AMD Vega 10 3DMark Fire Strike Benchmark Entry Spotted – 687F:C1 Device With 8GB of 700MHz vRAM & 1200MHz Core Clock
AMD's highly anticipated next generation Vega 10 GPU has been spotted in the 3DMark Fire Strike database. This is the very same GPU that AMD has showcased in numerous performance demos over the past several months.
So, let's jump straight into the details. First things first, the device ID of the graphics card in the discovered entry is 687F:C1. The device ID matches that of the earliest Vega 10 prototype that we know of. The tested graphics card has 8GB of video RAM running at 700MHz and the card's GPU clock is a modest 1200MHz.
This Is AMD's Earliest Public Vega 10 Engineering Sample
All of the specifications match Vega 10, allbeit an early engineering sample of the graphics chip. In fact, AMD used an identical Vega 10 687F:C1 prototype in its Vega performance demos, Including the DOOM 4K demo back in January and the Star Wars Battlefront 4K Demo in February. The 687F:C1 engineering sample board is at least five months old and was described by AMD's Raja Koduri as one of the earliest samples out of the fab.
The memory capacity of the card spotted in the 3DMark database again matches that of Vega 10 and the low memory clock speed once again falls right in the operating frequency range of HBM2.
The Vega 10 687F:C1 Engineering Sample 3DMark Fire Strike Performance Result
The 3DMark Fire Strike entry shows a graphics score of 17801, which is roughly 1400 points more than an R9 Fury X, nearly exactly the same as a Maxwell GTX Titan X and less than a couple hundred points below a GTX 1070.Reference
Putting The Vega 10 687F:C1 Performance Figures Into Perspective
Because this is an engineering sample we have to put its performance into perspective of what AMD expects its final production ready Vega 10 graphics cards to be able to deliver. Thanks to AMD's DOOM demo we can seamlessly connect the dots. We know that the Vega 10 graphics card that AMD has been using all this time in its demos over the past several months was in fact the same 687F:C1 prototype running at 1.2GHz.
1200MHz is a fairly modest clock speed, even when compared to Polaris 10. We know the tweaked version of Polaris 10, Polaris 20, which is now in every RX 580 graphics card out there runs at 1350MHz+. In fact many RX 580 graphics cards widely available today run at 1400MHz+ right out of the box.
As with any semiconductor product, between the very first 687F:C1 Vega engineering graphics board and the final production ready Radeon RX Vega are going to be many iterations of the silicon. Engineering teams work tirelessly during product bring-up to produce one iteration after the other as they optimize the silicon and iron out issues.
During this process and as the product matures overtime, more and more performance is squeezed out of the silicon before it's finally ready to ship. This is why we always see the final products, whether it be AMD, Nvidia or even Intel chips, end up running at significantly higher clock speeds and considerably higher efficiency than their engineering sample counterparts.
The Vega Architecture Is Designed For Higher Clock Speeds Than Polaris
It's no secret that the Vega architecture has been designed from the get-go to operate at higher frequencies than Polaris. If you're interested in reading more about what differentiates the Vega architecture from Polaris you can check out our Vega architecture article here.
Now, based on AMD's performance figure for its Vega 10 based Radeon Instinct MI25 accelerator we know that the company expects Vega to run at above 1.55GHz. The MI25's single precision compute performance figure is 12.5 TFLOPS and its half precision compute performance is rated at 25 TFLOPS. AMD's graphics guru Raja Koduri revealed that this is the reason the accelerator was in fact named the MI25
I should point out that thanks to a Linux driver patch that came out last week Vega 10's specs, including the core count, are no longer subject to speculation. The new driver released by AMD revealed Vega 10's specifications in intimate detail. Vega 10 has 64 compute units and 4096 stream processors, so it needs to operate at 1563MHz to deliver that 12.5 TFLOPS single precision compute figure, because each stream processor can execute two FP32 operations per cycle.
How AMD Expects Production Ready Radeon RX Vega Graphics Cards To Perform
Remember, the MI25 accelerator is a passively cooled device. Which in reality means 1563MHz is potentially a conservative clock speed for Vega 10 and actively cooled graphics cards should be able to run at higher clock speeds.
Now keep in mind that all of this is based on information AMD had made public early this year based on its expectations. Vega 10 may eventually punch above its weight or fall below those projections, we simply cannot know for certain just yet. What we can safely assume is that Vega 10 will not run at clock speeds below Polaris, based purely on the higher clock speed advantage of the Vega architecture and the maturity level of the FinFET manufacturing process.
Realistically speaking, if AMD manages to successfully hit its 1.56GHz projection, production ready Vega 10 will run at a clock speed that's 30% higher than what we're seeing from the 687F:C1 prototype. Performance doesn't necessarily scale in perfect linearity with the core clock speed however, as other factors play a major role like memory bandwidth and internal architectural bottlenecks that we know very little about at this point.
According to the JEDEC spec HBM2 should scale all the way to 1000MHz/2Gbps but currently available HBM2 stacks are limited to 800MHz/1.6Gbps. With that in mind, a 30% higher GPU clock speed and a 100MHz boost to the memory would move Vega 10 from its current 1070'ish spot to a more competitively favorable position with the GTX 1080.
Radeon RX Vega ( Vega 10 ) Is Set To Be A GTX 1080 Contender
Perfect frequency/performance scaling would put it 7% ahead of the 1080 in 3DMark Fire Strike. A more realistic estimate would put it right around the performance level of the GTX 1080. That is assuming the Radeon RX Vega flagship ends up at around the ~1.56GHz clock speed of the passively cooled MI25 accelerator. So without doubt there's still more wiggle room for the Radeon Technologies Group if they're able to push the frequency higher than that. Things like liquid cooling which they're potentially going for again can definitely help.
A cut down Vega 10 will compete with the GTX 1070. The gap between the GTX 1080 and GTX 1070 is sizable, definitely larger than #80 and #70 cards of past generations. This should help a lower binned Vega 10 -- let's call it Vega 10 Pro -- put on a good showing against the GTX 1070.
Before we close here it's very important to understand that what AMD is attempting with the Vega architecture as a whole does not begin with nor does it end with enthusiast gaming Radeon RX Vega graphics cards.
The new architecture means a lot more than that to the company. Vega's high bandwidth cache technology is a truly unique memory subsystem. A technology that is poised to shine brightest in professional compute arena, especially in applications that demand massive sets of data. Think AI, machine learning and many forms of data analysis. While the implications of this forward-looking architecture in gaming may be far into the future its potential can be immediately realized in the GPGPU compute space.
With the Vega launch approaching, it's critical that enthusiasts looking forward to the new GPU hit the pause button for a moment and manage their expectations. If you were expecting a monster that's going to dethrone NVIDIA's GTX Titan Xp, think again. If you were looking for GTX 1070 and GTX 1080 competition from the Red team to shake things up, you're going to get it.
Vega is officially slated for release some time between now and before the end of June according to AMD's CEO.