[Videocardz] AMD Polaris 11 SKU spotted, has 16 Compute Units

[BlitzWulf]

http://videocardz.com/58955/amd-polaris-11-sku-spotted-has-16-compute-units

 

[thilanliyan]

Wow at the mem bandwidth of Vega
(if true)

 

[EightySix Four]

Doesn't the R9 390x have like ~5.9 TFLOPs of computing power? I know that doesn't directly correlate to graphics performance, but Polaris 10 could be really disappointing with those numbers. That's between R9 380 and R9 390 performance...

 

[poofyhairguy]

Wow only 800hmz while the rest are 1GHZ? Hopefully that is pure overclocked headroom.

 

[thesmokingman]

Vega is the answer no matter the question.

 

[poofyhairguy]

Doesn't the R9 390x have like ~5.9 TFLOPs of computing power? I know that doesn't directly correlate to graphics performance, but Polaris 10 could be really disappointing with those numbers. That's between R9 380 and R9 390 performance...

Yeah that memory bus means maybe people should start dialing back expectations, at least at high resolutions. Polaris 10 might be a 1080p monster though, and we all know the 970 made a killing because of that.

I am very interested in that Polaris 11. Looks like it will be cheap with only a 128 bus.

 

[ShintaiDK]

Wow at the mem bandwidth of Vega
(if true)

It would require someone to deliver 2Ghz HBM2. Samsungs HBM2 is currently clocked at 1.4Ghz.

 

[swilli89]

These numbers look super bogus. A 230mm die on a very high clocking process is going to be 800mhz? I'd wager clocks will be much higher.

Also Vega memory specs seem too high given the information we know.

 

[tential]

Yeah that memory bus means maybe people should start dialing back expectations, at least at high resolutions. Polaris 10 might be a 1080p monster though, and we all know the 970 made a killing because of that.

I am very interested in that Polaris 11. Looks like it will be cheap with only a 128 bus.

I've been asking about this as well I'm curious if this could be the case and am wondering if there is anything architecture wise that looks like it would stop it from being a 4k card while I wait for vega.

Sent from my C6833 using Tapatalk

 

[monstercameron]

These numbers look super bogus. A 230mm die on a very high clocking process is going to be 800mhz? I'd wager clocks will be much higher.

Also Vega memory specs seem too high given the information we know.

Gcn has never been a high clocker and chances are this mighy only use pcie power.

Sent from my SM-G930T using Tapatalk

 

[IllogicalGlory]

Is AMD really sticking with their old 6000MHz RAM? What's the point of Polaris's new memory controller?

 

[monstercameron]

Is AMD really sticking with their old 6000MHz RAM? What's the point of Polaris's new memory controller?

1024 shaders clocked @800mhz scream low power to me. Im assuming theyre keeping clocks low to only use pcie power.

Sent from my SM-G930T using Tapatalk

 

[Mahigan]

GCN 4.0 will basically power down unused GCN cores and boost GCN cores that are in use. Vector ALUs will also intelligently match incoming workloads leaving very little inefficiencies compute wise:
http://www.freepatentsonline.com/20160085551.pdf

So 2,304 GCN 4.0 Alu's might act more like 1.5GHz+ GCN 3.0 Alu's. That means a significant improvement in compute performance over GCN 3.0. It all depends on how high they will boost.

Seems like we know one way that they will boost CU performance and compute efficiency.

Say we're comparing GCN 3.0 with 4.0 cores. A single GCN 4.0 core would be around 50% more powerful (if not more).

Airfathaaaaa sent me this link.

 

[thesmokingman]

These numbers look super bogus. A 230mm die on a very high clocking process is going to be 800mhz? I'd wager clocks will be much higher.

Also Vega memory specs seem too high given the information we know.

Technically that should be obvious because well, there's nothing official. But yall knew that already.

 

[Qwertilot]

It would require someone to deliver 2Ghz HBM2. Samsungs HBM2 is currently clocked at 1.4Ghz.

They've got a long time to wait for that to turn up before vega though

 

[swilli89]

Gcn has never been a high clocker and chances are this mighy only use pcie power.

Sent from my SM-G930T using Tapatalk

This could also be a SFF-Nano or low power SKU, possibly even something like a 490X-Mobile. The 800Mhz clock would make perfect sense if this is the case.

AMD definitely is getting solid performance/mm2 out of Polaris on 14nm as they went very conservative with their die sizes. Despite that as well despite the fact that the GlobalFoundries 14nm process is 10-20% denser than TSMC's 16nm, I fully expect AMD to clock the range topping desktop version of Polaris 10 at ~1200Mhz or more base. They'll safely be inside 1x8 pin power limits even with frequency pushed way up like this.

I'm bookmarking this post so I can come back and claim a correct prediction if that holds true

 

[DooKey]

This could also be a SFF-Nano or low power SKU, possibly even something like a 490X-Mobile. The 800Mhz clock would make perfect sense if this is the case.

AMD definitely is getting solid performance/mm2 out of Polaris on 14nm as they went very conservative with their die sizes. Despite that as well despite the fact that the GlobalFoundries 14nm process is 10-20% denser than TSMC's 16nm, I fully expect AMD to clock the range topping desktop version of Polaris 10 at ~1200Mhz or more base. They'll safely be inside 1x8 pin power limits even with frequency pushed way up like this.

I'm bookmarking this post so I can come back and claim a correct prediction if that holds true

Sounds good, but don't be a hypocrite. Come back and own up to being wrong if that's the case.

 

[swilli89]

Sounds good, but don't be a hypocrite. Come back and own up to being wrong if that's the case.

Well I mean I'm not claiming this as inside information I'm openly saying its my prediction.

 

[Silverforce11]

GCN 4.0 will basically power down unused GCN cores and boost GCN cores that are in use. Vector ALUs will also intelligently match incoming workloads leaving very little inefficiencies compute wise:
http://www.freepatentsonline.com/20160085551.pdf

So 2,304 GCN 4.0 Alu's might act more like 1.5GHz+ GCN 3.0 Alu's. That means a significant improvement in compute performance over GCN 3.0. It all depends on how high they will boost.

Seems like we know one way that they will boost CU performance and compute efficiency.

Say we're comparing GCN 3.0 with 4.0 cores. A single GCN 4.0 core would be around 50% more powerful (if not more).

Airfathaaaaa sent me this link.

I expect them to improve their "Boost" to be more smarter and easier to OC. This is good news.

The 800mhz was leaked earlier, ES base clock.

Could well be a mobile part, low power market.

Also, can we put all the threads on Polaris together, likewise for Pascal? This forum is getting very messy.

Edit: Read that patent quickly, very interesting, they are making each of their SP/SIMDs capable of running parallel instructions based on different types of workloads. That scheduler is basically hyper-threading the SPs individually, wow! Also optimization in their wavefront > SIMD, it can operate at peak efficiency at 4, 8, 16, 32, 64 threads. Lastly, each individual SIMD can be gated down or auto boost overclock based on the workload and power target. That is a lot of improvements, that ends up being less but more powerful and efficient SPs that handle bottlenecks much better. Very nice! Can't wait.

 

[RussianSensation]

It would require someone to deliver 2Ghz HBM2. Samsungs HBM2 is currently clocked at 1.4Ghz.

Considering you predicted and hyped up for months that NV's DX12_1 support was a killer feature spec over GCN's "lack of full DX12 support", how you predicted no APUs in all 3 next gen consoles, how you predicted AMD will not improve perf/watt with Fiji on 28nm (in fact as far as to say Fury X is just dual Tonga 2048 on 1 PCB), then forecasted AMD will not manufacture a flagship with > 550mm2 die sized, downplayed Async Compute impact on DX12 from day 1, downplayed the driver downfall that would cripple the entire Kepler line, downplayed any relevant impact of 3.5GB 970 VRAM-gimp, I am going to say that it's actually far wiser bet is to take your negative predictions and do a 180* opposite of them. That means Vega actually should nail 1TB/sec HBM2 chips if your track record of predicting the complete opposite continues.

---

As far as those specs go, the performance gulf between High-End and Enthusiast on that chart is too great. Polaris 10 is mid-range, with Vega 11 (or cut down Vega 10) high-end and Vega 10 enthusiast. There is no way the flagship Vega will have 4096 shaders and Polaris only has 2304 as it would mean a gargantuan gap in-between. AMD always has a killer high-end cut down part hitting the sweet spot; and that chart shows no such card.

HD5850
HD6950
HD7950
R9 290
AMD Fury

Each of those cards Max OC offered 90-95% of the performance of the flagship card OCed and it came directly from a cut-down flagship. That is the key missing link between a supposed 4096 Vega and a 2304 Polaris.

 

[Silverforce11]

Considering you predicted and hyped up for months that NV's DX12_1 support was a killer feature spec over GCN's "lack of full DX12 support", how you predicted no APUs in all 3 next gen consoles, how you predicted AMD will not improve perf/watt with Fiji on 28nm (in fact as far as to say Fury X is just dual Tonga 2048 on 1 PCB), then forecasted AMD will not manufacture a flagship with > 550mm2 die sized, downplayed Async Compute impact on DX12 from day 1, downplayed the driver downfall that would cripple the entire Kepler line, downplayed any relevant impact of 3.5GB 970 VRAM-gimp, I am going to say that it's actually far wider to take your negative predictions and do a 180* opposite of them. That means Vega actually should nail 1TB/sec HBM2 chips if your track record of predicting the complete opposite continues.

LOL oh man, schooled!

And he has the galls to misquote me in his sig without context, click on the link and I am correct in that post.

Basically there's a metric here, how unimpressed some individuals are with a feature or hardware = the better it will be.

ps. I believe he and his cohort also predicted how Mantle is useless, it will never get anywhere, nobody will use it... yet here we are, in the Vulkan, Metal and DX12 era. All gamers are gonna run AMD Mantle code underneath these APIs that appeared magically out of nowhere sharing 99% similarities with fancy names, like I called it years ago.

 

[kawi6rr]

Considering you predicted and hyped up for months that NV's DX12_1 support was a killer feature spec over GCN's "lack of full DX12 support", how you predicted no APUs in all 3 next gen consoles, how you predicted AMD will not improve perf/watt with Fiji on 28nm (in fact as far as to say Fury X is just dual Tonga 2048 on 1 PCB), then forecasted AMD will not manufacture a flagship with > 550mm2 die sized, downplayed Async Compute impact on DX12 from day 1, downplayed the driver downfall that would cripple the entire Kepler line, downplayed any relevant impact of 3.5GB 970 VRAM-gimp, I am going to say that it's actually far wiser bet is to take your negative predictions and do a 180* opposite of them. That means Vega actually should nail 1TB/sec HBM2 chips if your track record of predicting the complete opposite continues.

That was the best post I've read in years, you sir deserve an award, thank you.

Looking forward to all the new cards can't wait for the releases.

 

[96Firebird]

Wow Russian, nice attack. I don't see anything in ShintaiDK's post predicting anything, he is simply posting fact. You've hit a new low there... :\







Member callout. It's not allowed.


esquared
Anandtech Forum Director

 

[thilanliyan]

Shintai, as for no 2GHz HBM yet, is there some physical limitation for there to be none available by the time Vega launches early next year (ie. did anyone say it is not manufacturable)? How long between release of said HBM and launch would AMD need to incorporate it into their cards?

 

[thilanliyan]

Wow Russian, nice attack. I don't see anything in ShintaiDK's post predicting anything, he is simply posting fact. You've hit a new low there... :\

RS also posted some facts