2016+ CPU/GPU News

[MoroseTroll 24]

I suppose, the "dmap" function is barely multi-threaded and/or doesn't use SIMD at all. The results I was talking about earlier ("AMD Zen works usually faster with 128-bit SIMD than any Intel CPU") I've got by entering the "testSIMD" console instruction.

[Bikerdude 3738]

I've got by entering the "testSIMD" console instruction.

ok, for curiosities sake how do I do a usefull test on both systems?

[MoroseTroll 24]

Before I've started my work on SIMD optimizations, I did usually use vanilla Doom 3 (not sure about the BFG edition) for such an experiments: run the game, enter the console menu and type the instruction. But later, I compiled a standalone console executable that does almost the same. In case you run the last one, it would be nice to redirect its output to a text file, like "TestAll > Results.txt". But be careful: my test only run my SIMD code on AMD CPUs (on any of them, but not earlier Athlon, i.e. no K6 & K5 ), so if you run it on Intel (or VIA ), it'll run the old (not my) code, i.e. you'll see no difference.

[Bikerdude 3738]

Ive downloaded the little file, whats flags do I use to get a result that I can upload as a file?

[MoroseTroll 24]

What do you mean by saying "flags?" My method is simple:

 

1. Download "TestAll.7z" and unpack it somewhere.

2. Run "TestAll.exe > Results.txt".

3. Open "Results.txt" in Notepad or any other text viewer/editor and compare the results.

 

The executable has no flags at all. It's just a standalone version of the "TestSIMD" console instruction from Doom 3.

 

BTW, 7-Zip has been updated to 17.01 yesterday. Also, please don't forget to toggle on the "Use large pages" option in it, because this improves overall performance .

Edited August 29, 2017 by MoroseTroll

[Bikerdude 3738]

ah ok, I didn't realize it auto generated the txt file.

[MoroseTroll 24]

Um... Actually, its output is directed to console (DOS/*nix style), so the best way to anylize the results is to redirect them to a text file.

[Bikerdude 3738]

1. Download "TestAll.7z" and unpack it somewhere.

2. Run "TestAll.exe > Results.txt".

3. Open "Results.txt" in Notepad or any other text viewer/editor and compare the results.

done, see attached.

 

Results.txt

[Bikerdude 3738]

Due to the various issues and immaturity of the bios/drivers its going back for a refund. If anyone wants me to do any further tests they had better be quick, I will be raising the return with amazon on Friday.

Edited August 30, 2017 by Bikerdude

[MoroseTroll 24]

I think there will be no adequate BIOS/UEFI & driver support for Win 7 on Ryzen - we all know why.

Edited August 31, 2017 by MoroseTroll

[jaxa 222]

Intel Xeon-W

http://www.anandtech.com/show/11775/intel-launches-xeon-w-cpus-for-workstations

 

MediaTek Helio P23 and P30

http://www.anandtech.com/show/11770/mediatek-helio-p23-helio-p30-midrange-socs

 

AMD Threadripper 1900X

http://www.anandtech.com/show/11794/now-shipping-amd-ryzen-threadripper-1900x-ryzen-pro-cpus

[Bikerdude 3738]

Final opinion from the 6850k/x99 & 1600x/X370 -

 

Intel - 6850k/X99 (total cost £600):

 

Pros

• Multi CPU perf noticeably better.
• Motherboard beefier across the board -
  • 40 PCIe lanes on the 6850 compared to 16 on the 4790.
  • More PCIe expansion slots.
  • More ram slots (8 versus 4)

Cons

• Cold boot times significantly slower than my 4790k/Z97, 45s to desktop compared to 20s.
• Even overclocking the 6850k to 4.2Ghz, single thread perf not quite as fast as the 4790k @4.4ghz - so dmapping a tad slower.

Amd - 1600x/X370 (total cost £400):

 

Pros

• Multi CPU perf significantly better, way better than the 6850k (both CPU's were 6 core)
• Motherboard better in some respects -
  • 24 PCIe lanes versus 16 on the 4790.
  • USB 3.1

Cons

• Cold boot times significantly slower than my 4790k/Z97, 35s to desktop compared to 20s.
• AMD chipset drivers immature - getting weird incompatibility issues - but that to be expected seeing as the platform is only 6 months old
• Weird warm and cold boot issues -
  • on cold the system would do a double boot, some info online suggest this is normal and its the mobo checking the ram - either way I am NOT putting up with that (cold to desktop is well over a minute in that scenario).
  • Warm rebooting, randomly windows 7 would sit on the logo screen for waaaaay too long p - At a guess chipset drivers.
  • Shutting down sometimes it would a bit long - At a guess chipset drivers.
• Single thread perf @ 4.0Ghz noticeably slower than the 4790k @4.4ghz - so dmapping on my city map took over a minute longer

I will not be shoehorned into upgrading to Win8/10, all of the issues with boot times on both system would have probably been alleviated with the aforementioned OS's. And the weird cold/warm boot issues on the Ryzen setup would most likely not have even come up.

 

For this you want the best perf and 100% Win7 compatibility, you can still buy 4790k on newegg ($391, 3 year warranty), or if you want to save some cash CEX ($250, 2 year warranty)

[jaxa 222]

Ryzen Mobile launched:

 

https://www.anandtech.com/show/11964/ryzen-mobile-is-launched-amd-apus-for-laptops-with-vega-and-updated-zen

[jaxa 222]

Intel is likely to put out a 6-core mobile chip before AMD does, but the CPU and GPU improvements make these chips worthwhile.

 

The 200% CPU improvement is probably a best case scenario from switching from "2 modules" to 4 cores/8 threads in a 15W chip combined with the IPC gains. ~52% improvement from just the IPC gains is more realistic.

 

The 128% GPU improvement is realistic, with 114% being a minimum improvement (10 vs 8 GPU "compute units", 1300 MHz vs 758 MHz frequency. 1.25 x 1.715 = 2.14).

 

It looks like these could double TDM performance vs previous chips (GPU is the bottleneck, right?), or do a lot better than my A6-3400M which has gotten as low as 3-5 FPS during mission beta testing. The two chips being launched are only at a 15 W TDP, so you might have to wait for 30-35 W if you want more performance and not ultrabook-like specs.

Edited October 26, 2017 by jaxa

[jaxa 222]

GTX 1070 Ti, Nvidia's answer to the Vega 56, although it is $50 more expensive:

https://www.anandtech.com/show/11987/the-nvidia-geforce-gtx-1070-ti-founders-edition-review

[jaxa 222]

LOL it happened:

 

https://newsroom.intel.com/editorials/new-intel-core-processor-combine-high-performance-cpu-discrete-graphics-sleek-thin-devices/

 

The new product, which will be part of our 8th Gen Intel Core family, brings together our high-performing Intel Core H-series processor, second generation High Bandwidth Memory (HBM2) and a custom-to-Intel third-party discrete graphics chip from AMD’s Radeon Technologies Group* – all in a single processor package.

 

It’s a prime example of hardware and software innovations intersecting to create something amazing that fills a unique market gap. Helping to deliver on our vision for this new class of product, we worked with the team at AMD’s Radeon Technologies Group. In close collaboration, we designed a new semi-custom graphics chip, which means this is also a great example of how we can compete and work together, ultimately delivering innovation that is good for consumers.

Edited November 7, 2017 by jaxa

[AluminumHaste 966]

I was surprised to read that earlier today, after Intel squashed the rumours earlier in the year, now it's come to fruition. Good for AMD, this should really help.

[jaxa 222]

I was surprised to read that earlier today, after Intel squashed the rumours earlier in the year, now it's come to fruition. Good for AMD, this should really help.

 

Some of the AnandTech comments speculate that it WON'T help AMD.

 

The graphics look better than what will come with AMD Raven Ridge, and AMD has not added HBM to any of their consumer parts yet.

 

Add to that Intel's better cores and possibly 6 cores for these chips, and it is the king of high-TDP laptop chips.

[jaxa 222]

Still looking good for AMD?

 

https://newsroom.intel.com/news-releases/raja-koduri-joins-intel/

 

Intel to Develop Discrete GPUs, Hires Raja Koduri as Chief Architect & Senior VP

Edited November 9, 2017 by jaxa

[AluminumHaste 966]

I will don't see how Intel can do a discreet GPU, they hold no patents in GPU architecture.

[jaxa 222]

They made Larrabee which later became Xeon Phi which competes with Nvidia products (Knights Mill vs Nvidia Tesla for half-precision machine learning). And they make iGPUs. So they have a few patents somewhere.

Edited November 9, 2017 by jaxa

[AluminumHaste 966]

I see, the reason I thought that was because their patent agreement with Nvidia ended but they still have access to the patents for making their GPUs.

However, Intel seems to hate Nvidia, hence going with AMD.

[lowenz 554]

I will don't see how Intel can do a discreet GPU, they hold no patents in GPU architecture.

There are no more "GPU" with the "G", as Huang said

 

There are only high parallelized computation processors usable for graphics too

Intel (as NVIDIA, as AMD) will produce TENSOR processors that can be used to accelerate graphics computation too (thanks to the OS drivers).

Edited November 11, 2017 by lowenz

[Bikerdude 3738]

Another nail in the coffin for the Intel management engine - https://arstechnica.co.uk/information-technology/2017/11/intel-warns-of-widespread-vulnerability-in-pc-server-device-firmware/

[jaxa 222]

Another nail in the coffin for the Intel management engine - https://arstechnica.co.uk/information-technology/2017/11/intel-warns-of-widespread-vulnerability-in-pc-server-device-firmware/

 

People will just patch and move on, or forget to patch. Intel ME will become even stronger.

 

We now live in an age of total insecurity in computing. Updates might protect you somewhat from script kiddies and criminals, but intelligence agencies will find a way in if they want in. Similar backdoors will be found and exploited in AMD and ARM, comprising the overwhelming majority of personal computers (to include mobile phones). Let's see a widespread switch to MIPS or IBM's POWER. *crickets*

Edited November 26, 2017 by jaxa