5775c better gaming CPU than 6700k?

[DrMrLordX]

There is no reason why Skylake is architecturally better-suited for DX12 than Broadwell.

 

[dark zero]

There is no reason why Skylake is architecturally better-suited for DX12 than Broadwell.

Actually Broadwell is better only due EDRAM. But without that, Haswell could beat them. Skylake without EDRAM is better in that way. Is built to DX12 standards

 

[Shivansps]

This is simple, just wait for a edram Skylake, there is no rush to upgrade.

 

[DrMrLordX]

Is built to DX12 standards

What standards?

 

[know of fence]

Here is one example of Skylake on top, though still with a massive handicap for DDR4. Maybe there is something about driving games that depends on low latency.

At 4K DDR4 is ahead, multi GPU setups can benefit from high bandwidth memory, typically this is the only scenario where quad channel builds can get ahead (though not tested here). 4K multi GPU setups is also a case where a faster Level4 cache seems to help, because multi-GPU can only use the 4GB GDDR5 of one of the dual cards, meanwhile rendering really large frames.
If the bottleneck of current console to PC ports are unnecessary draw calls to the CPU, than a fast extra cache is just the thing.

 

[therealnickdanger]

Just wait for GT4e.

 

[LTC8K6]

Just wait for GT4e.

Will that be in a Skylake DT chip?

 

[therealnickdanger]

Will that be in a Skylake DT chip?

Take the details below with a grain of salt, but one thing we know for certain is that mobile Xeon will have it and the next NUCs will have it.

http://wccftech.com/intel-skylake-p...features-72-execution-units-128-mb-edram-llc/

 

[Zodiark1593]

What would be very interesting to see is Skylake-E on quad channel 4000+ mhz ddr4.

 

[know of fence]

So the Techreport explains the impact of eDRAM with this benchmark. When the graph flatlines on the right the test reached effective RAM speeds. They used 2166 MHz DDR4 RAM and they also use 1333 MHz ram for Sandy Bridge. Generally it shows just how tiny RAM bandwidth is compared to CPU cache, which is a reason why it rarely matters.

My main motivation for including this strange plot is to get you to consider the 64MB test block size. There, the purple line for the 5775C indicates higher bandwidth than any other CPU tested; the 5775C's bandwidth at this block size is roughly double the 6700K's. This data point is one spot where we can see the impact of the 5775C's 128MB L4 cache. Ooh, ahh. - Techreport Skylake Review

eDRAM shows about twice the bandwidth of DDR3 here, which isn't all that much, though still more than on quad channel X99 platform. But copying 64 MB chunks it also has significantly lower latency about half that of Haswell or Skylake. This is something that even a quad channel Memory can't even remotely match. (not actually tested/shown by the techreport)

Never mind the weird log2 scales, the 5775C is has almost half the latency at 32 and 64 MB. But it's also interesting that Skylake has lower latency than Haswell at those points, though RAM latencies (at 256MG and higher) are probably bigger.

Maybe we'll get some more results from AT memory scaling tests. To me it seems though that given the massive differences between DDR3 and DDR4, memory latency is kind of a big deal.
We know (integrated) graphics is bandwidth constrained, because it copies as much stuff as possible in real time, but for CPU constrained scenarios having quick data access seems important, so it's the purple line at the right of the last graph that tells why we see the 5775C ahead.