AMD Radeon VII Unboxing & Preview

[Vayra86]

This should have been your first comment, seems like you finally did understand.
Your argument is technically true, designing any new GPU costs money, but in my opinion it's not the same story as nvidia's RTX lineup.
I believe VII is a last minute product and Vega 20 silicon wasn't supposed to be a gaming focused silicon, I might be wrong though.

Last minute? There's no reason to believe that. I think Lisa is being as smart as she's been with the Zen launch: Hit em hard, and hit em fast, and make sure you don't sell air, but products. Radeon VII benefits from its 'hasty' launch: it needs the surprise element to sell. In the longer run, Nvidia will just eclipse it from below and from above with the 2080 and the 2070 either on price or on performance.

 

[Divide Overflow]

I'd like to see a 7 nm 8 GB 64 ROPS Vega instead of this cut down data center card.

 

[M2B]

it needs the surprise element to sell

Which it doesn't have. You can't surprise anyone with a flagship product that is only as fast as competitor's last gen flagship, after 2 years, at the same price.
GPU war isn't birthday party

And let me tell you something: Radeon VII exists mostly because RTX 2080 is too slow for the money.

 

[btarunr]

I'd like to see a 7 nm 8 GB 64 ROPS Vega instead of this cut down data center card.

So you want a cut-down cut-down data center card?

 

[EarthDog]

So you want a cut-down cut-down data center card?

Yes!

Performance will be the same (except perhaps at 4K UHD in some titles). It should be cheaper and would have a leg up on the competition.

 

[Mistral]

That cooler is a thing of beauty....

 

[R0H1T]

LOL. This is just a rebranded card. No real R&D.

You're making yourself look like a fool, this thing is expensive to make probably second only to 2080Ti
Do you also think 7nm chips landed in AMD's lap from the heavens, with no real R&D

 

[M2B]

You're making yourself look like a fool, this thing is expensive to make probably second only to 2080Ti
Do you also think 7nm chips landed in AMD's lap from the heavens, with no real R&D

1) Chillout
2) Radeon VII is a very small die, smaller than an RTX 2060 which means yields are probably very good for AMD - generally it's not a pig for mass production like the TU102 is (or even the TU104).
3)HBM2 is significantly cheaper than when original vega launched.
4) I did not say it's not expensive to make, just not as expensive as some of you think.
5) AND YEAH, R&D budget behind this product is nowhere near as much as nvidia's.

 

[AnarchoPrimitiv]

This thing is a godsend...Ive been waiting on a new videocard for a brand new 2700x build I started in December and I really didnt want to buy an RTX card so I don't reward Nvidia's avarice and because I really dont care about ray tracing right now....I'm buying one of these Radeon VIIs ASAP

 

[R0H1T]

1. Alright.
2. yields - we don't know how good or bad they are so I won't speculate on that given we have no hard facts on either product lines, from AMD or Nvidia.
3. HBM2 is still pretty expensive, it's 2x what was on Vega also HBM2 demand is even more now than 2 years back.
4. BoM is north of $500-550 ~ you're also forgetting then that packing costs with interposer are high, not to mention failure rates also go up.
5. How do you count R&D budget in the grand scheme of things? A $100 million fixed costs with 6 to 12 products, to recoup that spend, or $50 million for just the one? Nvidia will not only make more money off RTX, they'll also sell more RTX cards even after recouping the initial R&D & not lowering the prices. The point being R&D is proportional to the expected sales from such products, given what we've seen with RTX, prices & all. Nvidia will not only recoup their spend a lot earlier they'll also continue milking the top end for as long as they possibly can, which is virtually forever given lack of alternates from AMD.

 

[M2B]

A $100 million fixed costs with 6 to 12 products, to recoup that spend, or $50 million for just the one?

Nvidia has designed different chips with different configurations, heck, TU104 itself is only used in one gaming product, just like the Vega 20, so your statement is completely irrelevant, it's not like all turing cards are based on the same chip.

 

[IceShroom]

Looks like Nvidia offence force already in action.

Nvidia has designed different chips with different configurations, heck, TU104 itself is only used in one gaming product, just like the Vega 20, so your statement is completely irrelevant, it's not like all turing cards are based on the same chip.

Well TU104 is used in 6 different cards and Vega20 in 3 different cards.

 

[95Viper]

Keep it on topic, please.
Thank You.

 

[Anymal]

Just wait till review, pleasant surprise for nvidia it will be.

 

[phill]

I like it, much better and nicer than the damn blower cards we always see on release, I hope that this one can tame the beast (so to speak )

I hope there's a full cover water block similar colour for the card.. That will look rather nice and I'm sure it'll help with the thermals, after all, water is a lot more efficient and effective than air cooling for these things Looking forward to seeing the review. I hope that it will be as good as the RTX 2080......

 

[ratirt]

1. Alright.
2. yields - we don't know how good or bad they are so I won't speculate on that given we have no hard facts on either product lines, from AMD or Nvidia.
3. HBM2 is still pretty expensive, it's 2x what was on Vega also HBM2 demand is even more now than 2 years back.
4. BoM is north of $500-550 ~ you're also forgetting then that packing costs with interposer are high, not to mention failure rates also go up.
5. How do you count R&D budget in the grand scheme of things? A $100 million fixed costs with 6 to 12 products, to recoup that spend, or $50 million for just the one? Nvidia will not only make more money off RTX, they'll also sell more RTX cards even after recouping the initial R&D & not lowering the prices. The point being R&D is proportional to the expected sales from such products, given what we've seen with RTX, prices & all. Nvidia will not only recoup their spend a lot earlier they'll also continue milking the top end for as long as they possibly can, which is virtually forever given lack of alternates from AMD.

HBM is very expensive so is the GDDR6. It reflects pretty well the entire NV lineup or maybe NV will drop the price? I really don't think so. that would be a miracle. You can tell that by the 2060 Vram size. Cost cuts. (NV cuts the vram everybody is ok with it. AMD gives more Vram everybody is angry. I don't get that). I think AMD may drop the price and I also think, AMD released the Radeon Vega VII because Navi isn't ready yet.(nothing to show on the CES) They have encountered some problems with the chips and it probably will be delayed. Besides you don't know what's the improvement for HBM2 memory manufacturing now? I'm sure it has improved and the price might have come down due to the fact it's been in production for a while now.
Anyway, what I really want to know is how this card performs. I can't care less about if it's a rebrand or not. It would've been a nice purchase for my rig if the card turn out to be noticeably better than my V64
BTW. About that 60 CU's for RVII. Gamers Nexus has a theory that there's a correlation to the CU number and possibility to OC the card. They have mentioned it several times and proved it to be true with the V56, which can OC better than V64 due to the CUs. (if you wipe power restrictions). RVII is 7nm which gives a boost in frequency and maybe these cut down CUs from 64 to 60 give also some boost. Who knows, maybe we will be ale to squeeze more from RVII than what we've been used with Vega's? I'd really like RVII to have some OC headroom.

 

[F7GOS]

Dont know why but I really want one

 

[jabbadap]

Looks like Nvidia offence force already in action.

Well TU104 is used in 6 different cards and Vega20 in 3 different cards.

Uhm tu104 is used in quadro RTX 5000 and geforce rtx 2080. And well if you count mobile rtx2080+MaxQ so in four cards. There's no Tesla card out yet from that chip. Tesla T4 is tu106 as is rtx 4000, rtx 2070, 2060, other mobile chips. But yeah you are right on that it's not only a gaming product.

As for vega20, I'm sure there will be Radeon Pro Wx from it. Though it will probably have mi60 config rather than mi50. For visual designing 32GB of vRAM is very justified... Well yeah maybe they will release it with both configs, like there is full vega10 wx9100 and 56cu wx8200 now. On mi and pro lines that vrm section will probably be full used. Which reminds me AMD has changed vrm components from the past. So looking forward buildzoid's pcb breakdown.

 

[sepheronx]

Is the PCB just as tall as the total cooler for the RX 580 Sapphire Nitro +? Reason why I am asking is because due to the cooler being tall, it wouldn't fit in the Couger QBX case. If that is the case, this card would not fit in my current case.

 

[Dimi]

It looks so boring, why don't they ever try something new, design wise?

I also don't understand why reviewers aka influencers have to receive special editions that no consumer will ever get their hands on.

 

[sepheronx]

It looks so boring, why don't they ever try something new, design wise?

I also don't understand why reviewers aka influencers have to receive special editions that no consumer will ever get their hands on.

I wish cards were smaller. Offer more ITX cards. Or just make them a decent size we're they fit even in a Alienware x51

 

[Deleted member 158293]

I'd be first inline for a single slot Radeon VII NANO card with a pack of card sized waterblock.

AMD has done it before with the Fury Nano, they could do it again with HBM & enough demand.

 

[efikkan]

Last minute? There's no reason to believe that.

It has been mentioned several times that Vega20 would be a professional product, even as recently as last summer. The decision to launch Radeon VII wasn't literally last minute of course, but it was a backup plan put into action late last year, very likely due to the delays of Navi.

2) Radeon VII is a very small die, smaller than an RTX 2060 which means yields are probably very good for AMD - generally it's not a pig for mass production like the TU102 is (or even the TU104).
4) I did not say it's not expensive to make, just not as expensive as some of you think.

7nm is currently more than twice the price per wafer, so the move to 7nm isn't about saving costs, but about pushing density and efficiency.

 

[Anymal]

FFS, just innovate something new, GCN is pasé.

 

[anubis44]

LOL. This is just a rebranded card. No real R&D.

This is not a rebrand. Even if it's only a die-shrink of the 14nm Vega, it's not a rebrand. A rebrand is literally the same product on the same die-process, with maybe an overclock/added memory on the card to distinguish it from the previous GPU.