Intel Broadwell Thread

[Sweepr]

More Core i7-6950X tests: https://news.xfastest.com/intel/20138/intel-broadwell-e-i7-6950x-cpu-benchmark-test

 

[dark zero]

I am most disappointed with the last of sub-$999 8C options. With the new 10C EE, was hoping 8C would move downmarket to ~$600 or so. Unless you are building a new system for X99, or absolutely want (need) 10C, there isn't much here that would entice HW-E users to move over. The options are essentially the same as HW-E CPUs with an added SKU at $1599.

Intel just gave AMD the chance to comeback with Zen. If AMD solution cost only 599 for octacores, 399 for Hexas and just 249 for Quads (when gets released), Intel will suffer hard on HEDT

 

[Arachnotronic]

Intel just gave AMD the chance to comeback with Zen. If AMD solution cost only 599 for octacores, 399 for Hexas and just 249 for Quads (when gets released), Intel will suffer hard on HEDT

Intel can cut prices if it needs to.

 

[dark zero]

Intel can cut prices if it needs to.

Not too much. Also the MoBo prices hurts them hard.

 

[csbin]

http://www.overclock3d.net/articles/cpu_mainboard/asus_rog_rampage_v_edition_10_has_leaked_online/1

ASUS' ROG Rampage V Edition 10 has leaked online

 

[Lepton87]

Strange aesthetic for an ROG board. It seems it only differs from my by the inclusion of new ports like that U.2 which I only heard about recently but still didn't read what it does. All in all for the owners of the existing ROG X99 it would be wasteful to "upgrade". I was existed about broadwell at first, because I thought they would keep the price of flagship close to 1000$ and compress the pricing below but 1000$ for an 8 core? Hell no. You can buy an 8 core for 70$ on the used market only SB-E but it still is competitive if someone doesn't OC.

 

[IntelUser2000]

More Core i7-6950X tests: https://news.xfastest.com/intel/20138/intel-broadwell-e-i7-6950x-cpu-benchmark-test

These are *surprising* scores. In R11.5 the 6950X is 41.6% faster and in R15 its 44% faster. Other sites show better scores for 5960X but the difference is still in the 35% range.

That means its anywhere from 10-20% better with same number of cores.

Other sites are claiming the advantage is only 20% but that's comparing against Xeon E3 1680 v3, which has higher clocks than 5960X. You'll see from 5960X reviews that 5960X is slower. The difference between the 1680 and the 5960X is almost 10%. 6950X is 20% on top of that.

The possibility is that it has higher multi-core Turbo. That means at least at stock, the pricing is easier to swallow.

Update: The $587 6850K gets 1311 in Cinebench R15, or almost on par with the 5960X. So its essentially the same as Haswell-E versus Ivy Bridge-E. The pricing moves down one stack, and the performance at same price increases by 20%. Only this time, you get even faster one with 40% over previous generation.

Now if they are able make top OC clock 10% better than with Haswell-E(since base clocks are 10% higher), we'd be at the same spot.

So don't feel bad. You are still paying too much, but not more than you were with Haswell-E.

 

[IntelUser2000]

Intel just gave AMD the chance to comeback with Zen. If AMD solution cost only 599 for octacores, 399 for Hexas and just 249 for Quads (when gets released), Intel will suffer hard on HEDT

Let's analyze this:

Assuming Cinebench R15 gives us rough comparison of both chips.

5960X: 1337(AT result)
FX-9590: 728

6950X: 1852

According to another info the $587 6850K gets 1311. That means the 8 core Zen would have to be 80% faster to equal the 6850K.

They would need to price it lower, not just because of performance. AMD vs Nvidia for video cards have shown despite AMD having very good perf/$, Nvidia sells better. Likewise in the CPU world, Intel has better brand recognition plus fastest top chip.

 

[The Stilt]

Let's analyze this:

Assuming Cinebench R15 gives us rough comparison of both chips.

5960X: 1337(AT result)
FX-9590: 728

6950X: 1852

According to another info the $587 6850K gets 1311. That means the 8 core Zen would have to be 80% faster to equal the 6850K.

They would need to price it lower, not just because of performance. AMD vs Nvidia for video cards have shown despite AMD having very good perf/$, Nvidia sells better. Likewise in the CPU world, Intel has better brand recognition plus fastest top chip.

If Zen would have exactly 40% higher IPC than Excavator in Cinebench R15, it would need to be able to maintain 3700MHz - 4200MHz (25 - 10% SMT efficiency) clocks on all cores in order to match 6850K score.

Unless it has significantly higher IPC in Cinebench R15 it won't happen because of the TDP alone, not to mention that it is currently unknown if the design is even able to hit frequencies this high.

 

[IntelUser2000]

If Zen would have exactly 40% higher IPC than Excavator in Cinebench R15, it would need to be able to maintain 3700MHz - 4200MHz (25 - 10% SMT efficiency) clocks on all cores in order to match 6850K score.

Unless it has significantly higher IPC in Cinebench R15 it won't happen because of the TDP alone, not to mention that it is currently unknown if the design is even able to hit frequencies this high.

Bear in mind, that's the 220W 9590 I compared. The 8350 gets 640 points, meaning 1311 is 104.8% better.

 

[The Stilt]

Bear in mind, that's the 220W 9590 I compared. The 8350 gets 640 points, meaning 1311 is 104.8% better.

I calculated it based on Excavator figures.

 

[AtenRa]

Comparing Piledriver to ZEN in Cinebench is pointless at this time.
Not to mention comparing Intel vs PD and ZEN in CB is not the benchmark to draw conclusions.

 

[IntelUser2000]

I calculated it based on Excavator figures.

Ah, forget about that.

Comparing Piledriver to ZEN in Cinebench is pointless at this time.
Not to mention comparing Intel vs PD and ZEN in CB is not the benchmark to draw conclusions.

That is true.

Advancements happen fast, but not all at once. Especially at the last leg of Moore's Law where every issue regarding MPU scaling is creeping up to you at once, expecting a competitor so far behind to suddenly catch up is ever less likely to happen.

It's like watching an Olympic athlete grow. His advancement is very fast in the beginning. You have "average" Grade 8 kids achieving 14 second 100 meter run. And with vigorous training, by 18-20 he might do 10.5 seconds. But the world record is 9.58 seconds. Tens of milliseconds determine the winner and the 5th place.

 

[RussianSensation]

Let's analyze this:

Assuming Cinebench R15 gives us rough comparison of both chips.

5960X: 1337(AT result)
FX-9590: 728

6950X: 1852

According to another info the $587 6850K gets 1311. That means the 8 core Zen would have to be 80% faster to equal the 6850K.

They would need to price it lower, not just because of performance. AMD vs Nvidia for video cards have shown despite AMD having very good perf/$, Nvidia sells better. Likewise in the CPU world, Intel has better brand recognition plus fastest top chip.

Luckily, most of us don't play Cinebench 11.5 or 15. Ya, I run them for fun but I really couldn't care less if an i7 6950X OC would destroy an i7 6700K/i7 7700K in this test. At the end of the day, if an 6-8 core CPU cannot take down a 4-core CPU for games, while using 100-150W more platform power, it's not impressive in the slightest. Blame it on current game development, but it's the reality. The reality will dictate that i7 6700K OC will beat every single BW-E SKU OC in games, while using less power. It's the primary reason why most people skip the X99 platform as there isn't enough of an incentive for them to increase their power usage without a tangible benefit in performance for 95% of apps/programs they run. This is actually a dangerous case for Zen as well as they are trying to push 8-cores down our throats but each of those cores will only be 40% faster than today's slow Vishera cores. Hard to be excited about that either.

Unless BW-E can overclock beyond 4.8Ghz, it will still lose to Skylake/Kaby Lake i7 in games. By the time SKL-E comes out, a year out we'll have an even superior architecture with Icelake. I also don't like that fact that SLK architecture will be almost 2 years old by the time SLK-E SKUs show up in 2017. It feels a bit strange to pay cutting edge prices in 2017 for a 2015 CPU architecture no matter how good it is. It's going to be hard to try to sell a 6-10 core BW-E to a gamer when a $320 MicroCenter i7 6700K, soon to be replaced by an i7 7700K, will still be a faster and more efficient gaming CPU. From now on, it will be a trade-off between the fastest gaming system you can get on the mainstream side and a workstation system that can be used for games + anything else.

Where Zen can come in is target Intel's $350 and below space because that's where they have the weak i3 and i5 non-K CPUs. Also, don't forget that AMD's motherboards have traditionally been much cheaper.

AMD already stated repeatedly that Zen is not a one shot wonder and no one objective truly expects Zen to outperform Intel's Broadwell/Skylake in single core performance. It's in our best interests though for AMD to do well in this space because it could push Intel to innovate more or deliver more cores at lower prices.

What's a bit ironic is that with the arrival of DX12, even more priority will be shifted towards the GPU side, suggesting that spending anything above $350 for a CPU is going to become an even greater case of diminishing returns.

I was existed about broadwell at first, because I thought they would keep the price of flagship close to 1000$ and compress the pricing below but 1000$ for an 8 core?

You already have a very good CPU. Unless you are using the PC for encoding, rendering and other productivity tasks that need as much CPU power as you can throw at them, chances are even SKL-E won't be a good enough upgrade for you.

I wonder how well BW-E will overclock. So far I only briefly saw results of it hitting 4.5Ghz which is no better than HW-E.

 

[SAAA]

I still don't understand how Broadwell-E looks like a >10% better IPC than Haswel when the desktop/mobile variant were a mere 3% (pls. ignore eDRAM models for this purpose).
What happened with the new uncore? Did they redesign something in the cores themselves? That % should be Skylake territory... no actually it might be better!?

Can someone test this (pretty please?): fixed clock for both platforms and single thread results, even better would be 4 core, 8 thread scores with the rest disabled so the bigger L3 cache doesn't matter much.

 

[IntelUser2000]

I still don't understand how Broadwell-E looks like a >10% better IPC than Haswel when the desktop/mobile variant were a mere 3% (pls. ignore eDRAM models for this purpose).

It probably doesn't perform that good per clock. Rather, its a case of higher Turbo clocks. 5960X has 3.2GHz 8 core Turbo. 6950X may reach 3.5GHz, or maximum its capable of. 5960X reach 3.5GHz maximum only with less than 8 cores.

By the time SKL-E comes out, a year out we'll have an even superior architecture with Icelake. I also don't like that fact that SLK architecture will be almost 2 years old by the time SLK-E SKUs show up in 2017. It feels a bit strange to pay cutting edge prices in 2017 for a 2015 CPU architecture no matter how good it is.

We wish it otherwise but this is what's probably going to happen.

Computex 2016: Broadwell-E
November 2016: Kabylake
Back to School 2017: Skylake-E
January 2018: Cannonlake

Now, from here they reduce the gap to pre-Broadwell levels. It also tells how rushed of a development Kabylake is. Perhaps the server focus is also why the HEDT series has such a gap. The HEDT has everything that a good scalable server chip needs, but most are useless for PC.

October 2018: Cannonlake-E
Spring 2019: Icelake
December 2019: Icelake-E

You'll see I added more than 12 months per release. That's been true historically. I doubt it'll change. Broadwell-E was supposed to be Q1, not late Q2. It's not going to happen with SKL-E, but hopefully the Optane DIMM support comes to the -E platforms, which would start justifying its existence. I am saying not SKL-E, because the Optane DIMM seems to be something that's included in the memory expander buffers the MP platforms have.

 

[Lepton87]

You already have a very good CPU. Unless you are using the PC for encoding, rendering and other productivity tasks that need as much CPU power as you can throw at them, chances are even SKL-E won't be a good enough upgrade for you.

I wonder how well BW-E will overclock. So far I only briefly saw results of it hitting 4.5Ghz which is no better than HW-E.

The only reason I wanted to upgrade is that I got a bad overclocker but even so I don't even notice the difference between running it overclocked and at "MCE" stock. 3.6GHz and 4.2GHz is not that big of a difference. If I won't be bother by the lack of mobo features or too little frames per seconds in games I might wait with upgrading till we have such terrific deals for X99 CPUs as we do now for SB-E Xeons with the 8-core models going for just 5% of their sticker price.

 

[ZGR]

I've been wanting more single-threaded performance for awhile. I know the upgrade is pretty poor, but I did it. I paid $330 (shipping included) for the i7-5775C! I plan on selling my current i5 soon after installing the new CPU. I've been wanting the 5775c for a long, long time but at $400+ it was a hard chip to swallow.

I hope I can hit at least 4.2 GHz....

The games I play are all DX9 mostly; RTS games that load all the AI on 1 core. I'm hoping the L4 cache will help a lot here. At the end game of Sins of A Solar Empire + Crazy Mod with ~16000 AI, the game putters around at 8-12 fps. If I gain ~3 fps, I will be happy. On newer games, my i5-4690k is a definite bottleneck in The Division and GTA V.

I hope Broadwell C and Z97 treat me well for the next 3 or so years. I'll be getting the chip probably next week or so on economy shipping.

 

[LTC8K6]

I've been wanting more single-threaded performance for awhile. I know the upgrade is pretty poor, but I did it. I paid $330 (shipping included) for the i7-5775C! I plan on selling my current i5 soon after installing the new CPU. I've been wanting the 5775c for a long, long time but at $400+ it was a hard chip to swallow.

I hope I can hit at least 4.2 GHz....

The games I play are all DX9 mostly; RTS games that load all the AI on 1 core. I'm hoping the L4 cache will help a lot here. At the end game of Sins of A Solar Empire + Crazy Mod with ~16000 AI, the game putters around at 8-12 fps. If I gain ~3 fps, I will be happy. On newer games, my i5-4690k is a definite bottleneck in The Division and GTA V.

I hope Broadwell C and Z97 treat me well for the next 3 or so years. I'll be getting the chip probably next week or so on economy shipping.

Tough choice against a 4790K though.

 

[dark zero]

Let's analyze this:

Assuming Cinebench R15 gives us rough comparison of both chips.

5960X: 1337(AT result)
FX-9590: 728

6950X: 1852

According to another info the $587 6850K gets 1311. That means the 8 core Zen would have to be 80% faster to equal the 6850K.

They would need to price it lower, not just because of performance. AMD vs Nvidia for video cards have shown despite AMD having very good perf/$, Nvidia sells better. Likewise in the CPU world, Intel has better brand recognition plus fastest top chip.

Err.. The Intel X Core (Haswell) has 16 threads... The FX only 8.

Also. Zen is gonna correct the punishment of CMT. So there is 20%. It will have HT. Being very optimistic it will be another 20%. Die shrink can help a lot. So there is 20%. Finally, but not the least. The IPC improvement can go around 25%. So expecting going from that 728 to 1572. Not as great as Broaswell-E, but near enough to be competitive again.

 

[The Stilt]

Err.. The Intel X Core (Haswell) has 16 threads... The FX only 8.

Also. Zen is gonna correct the punishment of CMT. So there is 20%. It will have HT. Being very optimistic it will be another 20%. Die shrink can help a lot. So there is 20%. Finally, but not the least. The IPC improvement can go around 25%. So expecting going from that 728 to 1572. Not as great as Broaswell-E, but near enough to be competitive again.

Excavator at 3400MHz scores 86 in ST. Multiply that with "up to" 1.4, and then by eight. On top of that you can add 10 - 25% for SMT.

Still off, but much much closer than you hyperoptimistic dream.

 

[DrMrLordX]

Excavator at 3400MHz scores 86 in ST. Multiply that with "up to" 1.4, and then by eight. On top of that you can add 10 - 25% for SMT.

If that's all we get then Zen is toast. 8m XV would have been better.

 

[The Stilt]

If that's all we get then Zen is toast. 8m XV would have been better.

From where would you expect to gain more, since up to 40% higher IPC is what AMD themselves expect Zen to have over Excavator?

With previous BD iteration (particularly SR & XV) AMD's own expectations regarding the IPC improvement were extremely well matched to the average difference we could replicate using various workloads. Sure there were cases where the IPC improvement was higher or lower, but the average was almost exactly the figure AMD themselves said it would be.

I see no reason why it wouldn't be the case with Zen too?

If Zen can match the average IPC of Ivy Bridge, reach >= 3.6GHz without destroying the power efficiency and provide continuity (further IPC, frequency and power improvements in Zen+) then I'm personally ready to call it as a success.

 

[Abwx]

Excavator at 3400MHz scores 86 in ST. Multiply that with "up to" 1.4, and then by eight. On top of that you can add 10 - 25% for SMT.

Still off, but much much closer than you hyperoptimistic dream.

So according to you this FPU, that score 155 when loaded with two threads and wich is recycled by Zen, will score 120 when loaded by a single thread in a Zen core..?..

To get to 155 they would need 29% gain with SMT, wich is quite high for Cinebench, and this is without taking account that they improved this FPU, one has to wonder why since you are stating that it will be used below the usage %age in Excavator...:sneaky:

To summarize you are stating that a Zen core will have less FP throughput than a EXV module...

 

[Sweepr]

Coolaler got 4.2 GHz out of his Core i7-6950X:

www.coolaler.com/showthread.php/336635

Insane MT performance. To put it in perspective: