[TheReg] Intel Management Engine: Is this why we have seen slower speed growth?

[ShintaiDK]

If we look on other x86 makers. Specially AMD due to size. The IPC gains there since K8 have been very small. And that includes time where they had money.

It seems to revert to the term that people have problems accepting. The performance/watt.

 

[maddie]

If we look on other x86 makers. Specially AMD due to size. The IPC gains there since K8 have been very small. And that includes time where they had money.

It seems to revert to the term that people have problems accepting. The performance/watt.

Using this example, can we use AMD's case as an argument that large IPC gains are a thing of the past?

Why is it that Intel is given a free pass?

May be the wrong person to ask, but can you say for certain that large IPC gains are no more, or are just inferring this claim based on Intel recent history of small IPC gains per generation?

 

[Fjodor2001]

Whatever helps you refrain from taking personal accountability.

Accountability of what? We're trying to have a discussion here. The only thing you've done is pulling a straw man and then when that failed you resorted to insults and personal attacks. If you've got something to add to the subject then do that instead of just derailing the thread as usual.

 

[ShintaiDK]

Using this example, can we use AMD's case as an argument that large IPC gains are a thing of the past?

Why is it that Intel is given a free pass?

May be the wrong person to ask, but can you say for certain that large IPC gains are no more, or are just inferring this claim based on Intel recent history of small IPC gains per generation?

All the data points to its part of the past. Unless you can show otherwise?

Power consumption cant be driven like previous. Where CPUs from uarch to uarch could go from 20-30W to 40-50W, then 65-75W, then 95W and finally 125-140W over the years. If we extrapolate even the FX9590 would be in the low end.

 

[Fjodor2001]

Power consumption cant be driven like previous. Where CPUs from uarch to uarch could go from 20-30W to 40-50W, then 65-75W, then 95W and finally 125-140W over the years. If we extrapolate even the FX9590 would be in the low end.

The problem is that increasing TDP does not allow for the frequency to be increased as much as before:

The thing is that previously you paid a much less TDP penalty per frequency increase than today. E.g. in ~2 years time you could often double the frequency at only 30% or so higher TDP (given the benefit of a later node too).

Do you think Intel could produce a 4790K replacement ~2 years after it was introduced that was running at 8/8.8 GHz instead of 4/4.4 GHz? At only 30% higher TDP, so 88*1.3=114 W? I don't think so. No matter if there was much higher consumer demand for faster desktop CPUs and it was the primary focus for Intel. Not even close.

 

[ShintaiDK]

Or IPC increase.

 

[MongGrel]

The TCO cost is also why people can find very cheap older generation Xeons as upgrades for their desktop. Compared to relatively expensive equal desktop parts.

Yeah, played around with the X5680 in here on the main the other day and it's running pretty nicely @ 4.7 now on 6 X 12.

I hadn't messed with pushing it up in awhile.

 

[mrmt]

Using this example, can we use AMD's case as an argument that large IPC gains are a thing of the past?

Why is it that Intel is given a free pass?

Certainly not only Intel, but why don't add SUN and IBM to the mix? Are they improving IPC of their processors at extremely high rates or are they betting on efficiency and adding more core/threads and better interconnects for their new processors?

The only ones adding a lot of extra IPC each iteration is ARM, but they are coming from a very low starting point and have a lot of low hanging fruits to pick.

The thing is, all these IHVs/IDMs aren't just betting on better CPU cores for these workloads, but also on GPGPU, FPGAs and dedicated accelerators for a given workload and the reason for that is that they yield better results than to extract more IPC from already very mature core designs.

 

[SarahKerrigan]

Certainly not only Intel, but why don't add SUN and IBM to the mix? Are they improving IPC of their processors at extremely high rates or are they betting on efficiency and adding more core/threads and better interconnects for their new processors?

Power8 has significantly higher single-threaded IPC than Power7 did. I've seen tens of percent improvement myself at equivalent clock on non-synthetic loads, which is generally at least close to IBM's claims.

Oracle did a tremendous IPC leap when they released the T4, but that's more due to finally getting a modern OoO core than some kind of multi-generation strategy.

Fujitsu continues to get significant gains at equivalent thread count for SPARC64, although they have a slower release cadence than Intel.

 

[mrmt]

Power8 has significantly higher single-threaded IPC than Power7 did. I've seen tens of percent improvement myself at equivalent clock on non-synthetic loads, which is generally at least close to IBM's claims.

Can these workloads make good use of the beefier L3 and L4 caches?

 

[SarahKerrigan]

Can these workloads make good use of the beefier L3 and L4 caches?

Core-local cache isn't beefier in P8 - it dropped to 8MB/core from 10MB/core in P7+.

i'm not sure what effect the Centaur L4 has, but i suspect the answer is "not much." Memory latency is still pretty high.

 

[mrmt]

Core-local cache isn't beefier in P8 - it dropped to 8MB/core from 10MB/core in P7+.

i'm not sure what effect the Centaur L4 has, but i suspect the answer is "not much." Memory latency is still pretty high.

IBM recently lost a bid in one of our subsidiaries for a SAP ECC server because E7 Xeons were beating POWER8 chips in terms of TCO and even raw performance, in the main company SAP runs on SPARC.

Some friends of mine were talking about financial applications where this L4 cache makes a lot of difference but I don't have any first hand info about it. My uneducated guess is that this L4 cache somehow makes a significant difference in quite a few performance cases, if not what's the point in making memory management even more complex? At least it should have very improved latencies when compared to main memory.

 

[Rakehellion]

Ignorance is a bliss? You do know if people and companies dont upgrade, Intel goes bankrupt very fast?

Duh. The point is they're upgrading to something that's pretty much the same as what they had last year.

And your average business is running a room full of Pentium IIIs that will sit on the shelf for years.

 

[Rakehellion]

I don't think it's greedy at all, considering how much they've exploited populations.

Have they exploited you? No? Then you're being greedy.

 

[Rakehellion]

It seems that some are only thinking in opposites and not being able to see a range of possibilities.

Its either Intel is purposely slowing down development or IPC gains are almost impossible to extract.
What about benign neglect due to it not being a very high priority? Just enough of a gain between generations to keep sales up but not enough to wow the customer.

Exactly.

The thing that keeps the market going is Intel selling you more than you need. Like buying a Lamborghini to commute to work every morning. Eventually the software will catch up to. Overpowered chips are harder to make now. Overpowered chips aren't popular anymore. What people want is small, thin, light.

If the 6700K is too underwhelming you can always pay $6000 for a 15-core Xeon. What? And I thought you called yourself an enthusiast.

 

[Arachnotronic]

If the 6700K is too underwhelming you can always pay $6000 for a 15-core Xeon. What? And I thought you called yourself an enthusiast.

Pfftt....$4859.95 buys you an 18 core beastie!

http://www.newegg.com/Product/Produ...cm_re=xeon_e5_2699-_-9SIA75M3GF8887-_-Product

 

[Rakehellion]

Pfftt....$4859.95 buys you an 18 core beastie!

http://www.newegg.com/Product/Produ...cm_re=xeon_e5_2699-_-9SIA75M3GF8887-_-Product

Yeah, but the 15-core chip has better single-threaded performance which you need for playing Minesweeper.

On an unrelated note, I wonder why I hear of people in the forums bragging about their Titan SLI setup but I've never once heard of anyone with 12+ cores on their desktop PC.

 

[mrmt]

Pfftt....$4859.95 buys you an 18 core beastie!

http://www.newegg.com/Product/Produ...cm_re=xeon_e5_2699-_-9SIA75M3GF8887-_-Product

http://download.sap.com/download.ep...218166335A6011727316427E851ED72C285D42D4F5BFE

Lenovo System x3950 X6, 8 processors / 144 cores / 288 threads,
Intel Xeon Processor E7-8890 v3, 2.50 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 45 MB L3 cache per processor,
2048 GB main memory

 

[Arachnotronic]

http://download.sap.com/download.ep...218166335A6011727316427E851ED72C285D42D4F5BFE

Lenovo System x3950 X6, 8 processors / 144 cores / 288 threads,
Intel Xeon Processor E7-8890 v3, 2.50 GHz, 64 KB L1 cache and 256 KB L2 cache per core, 45 MB L3 cache per processor,
2048 GB main memory

Can it raytrace Crysis?

 

[Arachnotronic]

Yeah, but the 15-core chip has better single-threaded performance which you need for playing Minesweeper.

On an unrelated note, I wonder why I hear of people in the forums bragging about their Titan SLI setup but I've never once heard of anyone with 12+ cores on their desktop PC.

You can't buy a 12 core CPU today that can overclock to ~4.5GHz. For games, it would be massive suck compared to say a 5820K @ 4.5GHz.

 

[ShintaiDK]

12 cores at 4.5Ghz, that would set a new record in power consumption and LN2 usage

 

[ehume]

Doesn't anyone have anything else to say about the IME?

 

[jhu]

Doesn't anyone have anything else to say about the IME?

How does it differ from AMD's version?

 

[jhu]

Using this example, can we use AMD's case as an argument that large IPC gains are a thing of the past?

Why is it that Intel is given a free pass?

May be the wrong person to ask, but can you say for certain that large IPC gains are no more, or are just inferring this claim based on Intel recent history of small IPC gains per generation?

These small IPC gains aren't recent. Here are where the large IPC gains happened:

1) 8086 to 80286
2) 80286 to 80386
3) 80386 to 80486
4) 80486 to Pentium
5) Pentium to Pentium Pro

IPC increases since Pentium Pro have been fairly small. So essentially we've had 20 years of small IPC gains.

 

[ehume]

How does it differ from AMD's version?

I don't know. Perhaps you or someone else would elucidate. And how much of a performance hit do we see in AMD and Intel's ME?