AMD chief say BD will offer only 35% not 50% more performance than previous gen

[AtenRa]

I will have to say here that i dont believe that we can measure IPC of the CPUs by running a benchmark even with a single core.

I would say that it is better and more accurate to say that CPU X has 50% more single core performance than CPU Y at that specific benchmark/s at the same frequency.

 

[Nemesis 1]

So basically, in 2 months from now, we'll all be able to buy an equivalent to an 8-core 2500k @ 4.0ghz, despite 5 years of non-competitiveness, no single AMD benchmark leak, no official explanation for 3x or so they delayed BD? Oh I know, I know exactly where I heard this before (the perfect alignment of all the stars in the universe with no disadvantages) -- like that non-sense claim that HD6970 was going to be the next "R300" and beat GTX580 by 40-50% in DX11

Thats not what I recall . I for one as many others at the time you referrance . Did not have a clue 580 was going to be released. I do believe the numbers your referring to were comparred to a 480 and the 6970 was a early debug model that turned out to be extremely close to the release product in performance . A rare thing with GPUs.

 

[Lonbjerg]

Im sorry to say but Q6600 was not a quad core design but 2 dual core dies in the same PGA (Pin Grid Array).

So by comparing the two processors in benchmarks and say that SB 2500K has ~50% more IPC is wrong.
In order to see the IPC of the two designs we have to measure ONLY a SINGLE core of each processor (edit: at the same frequency).

That ~50% between Q6600 and i5 2500K is performance and not IPC.

Is this the washed out "native quadcore...but still lose" Phenom argument with a new twist?

 

[AtenRa]

Is this the washed out "native quadcore...but still lose" Phenom argument with a new twist?

No, its a technical observation.

If you measure a single core of the Q6600 it will give you a higher performance per frequency percentage than 4 cores because you will eliminate the communication bottlenecks between the two different dies in multithreaded apps

 

[Nemesis 1]

I will have to say here that i dont believe that we can measure IPC of the CPUs by running a benchmark even with a single core.

I would say that it is better and more accurate to say that CPU X has 50% more single core performance than CPU Y at that specific benchmark/s at the same frequency.

I would say that many AMD backers would agree with this type of thinking. The turbo boost was one of the most entertaining cpus events to happen in recent years it was so amusing listening to the whinning of the red team . Than later more recently an AMD employee involved in marketing was telling us all about this super dupper 8th wonder of the world Turbo boost to be introduced on AMD 32nm . Didn't show up on llano and wasn't said to be either. What ya want to bet Intels Turbo makes AMD BD turbo look silly .

My wife only uses this computer on occassion . The whole of this week she is been all over me to put water on this browser . Its not going to happen . She gets all excited when she starts up this PC when its cold and she opens an app and watches Turbo go over 6ghz for a second or 2. She says the 2600K doesn't go anywere near that . Which is a fact . I not real sure whats going on with that . I was going to take this thing over to skinnies Lab . He not far away 20 min. But wife quickly pointed out that he would see that which I am hiding. Its something not being used but is in plain site. I want skinny to see it and test and report on it but not yet . I am still way to alive for that.

 

[Nemesis 1]

Is this the washed out "native quadcore...but still lose" Phenom argument with a new twist?

The AMD native wording is back in use with the All new 8core worlds first native 8 core cpu . Here for over a year I thought BD was a modular design .

 

[Phynaz]

I can't wait for BD and SB-E to come out so I don't have to hear all of this arguing and speculation anymore.

Please report whoever is holding that gun to your head to the authorities.

 

[LOL_Wut_Axel]

I don't recall ever stating that there is a perfectly linear scaling with increased frequency. I said it was 'excellent' or 'nearly linear' for SB and that it will still be there at higher clock speeds against Phenom II. As such, I believe that SB's IPC advantage will also matter against BD, which is the main argument at hand in this thread.



:thumbsup: 2500k does Turbo to 3.7ghz when only 1 core is used. However, take a look at what happens in the Multi-threaded Cinebench R10:

Q6600 2.4 = 9,681
2500k 3.3 (3.4) = 20,381 (the same 2.11x increase). <-- 49&#37; difference then.

So I don't think in that single-threaded bench the 2500k was running at 3.7ghz.

Here are more benchmarks:

Source
2500k is 54% faster than the X4 970 at similar clock speeds.

2500k is 55% faster than the X4 970.

2500k is 68% faster than the X4 970.

2500k is 44% faster than the X4 970.

2500k is 38% faster than the X4 970.

There are scenarios such as WinZip where 2500k is only 22% faster, for instance.

However, based on a multitude of benchmarks I provided from Anandtech, Hardware Secrets, Xbitlabs, Computerbase, SB's advantage in IPC over PhII is closer to 45-50%, not 35%.

You still don't get it, do you? You're not supposed to use multi-threaded benchmarks unless the CPUs been forced to use one thread and the same clock speed to directly compare architectures. You shouldn't use synthetics, either.

 

[bryanW1995]

You still don't get it, do you? You're not supposed to use multi-threaded benchmarks unless the CPUs been forced to use one thread and the same clock speed to directly compare architectures. You shouldn't use synthetics, either.

But with BD in single thread it will have the advantage of using the entire module, and advantage that will disappear with more than one thread (or possibly 4 depending upon how threads are scheduled). How do we deal with that?

 

[Troll Trolling]

But with BD in single thread it will have the advantage of using the entire module, and advantage that will disappear with more than one thread (or possibly 4 depending upon how threads are scheduled). How do we deal with that?

Run two instances of the same benchmark on the same module?

 

[LOL_Wut_Axel]

But with BD in single thread it will have the advantage of using the entire module, and advantage that will disappear with more than one thread (or possibly 4 depending upon how threads are scheduled). How do we deal with that?

Reviewers will know how we can deal with that. Bulldozer isn't here still.

As it goes now, comparing this way is the best for all current CPU architectures.

 

[Riek]

But with BD in single thread it will have the advantage of using the entire module, and advantage that will disappear with more than one thread (or possibly 4 depending upon how threads are scheduled). How do we deal with that?

and that is exactly the same as intel with HT. one thread running is full speed, if the thread runs over HT it becomes crawl speed in single thread performance over that core. (60&#37; ipc/thread left)

:thumbsup: 2500k does Turbo to 3.7ghz when only 1 core is used. However, take a look at what happens in the Multi-threaded Cinebench R10:

Q6600 2.4 = 9,681
2500k 3.3 (3.4) = 20,381 (the same 2.11x increase). <-- 49% difference then.

So I don't think in that single-threaded bench the 2500k was running at 3.7ghz.

You don't see the flaw in your logic?
The single threaded IS running at 3,7GHz. What you apparently still don't understand is that ipc fluctates over design, over cores, over frequency. This is again the proof you use against others that disproves your own logic.

What you are seeing in that test is that the ipc/c changes according to the load.

1c load performance difference is x, 2cores in use performance difference is y, 3cores in use performance difference is z, 4cores in use performance w

Where x != y != z != w

Thats right not every cpu scales the same with frequency
scales the same with memorybandwidth (p4 anyone)
scales the same with latency (K8 anyone)
scales the same with cache seze
scales the same in multi cpu configuration
scales the same in the same application

IPC is NOT constant figure.

 

[996GT2]

The AMD native wording is back in use with the All new 8core worlds first native 8 core cpu . Here for over a year I thought BD was a modular design .

Didn't they also claim Phenom I was the first native quad-core?

Look how that performed against Core 2 Quad

 

[AtenRa]

But with BD in single thread it will have the advantage of using the entire module, and advantage that will disappear with more than one thread (or possibly 4 depending upon how threads are scheduled). How do we deal with that?

BD will not use the second INT unit in a single thread situation, so I don&#8217;t see what the problem is.

 

[Spikesoldier]

i wish someone would make that THG chart with superpi instead of 'THG total time for benchmarks'.

 

[LOL_Wut_Axel]

i wish someone would make that THG chart with superpi instead of 'THG total time for benchmarks'.

SuperPi would be horrible to use, to say the least. It's a benchmark that is heavily biased in favor of Intel processors because of the instruction set it uses. That's probably why most reviews don't even care to mention it.

Also, what's wrong with compiling all the benchmark times so we can see how much time each CPU took overall? It's probably the best way to do it.

 

[RussianSensation]

EDIT: To add, since we have touched upon turbo and clock-to-clock comparisons, I would like to express my frustration in this same endeavor that you are trying to accomplish: IPC comparisons of processors ever since "Turbo" became widespread.

I remember spending more time than usual (and more frustrating than usual) looking for clock-to-clock comparisons, because that is what I need for my usage since overclocking comes into play with Turbo off. I know why review sites bench with turbo on, because that is the "out-of-the-box" performance, but as an enthusiast who will only run the chip at its highest stable clock, I don't care about the "out-of-the-box" performance, and for me to know what I need to know, I need clock-to-clock comparisons.

With all the dynamic and hard-to-predict "turbo" going on, it's not as easy as before when processors only came with one speed setting, unless a reviewer specifically allots a section for IPC comparisons, and it doesn't happen often at all.

Agree with you completely. :thumbsup: Each new major CPU launch should have 2 additional sections in the review:

1) Clock-for-clock comparisons between various architectures
2) The 2-3 most modern competitors should be compared in their top overclocked operational modes on air cooling (as Xbitlabs did by overclocking i7 875, 990X, 1100T, 2500k and 2600k) <-- this prob. is the most useful review scenario for us overclockers.

 

[RussianSensation]

Im sorry to say but Q6600 was not a quad core design but 2 dual core dies in the same PGA (Pin Grid Array).

So by comparing the two processors in benchmarks and say that SB 2500K has ~50&#37; more IPC is wrong.
In order to see the IPC of the two designs we have to measure ONLY a SINGLE core of each processor (edit: at the same frequency).

That ~50% between Q6600 and i5 2500K is performance and not IPC.

Whether or not the Q6600 was a "true" quad core is irrelevant. I'll explain.

The intention to bring Q6600 in this thread was solely because its performance to the Phenom II X4 is almost identical at similar clock speeds (See this review for details). Once I established that Q6600 had similar performance per core to the Phenom II, I then proceeded with at least 4 different reviews to verify that 2500k has a 45-50% faster performance per core vs. either the Phenom II X4 or the Q6600. Notice I independently compared 2500k vs. Q6600 and then 2500k vs. Phenom II X4. Since we already knew that Q6600 ~ Phenom II per clock, we should have expected 2500k's advantage to be almost exactly the same over either the Q6600 or the Phenom II X4. The 4 reviews showed exactly that.

In each independent review comparison of 2500k vs. Q6600 and of 2500k vs. Phenom II X4, the average performance differences were 45-50% per clock (once adjusted for math). This was true across Anandtech, XBitlabs, Computerbase, and Hardware Secrets. In other words, not a single one of those reviews deviated from the hypothesis range. They all showed similar information of SB performing approximately 45-50% faster on average per clock vs. previous architectures.

 

[RussianSensation]

Thats not what I recall . I for one as many others at the time you referrance . Did not have a clue 580 was going to be released. I do believe the numbers your referring to were comparred to a 480 and the 6970 was a early debug model that turned out to be extremely close to the release product in performance . A rare thing with GPUs.

Claims were made that HD6970 would be "R300-like" vs. GTX580. I then provided benchmarks referencing the fact that 9700Pro/9800Pro were 40-50&#37; faster than GeForce 5800/5900 series in DX9 applications (the latest API at the time). As such the likelihood of HD6970 to be R300 like was nil as far as logic dictated.

In other words, to claim R300 like performance implied that HD6970's DX11 performance would eclipse GTX580 in DX11 by a similar delta of 40-50%. Despite the massive amount of hype and fake HD6970 specs (1920 SPs), the HD6970 is really a competitor to the GTX570, and not the GTX580......Yet until the last week before HD6970 was launched, a lot of claims were made that it would blow the doors off the GTX580.

She gets all excited when she starts up this PC when its cold and she opens an app and watches Turbo go over 6ghz for a second or 2. She says the 2600K doesn't go anywere near that .

What kind of a 2600k does your wife have? From Haswell generation?

 

[RussianSensation]

You still don't get it, do you? You're not supposed to use multi-threaded benchmarks unless the CPUs been forced to use one thread and the same clock speed to directly compare architectures. You shouldn't use synthetics, either.

In other words, when I pick apps that run much faster on a higher IPC architecture, i.e., Core i3-2120 stomps all over the Phenom II X4 in games, you claim its unfair because games aren't perfectly multi-threaded.

Then I show 2500k smoke both the X4/X6 1100T in real world multi-threaded apps, and you still claim it's also unfair because I am not comparing them in a single threaded scenario?

Like I told you before, let's for a second accept your claim that only a 35&#37; IPC difference exists (and not a 45-50%), then a 4.2ghz 8-core Bulldozer with identical IPC to Phenom II would be equivalent in performance to an 8-core 2500k Sandy Bridge clocked at 3.1ghz (4.2ghz / 1.35 = 3.1ghz). That's your insinuation then? And we'll be able to buy such a beast of a processor for just $300 in 2 months, correct? Sign me up for 100 of those. ^_^

i wish someone would make that THG chart with superpi instead of 'THG total time for benchmarks'.

That's never going to happen. Unlike the time when Athlon 64/X2 dominated Pentium 4/D in Super Pi and it was "fashionable" to compare Super Pi times, the current AMD owners dismiss Super Pi as "biased" and irrelevant.

 

[LOL_Wut_Axel]

[I]I then proceeded with at least 4 different reviews to verify that 2500k has a 45-50&#37; faster performance per core vs. either the Phenom II X4 or the Q6600.

That's wrong. You provided images to benchmarks, many including synthetics, that show the differences in multi-threaded performance. We're comparing architectures here. We're discussing architectures and instructions per cycle, which you can't compare with the CPUs using four cores because there's many other factors to take into account, not to mention it doesn't test the architecture at the same raw level.

Again, if you want to make a good argument regarding architectures and IPC, you do not (1) use synthetic benchmarks to base your general argument , (2) you set the CPUs to work with a single thread,and (3) you set the CPUs to work at the same clock speed. As it's been shown already, though, across 18 benchmarks with the CPUs running a single thread at 3GHz, Sandy Bridge has 39% higher performance. I'll add the image again:

Here's the article from which the study was made:
http://www.tomshardware.com/reviews/processor-architecture-benchmark,2974.html

The bottom line: Sandy Bridge has 39% (40% if you want to round it out) instructions per cycle than K10.5.

 

[notty22]

Didn't they also claim Phenom I was the first native quad-core?

Look how that performed against Core 2 Quad

Yes they did.
I believe it was the other thread where marketing got in to the discussion.
Harping on the native quad core VS Intel's design makes technical sense if not really important.
But I always disliked how I learned about the tri-core, or how it was marketed. With AMD's design, they were faced with more imperfect cores/wafers and had to offer the tri-core to recover costs, imo.

I looked back at the AMD press releases.

http://www.hexus.net/content/item.php?item=9881
Hot off the back of the recent Barcelona launch, AMD's singing about its up-coming desktop line of Phenom processors. The company has said it will be bringing a three-core version to the market early 2008.
Yes, that's right, an odd*, prime number of cores will be coming to a desktop near you, next year.
So what's a three-core CPU good for? Well, when four's too much and two ain't enough, would be the obvious answer.
AMD's reasoning, of course, is a little more detailed than that.
The company cites a Mercury Research paper in which it is said quad-core CPUs only accounted for 2% of desktop CPU shipments in Q2 2007. AMD reckons "this suggests a need for greater choice and wider selection of multi-core solutions".
The 65nm based chips will contain three cores on the same die - a point AMD is pushing as an advantage over some of Intel's "multiple die on a package" chips. They'll sit on the proven Hyper Transport interconnect and have access to an on-die DDR-2 controller, much like current offerings.

 

[LOL_Wut_Axel]

In other words, when I pick apps that run much faster on a higher IPC architecture, i.e., Core i3-2120 stomps all over the Phenom II X4 in games, you claim its unfair because games aren't perfectly multi-threaded.

Then I show 2500k smoke both the X4/X6 1100T in real world multi-threaded apps, and you still claim it's also unfair because I am not comparing them in a single threaded scenario?

Like I told you before, let's for a second accept your claim that only a 35&#37; IPC difference exists (and not a 45-50%), then a 4.2ghz 8-core Bulldozer with identical IPC to Phenom II would be equivalent in performance to an 8-core 2500k Sandy Bridge clocked at 3.1ghz (4.2ghz / 1.35 = 3.1ghz). That's your insinuation then? And we'll be able to buy such a beast of a processor for just $300 in 2 months, correct? Sign me up for 100 of those. ^_^

I never claimed that. Whatever comparison I made for the Phenom II X4 955 was about it being a faster CPU in some programs and slower in others, not about it having a faster architecture. It seems to me that you can't distinguish between having a faster CPU and having a faster architecture.

Also, on some multi-threaded applications the Phenom II X6 1100T is the same speed as the Core i5 2500K, if not a tiny bit faster:

And your idea of an FX 8-core CPU having the same multi-threaded performance as a theoretical 8-core Sandy Bridge CPU given Nehalem IPC is false. Like I said earlier, an FX-8150 at 3.6GHz base would be a bit faster than a Core i7 990X at 3.46GHz base in multi-threaded performance. The FX CPU would have two additional cores, but the Core i7 would have HyperThreading. Having those many cores decreases performance scaling somewhat, but HyperThreading adds an average of about 20% higher performance. Those two additional cores would add a similar amount as well. So, all in all, given Nehalem IPC, the FX-8150 would be somewhat faster than a Core i7 990X. In multi-threaded performance it would win by a noticeable amount against a Core i7-2600, and in single-threaded performance it would lose slightly.

 

[bridito]

...So, all in all, given Nehalem IPC, the FX-8150 would be somewhat faster than a Core i7 990X...

With the utmost respect and reverence for your thoughtful calculations, I beg to differ... and trust that the benchys will prove my point (if I live long enough to see them). However, I would be very pleased if FX-8150 out of the box would provide the level of performance as I'd likely buy it on the spot on Day One of release.

Benchys talk, and... er... well... benchys talk! That's it.

 

[waffleironhead]

Yes they did.
I believe it was the other thread where marketing got in to the discussion.
Harping on the native quad core VS Intel's design makes technical sense if not really important.
But I always disliked how I learned about the tri-core, or how it was marketed. With AMD's design, they were faced with more imperfect cores/wafers and had to offer the tri-core to recover costs, imo.

I looked back at the AMD press releases.

How do you know intel doesnt suffer just as many problems with their production? Do they release % numbers of valid chips?