Separate names with a comma.
Discussion in 'CPUs and Overclocking' started by Fjodor2001, Aug 1, 2012.
Thats roughly 6 cores of work there, looks to scale bad if you dont have a bottleneck somewhere
Yes, because they added more CPU cores and an iGPU. See this picture of the Ivy Bridge (and Sandy Bridge) CPU die layout:
As can be seen, almost all of the available space is occupied by CPU cores, iGPU and caches. Adding integrated memory controller, VRM and such does however not require that much silicon real estate at all.
So since Intel within 4-5 years (2 node shrinks) will be able to fit 4 times the amount of transistors on the same CPU die real estate, the question still remains what Intel intends to do with all those transistors? As I see it they can keep increasing the iGPU and/or add more CPU cores.
It's not about that at all. But maybe you want to turn it in that direction?
From my side, I truly want to know when we can expect Intel to start delivering 8 core mainstream CPUs. I am actually hoping it will be already with the next node shrink (Broadwell), but absolutely no later than the following one (Skymont).
Actually not true. Recently AMD has been going for more cores, but Intel hasn't. And yeah, I know AMDs 8 core chips do not have 8 "true" CPU cores, but still.
As I wrote before:
"Don't we have a chicken and egg scenario here? If the Intel mainstream CPUs would be 8 cores instead of 4 cores, don't you think the SW would be adapted to make use of that?
For example, BF3 makes good use of 4 cores. Do you think that would still be the case even if Intel would have decided to stay at 2 cores for their mainstream CPUs, instead of transitioning to 4 cores?"
Also, it's really funny that some people seem to think that 4 cores is exactly what is needed - no less no more. They are very happy that we transitioned from 2 to 4 cores, but think it is pointless to ever go beyond that. I just don't get it! If Intel and AMD would have decided to not transition from 2 to 4 cores, then I guess the same crowd would be saying it is pointless to ever go beyond 2 cores...
You do know AMDs future bet is 2 modules/4 threads APUs? AM3+ platform gets one last CPU with 4 modules/8 threads and thats it. No more there, deadend.
AMD is going for less cores and more iGPU in the future.
And you compare SB with IB cores in size? I told you to compare Conroe with IB. How many transistors does a Conroe core use? And an IB core?
No, I'm not. The picture just happened to contain an image of Sandy Bridge as well, which was why I put "(and Sandy Bridge)" within parenthesis in the text. The intention was just to show the layout of the Ivy Bridge die, nothing else. Sorry for the confusion.
Anyway, Conroe was 3 node shrinks ago (we have transitioned from 65->45->32->22 nm). I.e. Ivy Bridge can fit 8 times the amount of transistors on the same die area as Conroe. Are you saying each Ivy Bridge CPU core has 8 times the amount of transistors compared to a Conroe CPU core (or anywhere close to that)?
I just happened to take the screenshot when the cores weren`t at 100%.
The graph shows that it didn't hit 100% at anytime during your measurement...the data is right there!
Good. AMD should not be fighting losing battles.
Can it be assumed they they're also looking into higher IPC, or is the iGPU their only concern?
Higher IPC/clock too.
One interesting aspect is that Intel will introduce hardware transactional memory (TSX) in Haswell. See e.g.:
The article says on page 4:
"Generally, Intels TSX should be helpful for improving the programmability and scalability for concurrent workloads. Even with a modest number of threads, locks can easily limit the benefits from additional cores. While that is not a problem for 2-4 core processors, it is a much bigger factor going forward."
So couldn't the introduction of TSX be a sign that Intel is preparing for more cores in later CPU generations?
Also, as previously has been stated, 4-5 years from now Intel will have 4 times the amount of transistors on the same die area (due to two node shrinks). They can of course use that to keep growing the iGPU, but it is starting to reach a point where its performance is sufficient for most users, and those that require more (primarily for games / video editing) will buy a discrete GFX card anyway. So wouldn't it be reasonable to assume that at least half of the 4 times amount of transistors will be consumed by additional cores?
Actually it's Beckton or Nehalem-EX... but yes 2010...
Westmere-EX is supposed to bring 10 cores to the party.
3ds Max mental ray and After Effects can certainly make quite good use of an n-core system. Unfortunately, I don't have money to upgrade to anything beyond 4 cores.
Either that, or they could still sell 4 core chips and just make more money per wafer.
First off, Intel already has 8 core processors available in servers. So the have the technology and could theoretically bring that to the desktop (probably at a high price though) any time they want...if there were demand for it.
Which brings me to my second point. It's not really true that there have been no performance increases. Today's CPU's blow away previous generations in every regard except for one. And that is real world software that actually uses all that processing power. For example for gaming, there is a lot of evidence to support the idea that going above a 2500K is a waste of money as the 2500K is good enough for any game out there when used with the appropriate GPU. And for most home users, gaming is about the most intensive thing they will do with the machine. Once you hit that good enough for the task point, what is the point of more cores? Everyone with a Sandy Bridge generation quad core probably has a cpu that is good enough for several years to come.
Absolutely. dinker99's bad multi-core scaling is a result of the lack of support for hardware transactional memory and lock elision, not bad software design. More cores could even lead to lower performance due to the overhead of handling quadratically more interactions between cores/threads. So we need TSX first to lower the overhead and facilitate the programming, before it makes sense to have more cores.
Meanwhile a quad-core Haswell CPU will still be a very significant upgrade, thanks to AVX2. It can easily make applications like Sony Vegas run twice as fast.
Both those application will benefit from AVX2's double vector throughput just as much as from doubling the number of cores, if not more.
And so the plan is that games will just stagnate where they are now? We've hit the limit??
Systems are fine now for 1080p gaming, but with the push to higher resolutions and of course continuing improvement in physics, lighting, ai, tomorrows game will always run not as well on yesterdays tech. Might still get that min 30 fps but how many pieces of eye candy do you have to turn off?
No, Haswell's AVX2 doubles the throughput per core, and its TSX extension prepares for more cores in the future.
If there was no demand for CPUs that perform well in multi-threaded scenarios, then how come Intel 3770K is selling well? If there was no demand for it then everyone would just be getting the 3570K instead.
I know 3770K only has Hyper-Threading support and not 8 "true" CPU cores, but still the point is that it only performs better than 3570K in multi-threaded scenarios (apart from some minor frequency difference). And there's clearly a demand for it, since the 3770K is selling well...
That's about like saying since cars that go 120mph sell well in the mainstream, cars that go 240mph (and the resulting price) would sell well in the mainstream...
Sorry, no. You want 8 Intel cores, just buy a Xeon, there have been 8 core (+HT) models available for almost two and a half years. The product is available.
The 3770K is a 100Mhz speedbin + 2MB cache and thats why. And those that buys it for HT does it due to having software that enables them to utilize it.
We already got CPUs capable of all those threads. Its software that lacks.
It doesnt help to use more cores if there aint a benefit of it.
Intels Mitosis project is a good example on how far behind software is. And how much software simply cant be multithreaded in a beneficial way. Atleast not past a low amount of cores. Thats when Intel thought about basicly throwing endless execution resources to get incredible small performance gains for the software that couldnt. We talk scenarios with 800% more resources to get a 10% speed boost. Mitosis today is dead and all money and time wasted.
AMD also burned its fingers in that game. Their result was basicly just idle cores in CPUs that added extra costs and limits. And now they change back to 4 threads max.
Transactional memory doesn't fix the potential quadratic scaling problem of adding more cores. Those are very independent issues.
Those are not mainstream CPUs, which is the topic for this thread. Most people are not prepared to pay an "Xeon-price tag" to have 8 cores. If there was only a $100 difference to get 8 cores, things would be different...
So we are back to wanting it, but not pay for it.
The 100 Mhz translates to a 100 Mhz/3400 Mhz = 3% performance increase. Do you really think that's why people are prepared to pay $100 more?
We're back to the chicken and egg scenario again. And no, we don't have mainstream CPUs that are capable of 8 "true" cores/threads.
The problem is that SW is optimized for the current mainstream Intel CPUs, which accounts for 80-90% of the market. Currently that means 4 true cores. So there is no point for AMD to lead the way and introduce 8 cores on it's own, since SW won't follow until the mainstream Intel CPUs have 8 cores as well. Also, note that AMD is one node shrink behind Intel, so they only have 50% the amount of transistors per die area which makes it harder for them to introduce true 8 core CPUs (they have to consider iGPU and TDP as well to match Intel).