Overclock.net › Components › CPUs / Processors › Amd Ryzen 7 1700X 8c/16t Processor

Amd Ryzen 7 1700X 8c/16t Processor

Posted

Pros: Scales with better cooling up to a point, much higher IPC than Bulldozer , much less expensive compared to Intel i7-6900k

Cons: 95W TDP vs the 65W TDP of Ryzen 7 1700, low overclocking ability, no included RGB cooler



PROS
8 cores and 16 threads , 3.4GHz base clocks.

Was purchased at Microcenter with a combo deal for $100 off. Without that deal I would suggest a Ryzen 7 1700 instead.

Comparable performance to i7-6900K in professional workloads.

Am4 platform will be supported with Ryzen revision

3.9GHz XFR boost on up to 2 cores and 3.5GHz all core XFR boost is minimally less than the highest SKU Ryzen 7 1800X (4.1GHz on 2 cores with XFR , 3.7GHz all core)

AM4 platform is a longterm socket , plus overclockable B350 boards are cheaper than X99 boards, so a motherboard purchase isn't a deadend unlike buying into X99

CONS
Lower IPC than Intel Kaby Lake (roughly Haswell/Broadwell)
Slightly lower NVMe SSD performance than Intel Z270 per SSD review sites (supposedly for random past QD=8)
Limited overclocking ability (average is 4GHz , which is not much over the 3.5GHz out of the box)
Higher idle clock than Ryzen 7 1700 (2.2GHz rather than 1.55GHz)
On motherboards with BCLK overclocking, there are no straps so that means any storage devices and/or wifi may suffer corruption and error issues ; PCie slots downgrade to PCie 2.0 when past ~ 107 BCLK per users' experiences on ASUS Crosshair VI Hero and Asrock X370 Taichi
Memory latency per AIDA64 on 3200MHz CL16 is around 75ns which is still behind Intel
Dual channel memory can be problematic if you use low memory clocks (i.e. DDR4 2133 or DDR4 2400) instead of 2666MHz+
Memory capacities above 32GB are not well supported with over DDR4 2666Mhz at this time
Total PCI-e lanes lower than the Intel X99 platform , which is why so few motherboards have more than one M.2 NVMe SSD slot
DX11 viewport workloads and games do not make use of this amount of threads
Requires GPU (not really a con but may be a problem for some people)
Requires DDR4 3200MHz or above to reach full potential of Infinity Fabric
To make full use of the 25MHz granularity of SenseMi technologies you cannot overclock it. Overclocking disables Precision Boost and XFR.
AM4 motherboards still need work across the board

Other notes

Assuming that an average i7-6800K or i7-6850K Broadwell-E achieves 4.5GHz (http://hwbot.org/hardware/processors#key=core_i7_6850k) on air or water, this is competitive in multi-thread even vs an overclocked one. Not accounting for software & SMT efficiency, we can estimate by the IPC differences.
i.e. 6 cores x 4.5GHz x 1.05 (IPC adjustment) ---> 3.54GHz 8 core Ryzen 7
6 cores x 4.5Ghz x 1.1 (IPC adjustment) ---> 3.7125GHz 8 core Ryzen 7
Of course there is the disadvantage that in single threaded workloads 4.5GHz vs 3.5GHz is more than +33% after accounting for IPC. Due to XFR this is less pronounced since at the 3.9GHz XFR clockspeed , you have ~25% difference after counting IPC.
* A Ryzen revision or stepping could change the maximum XFR or the maximum clockspeeds achievable with overclocking. If that happens before any future 6+ core Intel CPUs with hyper-threading appear then AMD stands to gain market share.Without a step up in clockspeeds for XFR it is only going to compete with Intel stock clocks vs stock clocks and fall behind slightly due to lower all core boost clocks + slightly lower IPC.
Skylake-S Update June 20,2017 (Click to show)
Skylake 8 core parts are rather competitive pricing wise. However, platform cost is higher overall due to the cost of motherboards. Even the cheapest X299 motherboards cost in excess of $200.

The i7 7820X is a 8 core part with quad channel DDR4 (2400MHz is official spec). MSRP is $599

The i7 7800X is a 6 core part with quad channel DDR4 (2400MHz is official spec). MSRP Is $389


In Windows 10 PE (LiveUSB) I ran AIDA64 Engineer 5.9.4200.0 and obtained the following with DDR4 3200MHz CL16 (Dual DDR4-3200 16-18-18-38 CR1)

CPU Queen 80940
Photoworxx 22980 MPixel/s
Zlib 643.0 MB/s
AES 62071 MB/s
Memory latency 74.4 ns --- so-so result , subsequent testing resulted in 73.6 or 73.7ns after dropping tRC to 56 with AGESA 1.0.0.6 , overclocking to 3.925GHz resulted in ~73ns also (16-16-16-38 also results in ~72.6ns at 39.25x multi)
Memory copy 41625 MB/s
Memory read 47482 MB/s
Memory write 45912 MB/s

Cinebench R15 (CINEBENCH R15.038_RC184115) multicore: 1558 cb
Cinebench R15 (CINEBENCH R15.038_RC184115) single core: 155 cb

Cinebench R15 4GHz multicore: 1760cb or 13% (Click to show)

SuperPi 1M : 11.9seconds with DDR4 3200MHz CL16 , I had a subsequent run of 11.122 seconds so this is down to variance I believe ; 11.030s in Windows 7 x64 SP1
SuperPi 32M: ~ 9 min 46 seconds , subsequent run of 9 min 42.873 seconds so it's about that time , in Windows 7 x64 I obtained 9 min 18.403s

Sisoft Sandra 2017.04.24.18 with DDR4 3200 CL16, Processor Arithmetic (MT) = 220.84 GOPs
Sisoft Sandra 2017.04.24.18 with DDR4 3200 CL16, Processor Multimedia (MT) = 436.67 MPix/s
Sisoft Sandra 2017.04.24.18 with DDR4 3200 CL16 , Processor Financial Analysis (MT) High / Double Precision = 37.57kOPT/s Aggregate, 45.67kOPT/s Binomial Euro Option pricing , 30.91 kOPT/s Monte Carlo Euro Option Pricing
--> Single Precision = 65kOPT/s aggregate , 232.49 MOPT/s Black-Scholes Euro option pricing, 86.16kOPT/s Binomial Euro options pricing, 49kOPT/s Monte Carlo Euro option pticing
Sisoft Sandra 2017.04.24.18 with DDR4 3200 CL16 , Processor Scientific Analysis (MT) High / Double Precision = 21.8GFLOPS aggregate, 62.08 GLOPS GEMM FMA, 7.66 GFLOPS FFT FMA, 86.9 GFLOPS NBDY FMA
Sisoft Sandra 2017.04.24.18 with DDR4 3200 CL16, Processor Image Processing (MT) = 209.8 Mpix/s Aggregate

Sisoft Sandra 2017.04.24.18 with DDR4 3200 CL16, .NET Arithmetic (MT) = 72.64 GOPs Aggregate
Sisoft Sandra 2017.04.24.18 with DDR4 3200 CL16, Java Arithmetic (MT) = 246.13 GOPs Aggregate

Sisoft Sandra 2017.04.24.18 with DDR4 3200 CL16, Memory Bandwidth Multithread = 34.48GB/s Integer and 34.82 GB/s Float
----> Global Data Memory : In-Page Random Access Pattern Memory Latency 60.3ns

Preliminary CPU all core overclock results:
34x multi (~3.4GHz) = <1.2V using dynamic voltage +0.00625V , CineBench R15 stable
35x multi (3.5GHz) = 1.225V (auto)
35.75x multi (~3.575GHz) = < 1.344V
36x multi (~3.6GHz) = < 1.344V
37x multi (~3.7GHz) = < 1.344V
38x multi (~3.8GHz) = < 1.344V
38.5x multi (~3.85GHz) = < 1.344V
38.75x multi (~3.875GHz) = ~ 1.344V - 1.356V
39.25x multi (~3.925GHz) = ~ 1.344V - 1.356V ... 1 hour AIDA64 stable
39.50x multi (~3.950GHz) = ~ 1.344V - 1.368V
39.75x multi (~3.975GHz) = ~ 1.356-1.368V (1.35V in BIOs + medium LLC) , Cinebench R15 stable when RAM Is at 2133MHz CL15
--> appears to require 1.38ish volts for AIDA64 1 hour (used 2133Mhz RAM speed to isolate RAM overclock from CPU overclock)

40.50x multi with fixed voltage = ~ 1.39V but doesn't seem to be stable enough
AMD Work in Progress Ryzen 7 1700X @ 4.040GHz (Fixed voltage is bugged , this booted at 1.39V , testing in Windows 10 PE so no GPU driver)
40.75x multi , CPU-Z reads 1.392V but is not stable enough

41.00x multi with fixed voltage = ~ 1.41V but doesn't seem to be stable enough
Ryzen 7 1700X @ 4.090GHz , booted at ~ 1.41V
Amd Ryzen 7 1700X 8c/16t Processor
By:
Description:

AMD SenseMI Technology AMD XFR (eXtended Frequency Range) Technology Socket AM4 Max Turbo Frequency 3.8 GHz 16MB L3 Cache 4MB L2 Cache DDR4 Support Unlocked Processor Thermal Design Power 95W Without Fan and Cooler

Details:
DetailValue
Models:
Model Name/TypeMPNEAN/UPC
Overclock.net › Components › CPUs / Processors › Amd Ryzen 7 1700X 8c/16t Processor