Intel Skylake / Kaby Lake

[Sweepr]

Pentium G4560 really is a little gem:

 

[Sweepr]

ComputerBase's Core i7-7700T Review

www.computerbase.de/2017-01/intel-core-i7-7700t-test-kaby-lake

35W TDP desktop CPUs are becoming very capable. 91% / 99% Core i7-6700K performance in applications / games at a fraction of the power.

 

[Jean-Sebastien Blais]

On my Asus Strix Z270G the BCLC has settings from 40 to 650 mhz. With my i5 6500 I can only raise it to 102.9. 103 will not post. Maybe Asrock did better than Asus.

u have to disable turbo boost and c-state in your asus bios in order for block oc to work

 

[Conroe]

u have to disable turbo boost and c-state in your asus bios in order for block oc to work

I've disabled everything and 103 is still a no go.

edit...
But it does work with my 7700k I just installed.

 

[IntelUser2000]

Yep. Cheap chips for "developing markets", but soak the USA customers with more expensive chips. Same MO that Big Pharma uses, with medications half the price being sold in Canada.

I don't think this is the case. Developing markets are perfectly capable of buying more than Pentium chips. 2011 was the year when Sandy Bridge brought big success for Intel. That year, you started to feel it. People around you bought Sandy Bridge systems. I saw a documentary of a person in a third world Asian country working for those nasty, low paid industrial age jobs. Then they showed you the picture of their living room. And guess what? I noticed a 2nd Gen Core sticker on the desktop! A person living in a poor third world country working a crappy job had the latest, greatest computer that year.

I am in Canada. In the first few weeks of release, Newegg and NCIX had Pentium G4560 chips available. Now, its all backordered. It may be that its actually that popular. Certainly, I see all kinds of Kaby Lake chips available to buy. Just not G4560. And few models are already offering discounts. Some discounts make sense. Like the i3 7100, 7350K. 7350K is too expensive, while 7100 mostly got usurped by the much cheaper G4560. None of the Pentium chips are offering discounts. Above $150, most models are just frequency increases, and people who would buy them from Newegg likely buys the lowest frequency part offering all the features they want, since the difference in clock is minimal.

I have an urge to replace my 2600K system with a Kaby Lake one, assuming Optane Memory arrives and it offers benefits they claim it does and its affordable(say $50 for 16GB, and $89/99 for 32GB. Anything over doesn't make sense - since its a hard drive caching device). I would probably go with the i3 7100, because i3 offers AVX and while I value single threaded performance which KBL will offer 2600K, losing AVX with G4560 is not attractive. But I assume most people are different.

 

[Sweepr]

DigitalFoundry: Intel Kaby Lake: Core i3 7350K review
Is an overclockable dual-core CPU faster for games than an i5 quad?

We don't have the 3.0GHz Core i5 7400 to hand, but we do have the slightly faster Core i5 6500, built on the very similar Skylake architecture, and with the same 3.6GHz boost clock. The Core i3 is generally considered a budget option, so we've also stacked it up against a processor many consider to be the new budget king - the Pentium G4560. Like the i3, it has two cores and four threads, but its clock speeds are limited to 3.5GHz - but that's fine as it costs just 36 per cent of the retail price of the i3 7350K. We picked one up for £63, and it's pretty amazing for its price.

The single-thread king - Far Cry Primal - sees some excellent results for the i3. Whether in stock configuration or with the overclock in place, it handily beats the Core i5 6500. And across the board, some of the results at 4.8GHz see the i3 get very, very close to the locked Skylake i5. Ashes of the Singularity and Crysis 3 even record wins against the locked quad-core processor. However, Far Cry aside, there is no 'killer blow' for the i3 7350K. Essentially, we've pushed a dual-core, quad-thread chip to its limits, throwing power efficiency out of the window and requiring a solid thermal solution to provide mixed results against the i5 6500.

...Kaby Lake does indeed revolutionise the dual-core processor line, but it's the addition of hyper-threading to select Pentiums that proves to be the game-changer.

Similar to every Intel CPU we've tested, performance also scales in line with memory bandwidth - as this table demonstrates - so to get the most out of the processor, regardless of how far you overclock it (or not), there's extra frame-rate there for the taking by pairing the CPU with faster memory.

Rather than spend the extra money on a heavy duty thermal solution, you can actually save some cash by opting for a Core i5 7500 - which won't require any additional expense on a heatsink and fan and should outperform the Core i5 6500 we used for our locked quad comparisons in this article.

But the genuine surprise from our testing is just how impressive the Pentium G4560 performs bearing in mind its tiny price-point. Historically, Intel's Pentiums and Celerons have been limited to just two threads, leading to major in-game stutter on modern game engines - the main reason it's hard to recommend the overclockable Pentium G3258. This is not an issue with the hyper-threading enabled G4560 - initial results look highly promising and we'll be stacking it up against the other budget processor options in a full review soon.

www.eurogamer.net/articles/digitalfoundry-2017-intel-kaby-lake-core-i3-7350k-review

Aside from the Core i5 recommendation, DigitalFoundry says major in-game stutter on modern games is not an issue with the new budget king Pentium G4560. I have little doubt Intel will be offering quad-core Core i3 next with Coffee Lake-S.

 

[frozentundra123456]

DigitalFoundry: Intel Kaby Lake: Core i3 7350K review
Is an overclockable dual-core CPU faster for games than an i5 quad?

www.eurogamer.net/articles/digitalfoundry-2017-intel-kaby-lake-core-i3-7350k-review

Aside from the Core i5 recommendation, DigitalFoundry says major in-game stutter on modern games is not an issue with the new budget king Pentium G4560. I have little doubt Intel will be offering quad-core Core i3 next with Coffee Lake-S.

Just look at that memory scaling. Granted it varies from game to game, but the average improvement from 2133 to3000 mgz memory is almost 11%. And some posters in these forums are still quoting Anand's lame skylake test with gimped memory to show how little skylake improved over haswell/broadwell.

 

[majord]

ComputerBase's Core i7-7700T Review

www.computerbase.de/2017-01/intel-core-i7-7700t-test-kaby-lake

35W TDP desktop CPUs are becoming very capable. 91% / 99% Core i7-6700K performance in applications / games at a fraction of the power.

Except it's not at a fraction of the power as evidenced by the power consumption graphs.

7700T uses more power than a 6600K and 7600K under CPU only load, but is only 2% faster. Only when the IGP+CPU are stressed simultaneously would you see any tangible difference.

If anything it shows how efficient the HT-less models are .

-edit. Also, HWinfo presents a perfectly usable , cpu core and package power readings from the Intel PMU itself, why these reviewers don't graph this along side their physical testing for processors that support it, I don't know. it would make it easier to assess.

 

[IntelUser2000]

7700T uses more power than a 6600K and 7600K under CPU only load, but is only 2% faster. Only when the IGP+CPU are stressed simultaneously would you see any tangible difference.

If anything it shows how efficient the HT-less models are .

I think this is good evidence showing the effect of binning is in practice, dead. Earlier on in CPU development there was much for everyone involved in the project to learn and of course some results will be drastically better. Binning therefore, had significant difference since it was much far away from perfection.

But nowadays its at the apex of development in both design and process and binning brings almost no difference. The lower TDP models are there just for a guarantee, nothing else. Now you need more of a complete overhaul just to get the same effect you did previously. No, even that was the past. Now you need more work to achieve the same thing. Remember what Nvidia said with Pascal? They overhauled the circuitry to optimize for frequency. A decade ago, that kind of frequency gains came with just porting onto a new process. I am grossly oversimplifying here but the kind of work that went on Pascal would bring far more than Pascal would have brought if it was 10 years ago.

 

[Dave2150]

Just look at that memory scaling. Granted it varies from game to game, but the average improvement from 2133 to3000 mgz memory is almost 11%. And some posters in these forums are still quoting Anand's lame skylake test with gimped memory to show how little skylake improved over haswell/broadwell.

I think the writer of the Skylake review was just clueless to the memory speed scaling, Anandtech's reviews used to be best in class, now they are pretty terrible. And that's if they even release a review, still waiting for the Polaris full review/architecture dive etc....

 

[Sweepr]

Meet Skylake-EP - Intel(R) Xeon(R) Gold

2x Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz (18C 36T 3.7GHz, 2.4GHz IMC, 18x 1MB L2, 24.75MB L3)
- Processor Multi-Media: 4040.67Mpix/s
- Processor Multi-Core Efficiency: 168.96GB/s

New cache structure in place. This is a 18C/36T 2.7 GHz (3.7 GHz Turbo) model in 2P configuration.

 

[frozentundra123456]

I think the writer of the Skylake review was just clueless to the memory speed scaling, Anandtech's reviews used to be best in class, now they are pretty terrible. And that's if they even release a review, still waiting for the Polaris full review/architecture dive etc....

True, Anand is no where close to the review site it used to be. More annoying though is the few posters who continuously quote that flawed, very initial test to prove how little improvement there was, with no mention (or even willingness to acknowledge it when other bring it up) that Skylake performs a lot better with fast memory.

 

[John Carmack]

So that review basically mirrors what people have been saying on the forum for the past couple of weeks: The new Pentium is a bargain, the 7350K is a waste of time?

 

[Sweepr]

Meet Skylake-EP - Intel(R) Xeon(R) Gold

2x Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz (18C 36T 3.7GHz, 2.4GHz IMC, 18x 1MB L2, 24.75MB L3)
- Processor Multi-Media: 4040.67Mpix/s
- Processor Multi-Core Efficiency: 168.96GB/s

New cache structure in place. This is a 18C/36T 2.7 GHz (3.7 GHz Turbo) model in 2P configuration.

Quick comparison, here's the highest score for 18C/36T Broadwell-EP @ 2.3-3.6 GHz:

2x Intel(R) Xeon(R) CPU E5-2697 v4 @ 2.30GHz (18C 36T 3.59GHz, 3.6GHz IMC, 18x 256kB L2, 45MB L3)
- Processor Multi- Media: 2099.19Mpix/s 72T

Skylake-EP almost doubles the multi-media score. I'm assuming AVX-512 is at play here.

 

[FlanK3r]

7700K, my chip nr3

 

[nvgpu]

7350K is expensive compared to the G4560. Coffee Lake i3 really needs to be a 4C/4T chip for the price.

 

[Arachnotronic]

7350K is expensive compared to the G4560. Coffee Lake i3 really needs to be a 4C/4T chip for the price.

I think that's the most sensible option. i3 = 4C/4T, i5 = 4C/8T, i7 = 6C/12T.

 

[Arachnotronic]

Quick comparison, here's the highest score for 18C/36T Broadwell-EP @ 2.3-3.6 GHz:

2x Intel(R) Xeon(R) CPU E5-2697 v4 @ 2.30GHz (18C 36T 3.59GHz, 3.6GHz IMC, 18x 256kB L2, 45MB L3)
- Processor Multi- Media: 2099.19Mpix/s 72T

Skylake-EP almost doubles the multi-media score. I'm assuming AVX-512 is at play here.

Xeon Gold, nice. I wonder if the HEDT series will get the "gold" treatment as well.

 

[TheELF]

I think that's the most sensible option. i3 = 4C/4T, i5 = 4C/8T, i7 = 6C/12T.

Would be nice but very doubtful that intel will change their whole lineup.
It's more probable that everything stays as it is other then the top i7 model being a 6/12 one.

 

[coercitiv]

I think the writer of the Skylake review was just clueless to the memory speed scaling, Anandtech's reviews used to be best in class, now they are pretty terrible. And that's if they even release a review, still waiting for the Polaris full review/architecture dive etc....

The author of the review is anything but clueless, but I think it's time to face the music: at this time I have every reason to believe Anandtech is no longer in the (financial) position to properly review new CPUs & GPUs. Their reviews are still worthy of our full attention and including stock configuration memory is a good thing, but only as long as overclocked info is also present (both CPU and RAM).

 

[lopri]

Just look at that memory scaling. Granted it varies from game to game, but the average improvement from 2133 to3000 mgz memory is almost 11%. And some posters in these forums are still quoting Anand's lame skylake test with gimped memory to show how little skylake improved over haswell/broadwell.

By "gimped memory" you mean memory specs officially supported Intel?

I saw the memory scaling charts from Digital Foundry and I am not very impressed. First of all, there are only marginal cases of a couple of game that can be said to reasonably respond to faster memory. Second of all, the fact that a processor is more sensitive to memory bandwidth is not necessarily a positive (as many assume that to be such). Agena (Phenom I) scaled better with memory than Yorkfield (45nm C2Q), does that make Phenom I a better CPU? Not really. Lynnfield scaled better with memory than Sandy Bridge, does that mean Lynnfield has a better memory controller? No.

Memory scaling in and of itself does not say much in a vacuum. It could be because of unbalanced core design, unbalanced platform design, superior memory controller, sub-par cache design, etc. any of them can be a reason.

One way to get a better idea is to compare Haswell - Skylake - Kabylake chips on even footing. By even footing I mean same frequency, same cache size, same uncore frequency, and same memory bandwidth/latency under same CPU/memory ratio (divisor) . There are bandwidth overlap between DDR3 and DDR4, so while it may not be ideal it is still something to build on.

If I take some of the extreme memory scaling results posted by certain review by their face value, I can foresee some difficult-to-explain results, such as previous generation chip performing better than later generation chip with the same rated memory sticks due to later generation chip's performance is much more volatile with regard to bandwidth. But as it is now, only thing that appears valid is that there are a couple of games that disproportionately respond to memory bandwidth increase/decrease.

Looking closely to the numbers presented by the chart does not help, either.

Take Witcher 3. At stock frequency of 4.2 GHz, going from DDR3-2133 to DDR3-2400 nets you 9% performance increase. Yet at 4.8 GHz, the same memory upgrade only nets you 5% of performance increase. (isn't that strange that a lower clocked CPU, not a higher clocked one, gains more from higher clocked memory?) Going from DDR3-2400 to DDR3-3000, the 4.2 GHz CPU gains 6.5% but 4.8 GHz CPU gains 11% (!!). That is before I take into the theoretical bandwidth of 3 selected bandwidth points, DDR3-2133, DDR4-2400, and DDR4-3000. As you can see from the ratings, the gab between each memory setting is not consistent. The next higher grade past DDR4-2400 is DDR4-2666, not DDR4-3000. I cannot glean from theh table as to why DDR4-3000 instead of DDR4-2666 was chosen and how it was chosen. Was there bus overclock? That introduces a whole host of other factors that are distinct from memory bandwidth.

I did not dwell on it much but the numbers are simply too all over the places, and I do not think any meaningful analysis (e.g. where the bottleneck is) can come out of this data without further testing or disclosure. All I see from the above table is that, well, memory bandwidth does not do much in real world performance except for a few corner cases. As it has always been.

 

[Dave2150]

The author of the review is anything but clueless, but I think it's time to face the music: at this time I have every reason to believe Anandtech is no longer in the (financial) position to properly review new CPUs & GPUs. Their reviews are still worthy of our full attention and including stock configuration memory is a good thing, but only as long as overclocked info is also present (both CPU and RAM).

If the author of the review was knowledgeable, it wouldn't have cost much to include a few sentences to inform the reader that higher speed memory would have a dramatic effect on the benchmark numbers. The author failed to mention memory speed at all, even though all the Z170 boards QVL lists showed official support for 2400Mhz, 3000Mhz, ,3200Mhz kits right from release. It's simply ignorance.

 

[Sweepr]

MARCONI, the new Tier-0 system

• A3: finally, in July 2017, this system is planned to reach a total computational power of about 20Pflop/s utilizing future generation Intel Xeon processors (Sky Lakes).

www.hpc.cineca.it/hardware/marconi


Surprisingly you can still find ES of unreleased Intel CPUs on eBay. Someone sold a Skylake-EP ES for almost U$1.600 yesterday:

http://www.ebay.com/itm/Intel-Xeon-E5-26XX-V5/152402932924?autorefresh=true
http://www.ebay.com/itm/xeon-v5-269...5f7c16&pid=100011&rk=1&rkt=12&sd=152402932924

 

[nvgpu]

Truly a budget powerhouse the G4560.

 

[DrMrLordX]

One way to get a better idea is to compare Haswell - Skylake - Kabylake chips on even footing. By even footing I mean same frequency, same cache size, same uncore frequency, and same memory bandwidth/latency under same CPU/memory ratio (divisor) . There are bandwidth overlap between DDR3 and DDR4, so while it may not be ideal it is still something to build on.

X99 and the upcoming X299 would actually permit DDR4 testing on Haswell, Broadwell, and Skylake. Though Skylake-X may have "weird" cache configurations so I'm not sure the scaling results would be as useful as people would like.

Regardless that would iron out any difficulties imposed by comparing DDR3 and DDR4 numbers between microarchitectures.