1600+) by reaching the 1.8Ghz clock speed. The Northwood revision was the real shot-in-the-arm that Intel needed, and I do believe that the extra L2 cache and higher FSB account for the drastic performance uptake we saw at the time. It'd take some digging in old articles, but if someone thinks that an excess of a 1:1 improvement can be found in clock frequency alone, without touching ram/fsb access at all, I'd like to hear it.
'takes off' is subject to interpretations. I think he meant it that way, because AT is under NDA rite now.
As far a scaling, the A64 scaled something like 1.5 to 1 as compared to the P4. Meaning that for a 100 MHz bump on both processors, the A64 would see something like 50% more performance increase over the P4.
As far a scaling, the A64 scaled something like 1.5 to 1 as compared to the P4. Meaning that for a 100 MHz bump on both processors, the A64 would see something like 50% more performance increase over the P4.
Agreed...it seems to me that either the benches were done on a very old Barcelona chip, or they have been faked.
I still don't see why it says both Opteron 2332 and Agena on the chip as they are very different except at the basic core level.
1. Barcelona is HT 2.0, Agena is HT 3.0...
2. Barcelona is socket 1207 and multiprocesser, Agena is socket AM2+ and uniprocesser
I doubt those are fake benches, but the performance of the retail part isn't going to be as bad as they indicate. Nor do I believe they are going to be as good as the INQ story about 30k 3dMark06.
I'm guessing that overall, clock for clock it will be on par to 10% better than Penryn based on the overviews I've seen of the arch. But the question is can AMD scale the clocks high enough to be competitive?
Keysplayr
Elite Member
- Jan 16, 2003
- 21,219
- 56
- 91
True. Willamette was a heartbreaker. Until the 1.6A (Northy 478) came out, which wasn't a rocket either until it hit 1.8 or greater, Intel was in deep you know what. I remember the benches clearly. The Northy didn't begin to challenge the adjacent Athlons until 1.8GHz, and then as Intel ramped up clockspeed like they were insane, of course performance improved. Probably because the L2 cache operated at twice the core frequency, and finally reached a happy place at a certain threshold. Who knows.
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
Not exactly, I cannot go into the details yet, just imagine the cache/memory pipeline as being a supercharger on a car, you have an engine (same compression ratio/cubic inches for NA versus SC) that performs the same until you hit a certain RPM/air-fuel mixture where the SC comes online and the power curve changes dramatically compared to the NA engine. The same basis is occurring here, all of the changes/enhancements made to the core / HT/ cache / memory controller are basically "idle" in some cases (SC is flowing more air than than the engine can take advantage of at low RPM plus you have parasitic drag from powering the SC), if not a hindrance (low compression and mismatched gearing). An engine (CPU) is most efficient at its torque peak (wherever that happens to be based on gearing, displacement, compression, efficiency, etc) and in this case, it starts nearing it (torque peak) around 2.4GHz from all indications.
This is a very crude and simple example but about the only way I can state information at this time. The simple fact is, this core needs clock speed and until 2.4GHz or so, it is not that impressive in my opinion on the desktop. Certain transactions/operations will be noticeable/improved over some of the Core 2 family processors on a performance/wattage aspect from an enterprise overview, but this chip design is going to require clock speed before you start seeing some numbers that make sense in the consumer/workstation market. That is why Phenom will launch at higher clock speeds, different core configurations, and with chipsets designed to take full advantage of the core changes with consumer applications. That is my guesstimate based on performance up to this point, several of the larger OEMs received their final silicon chips this past week, most noticed another improvement in performance, how much, we will find out shortly but do not expect a leap frog over the Blue Bunnies yet.
I would not be surprised to even see supply shortages until early next year and still think BullDozer is the AMD CPU that will make or break the company. Barcelona and Shanghai, plus their desktop counterparts, are evolutionary designs (core basis) based off the K7/K8. This is why I have never really understood the hype around this launch, yes, performance improvements will come and we get some interesting new technologies like HT 3.0 /native quad core layout, but without clock speed improvements to match, you end up with A64 X4+ for a lack of simpler words.
formulav8
Diamond Member
- Sep 18, 2000
- 7,004
- 523
- 126
The new chips scaling very well is probably right on. The memory controller speed will increase reducing latency. AMD has also stated you can specifically overclock the northbridge and gain anywhere from 3%-10% increases in performance which reduces latency.
So, I can understand why Gary mentions it scaling the way it does. Just my opinion though
Jason
So, I can understand why Gary mentions it scaling the way it does. Just my opinion though
Jason
Originally posted by: SickBeast
Gary Key: Thanks very much for the information.
Seeing as the AT staff don't even have the lastest Barcelona cores, then I've gotta call shens on all of these 'preliminary benchmarks'.
Actually, based on the last chip we had, those numbers are in alignment with some of the results we noticed and others as well. I think they will be better than that at release and especially on a consumer board with Vista 64-bit. Until we have final silicon and boards, anything is a guesstimate at this time. I just think people are going to be disappointed in some ways, as were we, until the clock speeds come up.
And very expensive RAM to fuel this supercharged engine! AMD take-over is still ON. Show me 15 bucks.
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
while this is certainly not unexpected news, it is scary to think that phenom has to run at 2.4 to be competitive. I've actually been considering a phenom upgrade at the same time that I snag a penryn quad core in Q1, but I'm a little concerned about cost now if I'll need cas4 ddr2 1066 or cas5 ddr2 1333. One of the reasons that I'm so gung-ho to upgrade in Q1 is that I can just buy a cpu for penryn and a cpu/mobo for phenom. If I have to get into some expensive ram, too, then I would probably just keep my e6750 and guy a cheap mobo to run it with instead of getting a whole phenom system.
I assume you're referring to the SuperPi score, seeing as you already called 'shens' on the 3DMark '06 score of 30,000.Originally posted by: Gary Key
Originally posted by: SickBeast
Gary Key: Thanks very much for the information.
Seeing as the AT staff don't even have the lastest Barcelona cores, then I've gotta call shens on all of these 'preliminary benchmarks'.
Actually, based on the last chip we had, those numbers are in alignment with some of the results we noticed and others as well. I think they will be better than that at release and especially on a consumer board with Vista 64-bit. Until we have final silicon and boards, anything is a guesstimate at this time. I just think people are going to be disappointed in some ways, as were we, until the clock speeds come up.
If that's the case, Barcelona is in big trouble. I don't know who in their right mind would buy a new chip that can be outperformed by something a few years old.
You mentionned that AMD won't outrun the 'Blue Bunnies' yet. That's a terrible disappointment seeing as the current A64s come pretty close to the C2Ds when both are running at around 3.2ghz. What I take from what you're saying is that these chips are marginally better than current A64s clock-for-clock, and that is absolultely devastating news for AMD (especially considering that they have delayed this chip far too long).
Again, I greatly appreciate your being so forthcoming. It's nice to have a glimpse at the hardware before the full review goes live.
Except a CPU has static resources, whereas a car engine can burn more fuel because a supercharger fed it compressed air. CPU core frequency increases cannot never yield greater than a linear performance gain.
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
yes, but a k10 could have a greater performance boost from a given clock increase than a core 2 or penryn chip. from what gary says, that starts to become meaningful at around 2.4. for example, a barcelona at 2.0 might not perform any better than a penryn quad at 2.0, but the barcelona might get a 5% clock for clock advantage at 2.4, then 10 % at 2.8, etc b/c it could come closer to the theoretical 1:1 than intel's offerings.
A 10-20% increase in clockspeed is not going to yield a 50% improvement in performance, which is what AMD needs. If that CPUMark99/SuperPi results are real, then Barcelona is an absolute joke. My Celeron M laptop almost outscores it and my C2D absolutely kills it.
That is certainly not true today, when cpus have so many other factors that conincide with cpu frequency. Such as multiple cores, HT, integrated memory controllers, enhanced buffers, enhanced branch chain prediction and the etc. So cpu frequency in general in todays cpus is not the dominant factor that it once was.
Keysplayr
Elite Member
- Jan 16, 2003
- 21,219
- 56
- 91
If the L2/L3 cache runs at twice the speed of the core, then I can see this happening. At 2GHz, a Barcelona's cache runs at 4GHz (hypothetically). Pump up the core speed to 2.6 GHz and the cache will be at 5.2GHz. So with only a 600MHz core increase, the cache speed increased 1200MHz.
The Pentium 4 did this, but after cranking up the clocks so high, diminishing returns were experienced. Wall was hit.
I'm not saying Barcelona will be this way, but it could explain the "power band" of the processor above 2.4GHz.
The Pentium 4 did this, but after cranking up the clocks so high, diminishing returns were experienced. Wall was hit.
I'm not saying Barcelona will be this way, but it could explain the "power band" of the processor above 2.4GHz.
Gikaseixas
Platinum Member
- Jul 1, 2004
- 2,836
- 218
- 106
Gary has a good point. I mean if the chip scales on a "not so traditional way" then a 100mhz speed difference represent's a higher than normal bump in real life performance and that's why AMD has 1.9, 2.0, 2.1, 2.2, 2.3 and so on line-up...
It makes sense now
It makes sense now
P4 ran parts of the ALU at double the core frequency, not the cache. Even with that kind of pipeline, frequency increases cannot not yield greater than a linear increase performance.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter