Two new laptop GPUs launched in NVIDIA’s low-end MX line. This classification of products is designed to slide above the GPUs found on typical laptop CPUs by a wide enough margin to justify an extra chip, but not enough to be endorsed as part of their gaming line.
As such, pretty much the only performance number that NVIDIA provides is an “up-to” factor relative to Intel’s HD620 iGPU as seen on the Core i5-8265U. For reference, the iGPU on this specific CPU has 192 shader units running at up to 1.1 GHz. Technically there exists some variants that have boost clocks up to 1.15 GHz but that extra 4.5% shouldn’t matter too much for this comparison.
Versus this part, the MX250 is rated as up to 3.5x faster; the MX230 is rated at up to 2.6x faster.
One thing that I should note is that the last generation’s MX150 is listed as up to 4x the Intel UHD 620, although they don’t state which specific CPU’s UHD 620.
This leads to a few possibilities:
- The MX250 has a minor performance regression versus the MX150 in the “up to” test(s)
- The UHD 620 had significant driver optimizations in at least the “up to” test(s)
- The UHD 620 that they tested back then is significantly slower than the i5-8265U
- They rounded differently then vs now
- They couldn’t include the previous “up to” test for some reason
Unfortunately, because NVIDIA is not releasing any specifics, we can only list possibilities and maybe speculate if one seems exceedingly likely. (To me, none of the first four stands out head-and-shoulders above the other three.)
Like the MX150 that came before it, both the MX230 and MX250 will use GDDR5 memory. The MX130 could be paired with either GDDR5 or DDR3.
Anandtech speculates that it is based on the GP108, which is a safe assumption. NVIDIA confirmed that the new parts are using the Pascal architecture, and the GP108 is the Pascal chip in that performance range. Anandtech also claims that the MX230 and MX250 are fabricated under Samsung 14nm, while the “typical” MX150 is TSMC 16nm. The Wikipedia list of NVIDIA graphics, however, claims that the MX150 is fabricated at 14nm. While both could be right, a die shrink would make a bit of sense to squeeze out a few more chips from a wafer (if yields are relatively equal). If that’s the case, and they changed manufacturers, then there might be a slight revision change to the GP108; these changes happen frequently, and their effects should be invisible to the end user… but sometimes they make a difference.
It’ll be interesting to see benchmarks when they hit the market.
So Pascal’s Shader cores/SMs
So Pascal’s Shader cores/SMs can not dual issue INT and FP Instructions but Turing’s Shader cores can. But the Pascal MX150 is more ROP heavy compared to Intel and so is AMD’s designs compared to Intel.[see quoted Turing information below]
For Intel’s HD620 wikipedia says: 192:24:3
For Nvidia’s MX150 wikipedia says: 384:24:16 [Note: TPU’s TMU counts for the MX150 is 32 and not the 24 that wikipedia states]
So look at the difference in Shaders and ROPs and I sure hope that Raja and Intel are going to up their ROP counts or they, like AMD, will be behind Nvidia on that all important ROP count and that HIGHER Pixel Fill Rate metric that comes from MOAR ROPs!
Anandtech lists the MX250 as 384:??:16 and that 14nm/Samsung node is where possibily some of the extra performance is coming from. So the MX250 is rated as up to 3.5x faster and the MX230 is rated at up to 2.6x faster than the i5 CPU part with the Intel HD620 graphics as it compares to the new Nvidia MX 200 series SKUs! Intel clocks its HD620 graphics higher on non U series branded parts but really looking at Intel’s ROP count information from TechPowerUp’s database for HD620(Kaby Lake Mobile) it’s 3 ROPs at a Pixel Rate: 3.150 GPixel/s for HD620 with GPU Clock: 300 MHz base and Boost Clock:1050 MHz.
So Nvidia’s MX150 has a Pixel Rate of 23.49 GPixel/s compared to the HD 620’s 3.150 GPixel/s and its easy to see where the performance delta in gaming comes from. And The MX150’s Base Clock: 1227 MHz and Boost Clock:1468 MHz.
Who knows currently but after the MX 200 series SKUs release there will be folks that will find out for sure after stripping some MX 200 series die samples of their outer layers and getting out the microscope and camera. So if its Pascal GP108 on 14nm they will find out and Nvidia has been making use of both Samsung’s 14nm and TSMC’s 16nm for a good long while now.
“New Streaming Multiprocessor (SM)
Turing introduces a new processor architecture, the Turing SM, that delivers a dramatic boost in shading efficiency, achieving 50% improvement in delivered performance per CUDA Core compared to the Pascal generation. These improvements are enabled by two key architectural changes. First, the Turing SM adds a new independent integer datapath that can execute instructions concurrently with the floating-point math datapath. In previous generations, executing these instructions would have blocked floating-point instructions from issuing. Second, the SM memory path has been redesigned to unify shared memory, texture caching, and memory load caching into one unit. This translates to 2x more bandwidth and more than 2x more capacity available for L1 cache for common workloads.” [see Nvidia’s Turing micro-Arch whitepaper]