MD unveiled its new Radeon RX 5700 line of graphics cards with 7nm chips at E3 last month, and with just days to go before they launch on July 7th, the company has announced new pricing. In the "spirit" of competition that it says is "heating up" in the graphics market -- specifically NVIDIA's "Super" new RTX cards -- all three versions of the graphics card will be cheaper than we thought.
The standard Radeon RX 5700 with 36 compute units and speeds of up to 1.7GHz was originally announced at $379, but will instead hit shelves at $349 -- the same price as NVIDIA's RTX 2060. The 5700 XT card that brings 40 compute units and up to 1.9GHz speed will be $50 cheaper than expected, launching at $399. The same goes for the 50th Anniversary with a slightly higher boost speed and stylish gold trim that will cost $449 instead of $499.
That's enough to keep them both cheaper than the $499 RTX 2070 Super -- we'll have to wait for the performance reviews to find out if it's enough to make sure they're still relevant.
AMD fires back at 'Super' NVIDIA with Radeon RX 5700 price cuts
If you look at the numbers for Techspot's RX 5500 XT review, the 5500 XT 8GB and the GTX 1650 Super are effectively the same in performance, and you can see the very large jump in performance from the 1650 to the 1650 Super (37%), which is about the average that AMD claims the 5300 will have over the 1650, which is reasonable especially if clocks are able to be raised to 5500 levels, which is possible given it's the same GPU and has an 8 pin power connector.
The one saving grace for it is that there are likely a much larger stockpile of RDNA GPUs than there are Turing GPUs, especially since there has been no mention of a GTX 1650/60 replacement yet, though that will likely change once RDNA2 cards launch and enough dies accumulate which didn't bin high enough to reach RTX 3060 levels of performance...
Now if it's priced at $100, that's a whole different ball game, but anything north of $125 is going to make it another dust collector.
I'm betting they used a PCIe 8 pin so AIBs can make highly overclocked models since it is the same GPU as the 5500, and the last thing they want is to have another publicized "AMD Violates PCIe SIG Spec" hassle. I don't think we will be able to see a 75w version. To quote the TomsHardware review of the 5500 XT:
Lower clocks won't shave off 40w, and a cut down GPU won't perform well enough in 2020 compared to an iGPU to matter, especially when Renoir is out. Not even Turing GTX was that efficient...
Also, correct me if I'm wrong, but don't "eSports" people want 144 or 250FPS because they die if they don't?
With M.2 becoming much more of a thing though, 2.75 slot coolers are going to be the new normal as the first PCIe slot has 3 full slots of space, especially on higher end motherboards.
"So did you watch the RTX3000 series launch?
Obviously Nvidia marketing and we need to wait for benchmarks and analysis.
They claim 2.7x shader performance 1.7x RT-Core performance and 2.7x Tensor Core Performance.
They have built the GPUs on Samsung 8n Nvidia Custom Process and uses Micron GDDR6X memory.
They claim a 1.9x Performance / Watt improvement.
They are claiming ~ 1.6x - 1.8x performance in "RTXon" games like BFV.
Interesting to me, they are claiming > 2x performance in Blenders Cycles.
Not much discussion on basic Rasterising performance and many questions to answer"
Interesting stuff going on here. First and foremost.
"They claim a 1.9x Performance / Watt improvement."
Probably not. Looking at the chart they displayed at the reveal.
It is typical to compare GPUs at a fixed level of power consumption, not a fixed fps. If you compare 160W on both curves and look at the performance uplift it is more like 1.4X. But since there are no standards for this type of thing I guess they can measure it however they want.
"Not much discussion on basic Rasterising performance and many questions to answer"
This is also interesting, given the size of the dies and the number of CUDA cores they claim. It doesn't look like the dies are big enough to physically hold 2X the number of shaders. And in fact, they don't.
This is from a Q&A session on reddit.
- "Could you elaborate a little on these doubling of CUDA cores?
[Tony Tamasi] One of the key design goals for the Ampere 30-series SM was to achieve twice the throughput for FP32 operations compared to the Turing SM. To accomplish this goal, the Ampere SM includes new datapath designs for FP32 and INT32 operations. One datapath in each partition consists of 16 FP32 CUDA Cores capable of executing 16 FP32 operations per clock. Another datapath consists of both 16 FP32 CUDA Cores and 16 INT32 Cores. As a result of this new design, each Ampere SM partition is capable of executing either 32 FP32 operations per clock, or 16 FP32 and 16 INT32 operations per clock. All four SM partitions combined can execute 128 FP32 operations per clock, which is double the FP32 rate of the Turing SM, or 64 FP32 and 64 INT32 operations per clock.
Doubling the processing speed for FP32 improves performance for a number of common graphics and compute operations and algorithms. Modern shader workloads typically have a mixture of FP32 arithmetic instructions such as FFMA, floating point additions (FADD), or floating point multiplications (FMUL), combined with simpler instructions such as integer adds for addressing and fetching data, floating point compare, or min/max for processing results, etc. Performance gains will vary at the shader and application level depending on the mix of instructions. Ray tracing denoising shaders are good examples that might benefit greatly from doubling FP32 throughput.
Doubling math throughput required doubling the data paths supporting it, which is why the Ampere SM also doubled the shared memory and L1 cache performance for the SM. (128 bytes/clock per Ampere SM versus 64 bytes/clock in Turing). Total L1 bandwidth for GeForce RTX 3080 is 219 GB/sec versus 116 GB/sec for GeForce RTX 2080 Super.
Like prior NVIDIA GPUs, Ampere is composed of Graphics Processing Clusters (GPCs), Texture Processing Clusters (TPCs), Streaming Multiprocessors (SMs), Raster Operators (ROPS), and memory controllers.
The GPC is the dominant high-level hardware block with all of the key graphics processing units residing inside the GPC. Each GPC includes a dedicated Raster Engine, and now also includes two ROP partitions (each partition containing eight ROP units), which is a new feature for NVIDIA Ampere Architecture GA10x GPUs. More details on the NVIDIA Ampere architecture can be found in NVIDIA’s Ampere Architecture White Paper, which will be published in the coming days.."
So, there are actually physcially 0.5X the number of shaders in Ampere as listed. The data path for each SM allows either 64FP32 64INT32 just like Turing, or 128 FP32. So in a FP32 heavy workload the int32 path is shutoff, and the FP32 path clocked up to allow two executions per cycle. That's significant because, it will never be a straight 2X increase in performance. There is overhead that has to be programed for the engine to decide when to use 64 + 64 or 128. Depending on how well that works you could get close to having physically double the number of shades. Of course, in workloads that include a high number of integer instructions the engine will perform exactly like Turing.
So NVidia are getting a little creative with how they are marketing a FP32 "core".
