IPC comparison with Edram

[Lepton87]

I'm looking for an IPC comparison for Intel CPUs with eDRAM and without. Especially I'm interested in the performance improvement from eDRAM in skylake compared to broadwell. eDRAM in skylake is supposed to be improved because that eDRAM cache in broadwell could only acts as a victim cache. I'm curious if there are any relevant performance differences in practice or if the victim cache is just fine.
AFAIR Zen also has a victim cache and that could be an important piece of information that would show if we should be concerned about that victim cache.

 

[Yuriman]

The article just published on the front page might help you with this question:

http://www.anandtech.com/show/10602/memory-frequency-scaling-on-skull-canyon

 

[Lepton87]

The article just published on the front page might help you with this question:

http://www.anandtech.com/show/10602/memory-frequency-scaling-on-skull-canyon

Unfortunately there are no comparisons to the regular skylake clocked at the same frequency.

 

[DrMrLordX]

You raise an interesting question, Lepton, and I second your request. We don't have much information about eDRAM-equipped Skylake (or Kabylake). It looks like eDRAM Skylake on the desktop is MIA for the foreseeable future, which is a pity.

 

[ZGR]

One problem is that Broadwell and Skylake each implement eDRAM rather differently.

Skylake is able to utilize the 128mb L4 in a few different ways then Broadwell.

One helpful thing is that both Broadwell and Skylake can disable eDRAM in the UEFI BIOS.

One thing I would like to see is a large amount of benchmarks comparing Broadwell and Skylake compared clock for clock with eDRAm disabled and enabled.

Using L4 as a victim cache to L3 is the primary reason I got the 5775C. Skylake can utilize the L4's massive bandwidth even more by having the L4 be far more accessible to pass data back and forth.

I doubt we will see an in depth eDRAM article. We will need to gather the data ourselves. I can easily enable/disable eDRAM and so can the other 5775C users on this forum. A problem is Skylake-H is locked down tight. You can only use Skylake + L4 in a BGA format

We will need to find someone with Skylake H who is willing to disable and enable eDRAM; along with getting a good benchmark suite to test the two platforms. I am a big eDRAM fan so I will definitely want to learn more.

 

[superstition]

Anandtech found the 5675C at stock outperforming stock Skylake in some games as I recall. The 5775C in particular was held back by TDP throttling.

However, for gaming, Broadwell C's EDRAM appears to equalize or even better Skylake, despite a clock deficit, in many games — even with the TDP throttling. So, the 5775C put in a good showing, too.

If Intel had released a Broadwell C with Iris disabled but EDRAM enabled at a high TDP...

Also:

http://arstechnica.com/gadgets/2015...obbing-us-of-the-performance-king-we-deserve/

in memory-intensive workloads, such as some games and scientific applications, the cache is better than 21 percent more clock speed and 40 percent more power. That's the kind of gain that doesn't come along very often in our dismal post-Moore's law world.

Those 5775C results tantalized us with the prospect of a comparable Skylake part. Pair that ginormous cache with Intel's latest-and-greatest core and raise the speed limit on the clock speed by giving it a 90-odd W power envelope, and one can't help but imagine that the result would be a fine processor for gaming and workstations alike.

Intel could have had a Skylake processor that was exciting to gamers and anyone else with performance-critical workloads. For the right task, that extra memory can do the work of a 20 percent overclock, without running anything out of spec. It would have been the must-have part for enthusiasts everywhere.

One site that tested the 5775C's EDRAM overclocking found that 2000 was optimal.

 

[ZGR]

I just want to post again regarding 5775C OC and eDRAM OC.

First off is the is the 65W Broadwell-C and Skylake-R TDP; a very real performance limitation. A few posted articles regarding Broadwell-C OC along with general performance metrics generally did NOT remove all the restrictions needed to unlock the TDP. Out of the box, these 65W CPU's throttle hard! Even after applying a traditional overclock, the CPU will still throttle at load for brief moments.

For example, this is just what XTU displays after overriding the listed variables:
Draw 4096W I dare you i7!

Then there are a few other power restrictions to be lifted in BIOS that are not shown. All in all, my 5775C is now a ~125W TDP CPU with the 4.2 GHz OC. With all the necessary power restrictions lifted, this CPU will never throttle at load. And yes, the CPU still idles at around ~0.850v when not loaded. This is NOT a fair comparison to the 65W locked Skylake-H (The improved eDRAM that we all want in an LGA socket). Who would want Broadwell-C if it was restricted to 65W and had no overclocking abilities? That is what Skylake-H is.

Next is eDRAM Overclocking! We need a comprehensive list of games, benchmarks, and various software that is heavily impacted by eDRAM and the LACK of it. Like I said in a post above, eDRAM is easily toggled on and off in the UEFI BIOS. Nearly every eDRAM overclocking article I have seen does not publish any data with the eDRAM off. I see that as a big red flag. I do believe that 1.9-2.1 GHz is the effective limit for eDRAM (latency is the big problem past 2.1 GHz), but I do not know the performance benefits with eDRAM on vs off in many situations. I want to review 7z, Winrar, and CPU limited game titles. I do not believe some canned benchmark like Cinebench to benefit from eDRAM. What I care about is when I am in CPU limited situations where the cache is at the limit. We need a lot more info.

I also want to know what software is impacted the most by eDRAM. It would be nice to have an "eDRAM friendly" benchmark to really compare further improvements to eDRAM in the future. I sure want to know how much Skylake-H has improved over Broadwell-C regarding the L4.

 

[NTMBK]

I just want to post again regarding 5775C OC and eDRAM OC.

First off is the is the 65W Broadwell-C and Skylake-R TDP; a very real performance limitation. A few posted articles regarding Broadwell-C OC along with general performance metrics generally did NOT remove all the restrictions needed to unlock the TDP. Out of the box, these 65W CPU's throttle hard! Even after applying a traditional overclock, the CPU will still throttle at load for brief moments.

For example, this is just what XTU displays after overriding the listed variables:
Draw 4096W I dare you i7!

Then there are a few other power restrictions to be lifted in BIOS that are not shown. All in all, my 5775C is now a ~125W TDP CPU with the 4.2 GHz OC. With all the necessary power restrictions lifted, this CPU will never throttle at load. And yes, the CPU still idles at around ~0.850v when not loaded. This is NOT a fair comparison to the 65W locked Skylake-H (The improved eDRAM that we all want in an LGA socket). Who would want Broadwell-C if it was restricted to 65W and had no overclocking abilities? That is what Skylake-H is.

Next is eDRAM Overclocking! We need a comprehensive list of games, benchmarks, and various software that is heavily impacted by eDRAM and the LACK of it. Like I said in a post above, eDRAM is easily toggled on and off in the UEFI BIOS. Nearly every eDRAM overclocking article I have seen does not publish any data with the eDRAM off. I see that as a big red flag. I do believe that 1.9-2.1 GHz is the effective limit for eDRAM (latency is the big problem past 2.1 GHz), but I do not know the performance benefits with eDRAM on vs off in many situations. I want to review 7z, Winrar, and CPU limited game titles. I do not believe some canned benchmark like Cinebench to benefit from eDRAM. What I care about is when I am in CPU limited situations where the cache is at the limit. We need a lot more info.

I also want to know what software is impacted the most by eDRAM. It would be nice to have an "eDRAM friendly" benchmark to really compare further improvements to eDRAM in the future. I sure want to know how much Skylake-H has improved over Broadwell-C regarding the L4.

Nice OC How far can the GPU be overclocked with power limitations lifted?

 

[ZGR]

Nice OC How far can the GPU be overclocked with power limitations lifted?

I haven't pushed it past 1250 MHz (1150 Mhz is stock) which is nothing interesting. I am only using the iGPU in OS X, where graphics performance is abysmal compared to Windows. I've been meaning to bench off the Iris Pro 6200 vs my old Nvidia GT 650M 1GB and compare them when they are at max OC.

With the 65W limitation, the 6200 will perform around an Nvidia Geforce GTX 660M (which is an overclocked 650M). Would be interesting to see what happens with the restriction removed...

edit: I believe 1400-1450 MHz to be the limit. No idea how well performance scales.

 

[Hi-Fi Man]

The eDRAM on Broadwell C can't actually overclock on most motherboards. If you change the multiplier on the eDRAM it actually disables the eDRAM even though it still may show as enabled. In addition, overclocking the core to non-even frequencies (4.1GHz) also disables the eDRAM. I've tested this myself using the AIDA64 cache bench.

 

[ZGR]

The eDRAM on Broadwell C can't actually overclock on most motherboards. If you change the multiplier on the eDRAM it actually disables the eDRAM even though it still may show as enabled. In addition, overclocking the core to non-even frequencies (4.1GHz) also disables the eDRAM. I've tested this myself using the AIDA64 cache bench.

This is alarming news to me! I went back to my BIOS and tried a few different configurations while testing AIDA64's Cache and Memory Benchmark.

I now can verify that my eDRAM is not being disabled!

When I first got my 5775C I ran the same benchmark with eDRAM disabled becauseof an outdated bios. Updating the bios added eDRAM support and I never bothered to check to see if disabling eDRAm actually disables it...

Changing the eDRAM multiplier works, regardless of my CPU frequency. Testing at 4.2, 4.1, 4.0, and 3.9 GHz shows my eDRAM to be working. The big problem is I now have no way to disable it unless I flash back to the older BIOS. Thanks Asus

Lowering eDRAM to 800 MHz has a large negative effect on L1, L2, and L3 cache latency. Going above 2.0 GHz increases L4 latency and lowers L4 bandwidth by a bit.

Why is eDRAM support so flaky?

 

[Roland00Address]

i7-6770HQ has ed ram, notice it is 770

i7-6700HQ has no ed ram, notice it is 700

Both i7 6770HQ and i7 6700HQ are 2.6 ghz base, 5 turbo bins higher for 3 or 4 cores (3.1 ghz), 7 bins higher for 2 cores (3.3 ghz), 9 bins higher for 1 core

The only difference is ed ram or no ed ram, and the i7 6770HQ has 72 EU while the i7 6700 has 24 EUs. But in CPU tasks that are minimal for the gpu in theory the higher amount of EUs would automatically dynamically shut off most of those extra gpu cores.


So here are the anandtech reviews with the i7 6770HQ (effectively it is only the nuc, and he tested the nuc with different rams)
http://www.anandtech.com/print/10602/memory-frequency-scaling-on-skull-canyon

Here are the anandtech reviews of the non ed ram i7 6700HQ
http://www.anandtech.com/print/10116/the-dell-xps-15-9550-review (note this review is newer so it has many of the same benches side by side with the y700 in the same chart)
http://www.anandtech.com/print/10026/the-lenovo-ideapad-y700-laptop-review

Or use bench to make it easier to compare
http://www.anandtech.com/bench/product/1631?vs=1635

As you can see for CPU tasks ED ram rarely makes a difference. Unfortunately in the laptop reviews they did not show the tasks where memory latency matters the most, but understand such tests are actually the abnormal things in most work.

That said most things do not benefit much from EDRAM. This is because EDRAM is only barely faster on a latency perspective than ddr3 or ddr4 (both ddr3 and ddr4 are in the same general domain on latency) . Now ddr4 has higher bandwidth generally than ddr3 and edram has higher bandwidth than dual channel ddr4.

As an analogy.

Imagine you have a 2 lane highway. The highway can have a speed limit set at 50 mph, 60, 70, or 85 mph. This is latency

Bandwidth is whether you have a 2 lane highway or a 8 lane highway. Having a wider highway does not make the individual cars go faster, it does not change the speed limit it just allows more volume to occur. Most apps are not bandwidth limited, but decreasing the latency making it faster to retrieve things from memory can help certain types of programs that can't fit the data in the L1, L2, L3, etc caches.

----

Side rant about a new cool and upcoming technology. Intel / Micron's 3D XPoint, the thing like NAND but so much better.

3D Xpoint, also known as intel Optane or Micron Quant X has latency that is an order of magnitude lower than ram. That said it is two almost three orders of magnitude faster in latency than ssd.

The big benefit of 3D Xpoint is not bandwidth but once again latency. That and it is non volatile, you can access the files directly instead of reading the entire block of memory, lower power etc.

One of the big things in the future with 3D Xpoint or some similar technology from a different company is how certain workloads will now have things that are effectively ram drives instead of ssd which in turn replaced hard drives.

This is speculation. Depending on cost on price per gb and how many dies of 3D Xpoint and dies of NAND it would take I would not be surprised in the far future (not the next 18 months) that we will eventually see hybrid drives with a 128gb or so (OS or programs) are always stored in 3D Xpoint, and a SSD acting like a 1 TB or 2 TB Hard Drive but it is now SSD all on one package treated as a one drive letter. That said it will be at least 18 months or so, for right now the higher profit per each die of 3D Xpoint produced would be targeting the server workloads where they are working to pool ram and 3D Xpoint where it appears as 1 volume per software, and they stagger it so where it would be a big speed up for certain

3D Xpoint also may be benefits in low power devices (phones and internet of things) once the server market is saturated and you will see intel and micron target the more consumer market. Very small amount of ram (possibly no ram at all in IoT) and the rest being 3D Xpoint and little to no nand since data would be stored in the cloud and you do not need to use nand.

 

[superstition]

The eDRAM on Broadwell C can't actually overclock on most motherboards. If you change the multiplier on the eDRAM it actually disables the eDRAM even though it still may show as enabled. In addition, overclocking the core to non-even frequencies (4.1GHz) also disables the eDRAM. I've tested this myself using the AIDA64 cache bench.

Yes, it's important to get a motherboard that will let you separately clock the EDRAM and also bypass the TDP throttling.

That very few boards can do this is a testament to the substandard quality of the motherboard industry.

Why is eDRAM support so flaky?

The behavior you described isn't flakiness. It just shows that EDRAM has a specific zone of efficiency. It doesn't want to be clocked too lowly or too highly.

 

[ZGR]

Yes, it's important to get a motherboard that will let you separately clock the EDRAM and also bypass the TDP throttling.

That very few boards can do this is a testament to the substandard quality of the motherboard industry.

The behavior you described isn't flakiness. It just shows that EDRAM has a specific zone of efficiency. It doesn't want to be clocked too lowly or too highly.

I'm talking about why eDRAM would be disabled outright when the CPU is at specific odd numbered clockspeeds. Or when eDRAM can't properly be disabled on my motherboard unless I flash back to an older unsupported BIOS.
Or certain Motherboard manufacturers locking eDRAM to stock x18 multiplier.

I understand that eDRAM only operates efficiently in a narrow band of frequencies, but I don't understand why so many different motherboard manufacturers differ in eDRAM implementation and support.

Say I was to go back in time with the Z97 launch. If I was to ask the question of which OEM would have the best bug-free support of eDRAM with Broadwell, who would it be? I'm not even sure if I can answer that question today....

Being able to overclock eDRAM is not very important due to its sensivity, but I would like to be able to disable it outright to test certain games' performance.

 

[superstition]

I don't understand why so many different motherboard manufacturers differ in eDRAM implementation and support.

Planned obsolescence coupled with disinterest. It's not in Intel's interest to have Broadwell C embarrass Skylake anymore than it already has. Motherboard makers typically do the bare minimum to keep people buying their products.

There are a lot of examples of shoddy customer service in motherboards, from bologna hype ("military-grade") while obfuscating or lying about important information (VRM quality/specs), to not updating CPU microcode, to weird bugs (Gigabyte UD3P 2.0's refusal to post above a 22 multi), to silly designs (40mm VRM fan blowing the wrong way instead of a air/water combo sink), to extremely vague BIOS update descriptions, to hosting outdated drivers on their website, to 6 revisions of the "same" board with very different specs and only the tiny revision number on the box.

 

[tex_willer]

One problem is that Broadwell and Skylake each implement eDRAM rather differently.
Using L4 as a victim cache to L3 is the primary reason I got the 5775C.

I doubt we will see an in depth eDRAM article. We will need to gather the data ourselves.

QUOTE]

to some extend, we same similar machines, I also have 5775C, with NH D15, 2x8GB 2400Mhz C10 DDR3, Asus Z97 Deluxe USB3.1 (nfc&wlc). also I will try to delid this CPU but my friend need to deliver deliding tool for me next time he travels here. maximum OC which I get was 4.4 Core and 4.1 Uncore, more than that require some really extreme voltages which I wasn't interested to try. 4.3 Core and 3.9 Uncore is some sweet spot which really don't require any serious voltages.

Any playing with Edram ratio, any multiplier, or disabling it in Bios, yield totally negative effect on L4 speed.

I'm very interested to share any knowledge or tricks about this CPU

I just want to post again regarding 5775C OC and eDRAM OC.

First off is the is the 65W Broadwell-C and Skylake-R TDP; a very real performance limitation. A few posted articles regarding Broadwell-C OC along with general performance metrics generally did NOT remove all the restrictions needed to unlock the TDP. Out of the box, these 65W CPU's throttle hard! Even after applying a traditional overclock, the CPU will still throttle at load for brief moments.

For example, this is just what XTU displays after overriding the listed variables:
Draw 4096W I dare you i7!

Then there are a few other power restrictions to be lifted in BIOS that are not shown. All in all, my 5775C is now a ~125W TDP CPU with the 4.2 GHz OC. With all the necessary power restrictions lifted, this CPU will never throttle at load. And yes, the CPU still idles at around ~0.850v when not loaded. This is NOT a fair comparison to the 65W locked Skylake-H (The improved eDRAM that we all want in an LGA socket). Who would want Broadwell-C if it was restricted to 65W and had no overclocking abilities? That is what Skylake-H is.

Next is eDRAM Overclocking! We need a comprehensive list of games, benchmarks, and various software that is heavily impacted by eDRAM and the LACK of it. Like I said in a post above, eDRAM is easily toggled on and off in the UEFI BIOS. Nearly every eDRAM overclocking article I have seen does not publish any data with the eDRAM off. I see that as a big red flag. I do believe that 1.9-2.1 GHz is the effective limit for eDRAM (latency is the big problem past 2.1 GHz), but I do not know the performance benefits with eDRAM on vs off in many situations. I want to review 7z, Winrar, and CPU limited game titles. I do not believe some canned benchmark like Cinebench to benefit from eDRAM. What I care about is when I am in CPU limited situations where the cache is at the limit. We need a lot more info.

I also want to know what software is impacted the most by eDRAM. It would be nice to have an "eDRAM friendly" benchmark to really compare further improvements to eDRAM in the future. I sure want to know how much Skylake-H has improved over Broadwell-C regarding the L4.

after I had read your post here, I tested CPU and yes, it have little throttling (when I look at XTU) when you apply some moderate OC. but, turning on XPM in Bios, looks like it automatically adjust and turn off these limitations. When I turn on XPM and enter Windows and XTU, Turbo Boost Power Max and Processor Current Limit is set to maximum (4095W, etc.). so, I suppose that XPM resolve it automatically. but, if there is any other power restrictions in Bios, which have real effect, please let me know.

according to Aida64 Memory Test, I found that Uncore ratio have big impact on L4 speeds, especially write and copy speed.
when you let stock Uncore speed at 3.3, than this DDR3 2400Mhz even have higher write speed, and similar copy speed. but read speed stays way lower compared to L4 read speed, no matter is Uncore ratio at 3.3 or 4.1 or any speed between.

looks like we have to bump up Uncore ratio to have better benefits from Edram.

Cinebench C15 is affected with L4, at least I think so. with XPM turned on and RAM speed at 2400 C10-12-12-30-1T, I tested Cinebench with overclocked Uncore, stock Uncore, and modified Edram multiplier in Bios (which totally ruin L4 speed).

overclocked Uncore get definitely better average results compared to stock Uncore (improved L3 speed can also play role here).
but, with "ruined Edram" (which do not affect L3 speed, but totally decrease L4 speed), results are notably lower.

This is alarming news to me! I went back to my BIOS and tried a few different configurations while testing AIDA64's Cache and Memory Benchmark.

I now can verify that my eDRAM is not being disabled!

When I first got my 5775C I ran the same benchmark with eDRAM disabled becauseof an outdated bios. Updating the bios added eDRAM support and I never bothered to check to see if disabling eDRAm actually disables it...

Changing the eDRAM multiplier works, regardless of my CPU frequency. Testing at 4.2, 4.1, 4.0, and 3.9 GHz shows my eDRAM to be working. The big problem is I now have no way to disable it unless I flash back to the older BIOS. Thanks Asus QUOTE]

there is some very strange things about this which I do not understand. here is my experience and details.

any touching of Edram ratio in Bios result in very big drop of L4 speed. no matter if I set it to Disabled, or set any speed but Auto settings (if Auto settings are in use, CPU-Z show 1800Mhz L4 speed). is this some problem how Bios, motherboard, or whatever regulates Edram, I do not know.

also, when I set it to Disabled, CPU-Z do not read any L4 speed, but Aida64 Memory Test still measure and show L4 speed

I'm totally confused there.

That said most things do not benefit much from EDRAM. This is because EDRAM is only barely faster on a latency perspective than ddr3 or ddr4 (both ddr3 and ddr4 are in the same general domain on latency) . Now ddr4 has higher bandwidth generally than ddr3 and edram has higher bandwidth than dual channel ddr4. QUOTE]



http://img4.imagetitan.com/img.php?image=14_4441.png



at this test L4 get 41.8ns, but it varies and many times, in repeating tests, it goes around 50ns, sometimes even to 55ns. Is that normal or? and when I turn off XPM, than RAM speed drops to 1333 and RAM latency becomes around 75ns. If I do remember well, doing this test with XPM off, L4 will also have around 70-75ns, so it means it is somehow related to RAM or XPM settings. again, I do not fully understand how it all works.

this also raise question is 5775C with L4, when combined with fast ram like 2400 or faster, worth it at all.

 

[tex_willer]

sorry for confusion with Quote but I don't know how to correct it. you can anyway click to expand my post and I had put bold on my text to make it easier.

here is direct link to image of Aida64 Memory Test :


image hosting 12mb

 

[lolfail9001]

I mean seriously, Broadwell-C was overrated in initial Skylake reviews simply because about everybody ignored testing with different memory speeds IIRC.
Looking at Skull Canyon memory scaling article, all i see is eDRAM supplementing faster RAM, that's about it.
So, in ironic twist of fate, eDRAM would be best on highest end servers, as those tend to run with slowest of memory.

 

[superstition]

I mean seriously, Broadwell-C was overrated in initial Skylake reviews simply because about everybody ignored testing with different memory speeds IIRC.
Looking at Skull Canyon memory scaling article, all i see is eDRAM supplementing faster RAM, that's about it.
So, in ironic twist of fate, eDRAM would be best on highest end servers, as those tend to run with slowest of memory.

Like the way people ignore the way Broadwell C was able to match and beat the "revolutionary" Skylake and its requirement of replacing one's RAM and motherboard to become part of the revolution?

Broadwell C without the throttling and with iGPU disabled could have been a compelling part for enthusiasts, especially with solder.

As for early reviews... Hruska cited a slide from Anandtech that showed a bottleneck in one game in SLI due to slow RAM. Any DDR4 RAM 2400 or faster didn't bottleneck but 2133 did.

http://www.extremetech.com/extreme/...g-ddr4-to-3200mhz-improve-overall-performance

 

[lolfail9001]

Like the way people ignore the way Broadwell C was able to match and beat the "revolutionary" Skylake and its requirement of replacing one's RAM and motherboard to become part of the revolution?

Broadwell C without the throttling and with iGPU disabled could have been a compelling part for enthusiasts, especially with solder.

As for early reviews... Hruska cited a slide from Anandtech that showed a bottleneck in one game in SLI due to slow RAM. Any DDR4 RAM 2400 or faster didn't bottleneck but 2133 did.

http://www.extremetech.com/extreme/...g-ddr4-to-3200mhz-improve-overall-performance

Skylake is nothing revolutionary, so your point is already weak.
Broadwell C without the throttling and with iGPU disabled would perform exactly as Broadwell C with throttling and iGPU disabled with multiplier set to 42. Just broadwell things.

Last, the citation you provide is from Haswell ^_^.

 

[superstition]

Skylake is nothing revolutionary, so your point is already weak.

Restating my point and claiming it's a rebuttal of that point is... odd.

Broadwell C without the throttling and with iGPU disabled would perform exactly as Broadwell C with throttling and iGPU disabled with multiplier set to 42. Just broadwell things.

We've already been through this.

Last, the citation you provide is from Haswell ^_^.

Yes, it is. The article is also about how DDR4 is hardly revolutionary for the desktop.

 

[JoeRambo]

Yes, it is. The article is also about how DDR4 is hardly revolutionary for the desktop.

Hard to achieve revolutions when dram cell cycle is still 200Mhz for DDR4 3200 ( same as DDR3 1600 / DDR2 800 ). While bandwidth is great, it is still ~50ns affair to get data from memory in best case due to rising latencies with each tech. To make use of available bandwith, CPU needs to be smart with prefetch and stuff (while risking just burning energy) and of course not every workload will benefit from prefetch either.
EDRAM is a bit missed opportunity, latency is too high and bandwith is not that much greater than DDR4. And it comes with costs as well - on Broadwell it burns 2MB of L3 cache for L4 tags making some benefits dissapear.

 

[tex_willer]

so, conclusion is that 4790K with some very fast DDR3 (faster than 2400Mhz, with low latency) would perform better overall compared to 5775C?

take a look here :

https://www.techpowerup.com/forums/...ache-and-memory-benchmark-here.186338/page-25

pay attention to Haswell/Devil guys memory and cache tests. you will see that they have little lower score with RAM copy speed, but all of them have lower latency (when they use about equal RAM equipped on 5775C). about 10ns lower latency, or even little more.

I mean, for example, if we put DDR3 2400 C10 on both 4790K and 5775C, 4790K would have about 10ns better latency on RAM.

Haswell/Devil also have better L3 cache speed, when their North Bridge is clocked at same clock as 5775C. and not all 5775C can overclock their North Bridge to 4000, which is stock speed for 4790K (if I'm not wrong). it also have 2mb of L3 more.

 

[Hail The Brain Slug]

so, conclusion is that 4790K with some very fast DDR3 (faster than 2400Mhz, with low latency) would perform better overall compared to 5775C?

take a look here :

https://www.techpowerup.com/forums/...ache-and-memory-benchmark-here.186338/page-25

pay attention to Haswell/Devil guys memory and cache tests. you will see that they have little lower score with RAM copy speed, but all of them have lower latency (when they use about equal RAM equipped on 5775C). about 10ns lower latency, or even little more.

I mean, for example, if we put DDR3 2400 C10 on both 4790K and 5775C, 4790K would have about 10ns better latency on RAM.

Haswell/Devil also have better L3 cache speed, when their North Bridge is clocked at same clock as 5775C. and not all 5775C can overclock their North Bridge to 4000, which is stock speed for 4790K (if I'm not wrong). it also have 2mb of L3 more.

For reference, I have a 4770K at 4.3 core, 4.1 cache, and with DDR3 2400 at 10-12-12-32. If it would help a comparison I can change any settings requested.

The latency is roughly equivalent to the latency on the eDRAM on the above broadwell chip, the bandwidth is about 2/3 of the way there.

It could be really interesting to see some benchmark comparisons between that broadwell (underclocking it 100 mhz down to 4.3 to compare to this 4770k).

 

[tex_willer]

thank you, but you do not need to change anything as your results show exactly what I have said. RAM copy speed is better at my setup, read write is similar, and latency is much worse.

your L3 speeds are also better. is this due to your cpu have 2mb more of L3 cache, or Haswell/Devil have anyway higher L3 speeds than 5775C (no matter the cache size). I don't know what exactly determine that speed. if you know more about it, please tell me.

yes latency of your RAM is roughly equivalent to eDRAM on my chip, but as I already told, that latency vary between low 40 and 55ns. is this due to some benchmark problem or it really vary, don't know.

take a look at link I provided from Techpowerup forum, on page 25, there is 4790K with DDR 2800Mhz 11-13-13-31-T1. it have as expected, better speeds and latency (very low, under 40) compared to your setup. but using very fast RAM, I fear, will limit overclock potential.

I see also guys with Skylake with very fast DDR4, have very good RAM results. read speed is little lower or similar like eDRAM, write is much better, copy is similar od little better.

eDRAM still make sense if latency is about 40ns, but if it vary to 55ns than it make no sense, if you ask me

I have here on paper results of some testings, so I will write everything I have at 4.3 speed, to compare to your 4770K.

at 4.3 core speed, it make CPU-Z single core result 2082, and average result of multi core 8570. Uncore ratio don't seem to affect CPU-Z results.

at 4.3 core / 3.3 uncore - Cinebench C15 - average 904
at 4.3 core / 4.0 uncore - Cinebench C15 - average 918 (take into account, as I have said before, Uncore ratio affect L4 speed much).

Passmark 8 test results with 4.3 core / 4.0 uncore :

12521 overall (results below are best scored, so it maybe does not match perfectly with overall score, because I made more than one tests)
20880 integer
9152 floating point math
54.2 prime
59 SSE
16202 compression
2297 encryption
903 physics
9401 sorting
2484 single core

prime and physics are very affected by Uncore ratio, others are mostly affected by raw Core speed. so probably this 5775C have better prime and physics scores compared to other processors, because of L4 presence.